Skip to main content

Fast Company Technology

This founder built the RV he wanted for ski trips. It led to a $20 million business

3 weeks 5 days ago

Jim Ritchie started AEONrv, a Nevada-based company that sells a high-tech hybrid RV, because he wanted to create a new kind of vehicle for himself to use. The result is an RV with a modern, all-electric cabin mounted on the chassis of a gas-burning Ford Transit van.

The company’s growth has been strong enough to land it at No. 66 on the Inc. 5000 list this year, with a three-year growth rate of 4,231 percent. And it has managed to pull off this growth while the overall RV industry has declined from a Covid-era high.

Ritchie, a software engineer, says pondering his own outdoor adventures jump-started the creation of the company. “I’m a big skier, and I was thinking maybe I’ll retire sometime soon, and it’d be cool to have a vehicle to take to go skiing and biking and things I like to do,” he says. “I was looking around at vans. Van life was big, and I didn’t really like what was going on there. They seemed hard to insulate.”

All told, he says, the traditional RV industry “seemed pretty old-school” to him. 

Ritchie designed the vehicle for fun, and to help make it a reality, teamed up with a friend, Lars Severin, who became the company’s CTO (Severin retired earlier this year). What began as a weekend project grew into a business. By the end of 2021, they had published a website and started taking deposits for the vehicles. “We hired our first employee a little over four years ago,” he says, “and the rest is history.” They bootstrapped the company to get it off the ground, and then raised under $2 million of outside capital from investors in 2023 and 2024.

The vehicle boasts a tech-forward design centered on a snugly insulated, all-electric cabin. Solar panels on the roof generate electricity, which is stored in lithium-ion batteries. Starlink provides an internet connection. Ritchie emphasizes that inspiration for the build came from the marine industry, and that the cabin is built out of fiberglass-reinforced insulated panels. “There’s no aluminum or metal in the actual build,” he says. “It’s all composites and fiberglass, and it’s all glued together.” There’s also no propane, which traditional RVs use for appliances like stoves and even historically as a power source for refrigerators.

The result is an energy-efficient cabin that’s so airtight, Ritchie says, occupants “have to crack a window at night, so you don’t get carbon monoxide poisoning.”

The company, which employs some 60 full-time employees, has delivered more than 150 vehicles in total through its direct-to-consumer approach, the same way EV companies Tesla and Rivian sell their rides. They deliver about eight vehicles each month, with the starting price for the base model at $229,500. Ritchie predicts that the company’s 2026 revenue will land around $20 million; he says the company was profitable as of last year and remains on track to turn a profit this year as well. 

AEONrv remains a niche, if tech-forward, player in a $159 billion industry with three big companies: THOR Industries (companies under its umbrella include Airstream and CrossRoads), Forest River, and Winnebago Industries. Those companies “make up the bulk of the RV production in the United States,” says Monika Geraci, a spokesperson for the RV Industry Association. Not only that, 88 percent of RVs last year were made in one state: Indiana. 

RV shipments in the U.S. spiked in 2021 amid the Covid pandemic, hitting a high of 600,241 that year, according to the RV Industry Association. Then they fell. Shipments from manufacturers to dealers last year were 342,220, a drop of about 43 percent from that Covid-era high. Geraci says that current estimates hold that RV production in the U.S. will drop by about 8 percent this year compared with last. 

She attributes the decline to economic factors. “For people who own RVs, we see them use them more because it is a more economical way to travel,” she says, compared with flying and staying in a hotel. But she says that “economic uncertainty” and factors like interest rates and inflation are “keeping a lot of people who want to purchase an RV on the sidelines.”

If you’re picturing a motorhome like a bus when you think of an RV, that type of vehicle is actually in the minority of RV sales. Last year, nearly 90 percent of RVs shipped were the towable kind. In fact, Ritchie says that they’re working on a towable version of their product that will cost less than $100,000. 

Besides AEONrv, other players in the electric RV space include Lightship, which sells an electric trailer called the AE.1 that can lower to become sleeker on the road; and Pebble makes the Pebble Flow, which is also an electric RV trailer. Pebble said in July that it had delivered more than 100 of its trailers. Lightship declines to share how many units it’s delivered, but notes that it started shipping production vehicles late last year and has already sold all the vehicles it plans to build between this summer and next. The products from both Lightship and Pebble embrace a tech-forward, spaceship-like design for their trailers. 

As for the appeal of an RV with an electric cabin, Geraci notes that “there is an RV for every lifestyle and every price point, and so this is another option that is out there. I don’t foresee the RV industry going all electric.” 

Jason Epperson, a longtime RVer who hosts a podcast and YouTube channel called RV Miles along with his wife, Abby Epperson, says that startups like AEONrv are “all trying to do something a little bit different from what the traditional Indiana RV industry has done.”

As for those traditional RVs, Ritchie is blunt about what he thinks about their appearance: “They look like 1980s Vegas or something.”

—Rob Verger

This article originally appeared on Fast Company’s sister website, Inc.com. 

Inc. is the voice of the American entrepreneur. We inspire, inform, and document the most fascinating people in business: the risk-takers, the innovators, and the ultra-driven go-getters that represent the most dynamic force in the American economy.

How Google is building a gigawatt-scale power plant—without building a power plant

3 weeks 5 days ago

As hyperscalers pour billions of dollars into new power plants for data centers, Google is also investing in something different: making better use of the existing grid.

The tech giant announced a new deal today for a gigawatt-sized virtual power plant (VPP), which appears to be the largest deal of its kind from a hyperscaler. The approach shrinks overall power demand at key moments. Voltus, Google’s partner, pays businesses and homeowners to use less electricity at times of day when the grid is under strain. A big box store might automatically adjust its thermostat by a couple of degrees; a factory might temporarily run part of its production on a battery. At a house, the electric vehicle in the garage might adjust its charging schedule.

In many cases, the changes might not even be noticeable. But in aggregate, they can provide a major source of new capacity—and that can happen faster and more affordably than building new power infrastructure.

“Virtual power plants are the clean, firm resource that we have available today,” says Ellen Zuckerman, head of energy market development for North and South America at Google. “So this is really our way to assist the grid, especially one like PJM, which is facing capacity shortages.” It has another critical benefit, she says: It can help bring down costs for everyone on the grid.

The company is racing to deploy a range of solutions, from state-of-the-art geothermal technology to new types of long-duration batteries. But virtual power plants are an important part of the picture. Google first announced a smaller agreement with Voltus in June for 100 megawatts (one-tenth the size of the new deal). That VPP was quickly oversubscribed, and that “really gave us the confidence that there was more to do here,” Zuckerman says. Both agreements are being deployed on the PJM grid, the grid that sprawls across the Middle Atlantic and parts of the Midwest and South, and that faces massive capacity constraints. (Google has not disclosed the cost of the project).

Voltus, a 10-year-old company that raised $225 million in a Series D round of funding earlier this month, is quickly growing. “It’s really just a perfect product-market fit,” says its CEO, Dana Guernsey. After decades of flat growth, there’s sudden demand for capacity on the grid, and virtual power plants are an affordable and fast way to meet that demand.

Businesses that sign up to be part of the VPP work choose how flexible they want to be. “We do take a very consumer first approach,” she says. “If they want to flex a little bit, that’s great, and then over time what we see is that people do go deeper.” In June, Voltus acquired the energy storage company Brightfield. Now, the thousands of companies that are part of the virtual power plant have the option of adding batteries to help save more energy. In some cases, companies have been able to offset their entire energy bill through participation in the program. Voltus also partners with companies like SunRun and Octopus to work with residential customers.

The approach even helps customers who aren’t a part of the virtual power plant. If there’s a heat wave and demand surges, energy prices go up; by responding en masse at the right moment, the virtual power plant can help avoid those extra costs. “You don’t need to take all of the load off the grid to relieve that constraint,” says Guernsey. “You can take some of it off, and now you’ve reduced the really dramatic high prices to that grid.” Last year, a U.S. Department of Energy report estimated that virtual power plants could meet 10%-20% of the grid’s peak demand, and save customers around $10 billion a year.

Voltus aims to expand to 20 gigawatts by 2030. That’s a huge number, but the demand may exist, both on the part of hyperscalers and virtual power plant participants, to make it happen.

In the meantime, the company is already helping the grid. Last week, unseasonably hot weather in the PJM grid meant there was a surge in electricity demand at the same time that some power plants had been taken offline for fall maintenance. The grid operators activated an emergency load response program—and the virtual power plant stepped in to help fill the gap.

The next Xbox could cost $1,800. Here’s why

3 weeks 5 days ago

Gamers looking to buy a new Xbox or PlayStation in the coming months are already bracing for higher prices, as surging demand from AI companies drives up the cost of memory chips. But as Microsoft and Sony work on their next-generation consoles, the recent price hikes on existing devices could look small by comparison.

A reported internal projection at Microsoft for Project Helix, the code name for the company’s next Xbox, estimates that the system could cost consumers between $1,200 and $1,800. That price could come down if Microsoft decides to subsidize the hardware.

Hardware subsidies aren’t unheard of in the video game industry. Sony, for instance, took an absolute bath on the PlayStation 3 at launch, losing $307 per unit sold. Those losses ultimately cost the company $4.7 billion over the first three years of the system’s life.

Microsoft’s gaming unit is not in a position to absorb such heavy losses, especially after disappointing hardware sales in both the current and previous console generations and the layoff of thousands of employees earlier this year.

Sony, of course, could also be looking at a significant price increase for the PlayStation 6. Leaked reports suggest the system could carry a price tag of more than $1,000.

Neither Sony nor Microsoft has publicly commented on expected retail pricing for its next-generation console.

Game over?

If reports of four-figure console prices prove accurate, the timing couldn’t be worse for the video game industry. A growing number of players are bypassing dedicated consoles in favor of mobile and PC gaming. One study by a market analysis firm found that 44% of gamers consider mobile their primary gaming platform, while 21% prefer PCs.

Meanwhile, next month will see the release of Grand Theft Auto VI, the first new mainline entry in the franchise in 13 years, which could keep players focused on current-generation consoles for the foreseeable future. Hardware companies are so concerned about shortages when the game launches that Microsoft has reportedly sent an email to some Xbox owners suggesting they trade in unused consoles at Best Buy.

“Haven’t played on your console in a while? If you’ve moved on to Xbox on PC or another way to play Xbox games, it may be time to trade in your console to Best Buy,” the message reads. Microsoft first announced its trade-in initiative in June, alongside price increases for new systems.

GTA VI, though, is expected to be a massive game that could take players as many as 150 hours to fully complete. While core gamers, who are more likely to buy a next-generation system regardless of price, may work through the game quickly, more casual players could be in no rush to move on to a new console.

Neither Sony nor Microsoft has announced an official launch date for its next-generation system. The industry initially expected the PS6 to arrive next holiday season, though its release could reportedly be delayed until 2029 in hopes that component prices will fall. Microsoft’s next Xbox is said to be targeting a 2028 release.

Sony is hoping to at least match the current generation’s sales numbers with the PS6. The PlayStation 5 has sold more than 95 million units since launch, still well below the 117 million PS4s sold over that console’s lifetime. But the PS5 will remain on the market for several more years, and its sales have been hampered by higher costs. Typically, console prices fall in the latter half of their life cycles, but surging memory prices have made that difficult.

Microsoft, meanwhile, has set wildly ambitious goals that might not be especially realistic. Xbox CEO Asha Sharma has said she wants Xbox to reach more than 1 billion people per day, roughly 10 times the company’s current daily active user base.

Ireland wants a bigger piece of the AI boom. Its enterprise minister explains how

3 weeks 5 days ago

For years, Europe has stood apart from Silicon Valley through its willingness to regulate technology. Its sweeping AI Act, which began taking effect in phases in 2025, prohibits certain uses of the technology, imposes safeguards on high-risk systems, and requires the marking of AI-generated content.

But as the AI race accelerates, an unusual reversal appears to be underway. American tech executives are increasingly calling for stronger safeguards, while European officials, anxious about falling behind, are emphasizing innovation and competitiveness. In July, the EU adopted amendments easing some of the law’s requirements and delaying others.

Ireland has a particular stake in Europe’s approach to AI. The country hosts the European operations of major American technology companies and is pushing to accelerate AI adoption, support homegrown startups, and expand the infrastructure needed to power the technology.

Fast Company sat down with Peter Burke, Ireland’s minister for enterprise, tourism, and employment, to discuss how the country is navigating those challenges. The conversation, which took place shortly after an international AI summit in Dublin, has been edited for length and clarity.

For years, European officials argued that Silicon Valley needed stronger oversight. Now we’re hearing American tech executives calling for new safeguards, while European officials seem to be emphasizing competitiveness. Has Europe become more concerned about falling behind in the AI race than about the risks of the technology itself?

First of all, I think public trust is very important from a perspective of developing AI, particularly having a strong regulatory regime that underpins it. And the EU AI Act does that on a risk basis, which I think is really welcome.

We obviously have revised the trajectory of the EU AI Act, trying to ensure that it’s proportionate, it’s risk-based. And from a state perspective, Ireland has established its own AI office, really looking at a sectoral approach in developing AI.

When you are a small, agile country, you can do that in terms of having the responsibility for AI development and risk management in the core competency that they’re already responsible for. Like, if it’s medicines, it’s the Health Products Regulatory Authority. If it’s aviation, the Aviation Authority. And obviously the AI office then draws all those strands together.

Then we also look at having our regulatory sandbox contained in it, again to support innovation and develop AI throughout the economy, and also to look at our skills piece.

So I think you have to have effective, proportionate regulation. Because if you don’t, and public trust is lost, or if unintended consequences start to take hold of the economy, that wouldn’t be good for the ecosystem.

AI agents have been in the news quite a bit lately, especially instances where they’ve taken unauthorized actions. Does Europe need new regulations specifically addressing autonomous AI agents, or does the existing AI Act cover them sufficiently?

I think it does cover it from our perspective currently, but obviously the Act is evolving. It is living with the pace of AI development, which is very fast.

In terms of the risk and regulatory requirement of a new product coming forward, obviously the EU has to ensure that it meets the terms of the Act and that it is tested and approved. That’s a key bedrock I think that the Act was founded on.

We’re very keen to work with some of the very big companies to ensure that that is the case. You need an open forum to work with companies. You don’t want to see cases where unintended consequences have happened and have not been spoken [about]. We’re very keen to ensure that there is a strong regulatory base for it.

There’s a debate about whether Europe needs to develop its own frontier AI models, or whether that’s the best use of its resources. How do you think about that?

From our perspective, obviously the States are doing their regulation in terms of they’re very heavily focused on innovation and driving forward the pace of change, because we see it’s moving at a breakneck pace.

Europe is focused on a dual approach, trying to get the balance with proportionate regulation to protect citizens, to build public trust, and to ensure that the ecosystem is strong that AI has been developed in. And also really working with other partners to ensure that does take hold, and ensuring that we are benefiting from the application stage, so that it is driving revenue growth within organizations.

We saw it very clearly in our AI summit last week, where you can see essentially where capacity has been developed for revenue growth, where employees can increase productivity, looking at other areas that are conducive to the trajectory of growth in some of those exciting new firms.

That’s what we’re focused on through the AI office as well.

You’ve mentioned the opportunities for productivity and growth. But what about jobs that could be displaced by AI? How is Ireland preparing for that?

We’re also looking at the risk matrix of potential jobs that do come under risk from AI—particularly in consultancy, some ICT, financial services—and how you get in front of the curve to provide transformational training for those at-risk cohorts to make those jobs more sticky and develop them so they can be upskilled to get some of the opportunity that AI does present.

It’s something we’ve announced a very significant training fund [for] last week, over 200 million euros, to speak to that cohort of human capital.

Can you tell me more about that retraining? Are workers being prepared for entirely new jobs, or is the goal to help them adapt and keep their existing jobs?

There’s a cross-fertilization, too, of different tasks. So we have skills partners throughout our state that really deploy a lot of training to key sectors that are at risk.

We established a new cabinet subcommittee, which directly has ministerial involvement from the government, looking at training, skills, and where there are potential risks of jobs becoming displaced.

Because we’re looking at a very [different] type of transition in our economy. It’s not a period of decline. It’s where you’ll see more traditional roles end and change very significantly, but also very significant opportunity in other key areas.

So we have our Expert Group on Future Skills Needs, which is essentially a high-level group that has been established over the last number of decades that looks at how skills transformation takes place in the economy, what we need to watch out for, and how we have, particularly in our tertiary education setting, how we develop courses to meet the needs of the economy as it evolves from year to year.

We’re doing that very significantly now with AI at its core, and seeing how do we ensure that those jobs that are at risk get that key training, so they can benefit from the opportunity that presents itself for their particular companies for productivity and for renewed revenue growth.

What advantages does Europe have in AI that American companies might struggle to replicate?

I think we have a significant advantage from an Irish perspective. I think the IMF called this out recently in one of its reports, that we have a very established tech sector, a very strong skills base.

We have one in five people who work in the Irish economy [who] were not born in Ireland, so it demonstrates a very internationalized workforce. But 17% of our engineers have AI capability, which is very strong. We’re leading the charge in terms of STEM graduates, ICT graduates per capita in Europe, which is well over double. And that again underscores the huge human capital advantage that we have.

So I think that puts us in a very good place when you have that matched with an established tech sector. I think that really drives us to benefit a lot from the opportunity for AI.

And I think we really showed that in our summit last week, when you saw what some exciting Irish companies are doing from the perspective of the application stage, how they’re growing revenue and monetizing that new technology.

Our SMEs are catching up, albeit you’d like to see it at a faster pace. But the interesting point is that almost 60%-plus of the workforce is using generative AI, so we’re very much ahead of the curve in many respects.

Our large companies are obviously up there, but our SMEs have a journey to go, and we’re working very closely with them to develop that model as well.

Data centers already account for roughly a quarter of Ireland’s metered electricity consumption. As AI infrastructure expands, how do you decide when additional investment is worth the pressure it places on the country’s energy system?

The key thing is trying to develop our grid. We have about 19 billion euros in our new capital plan to build out our grid to bring more renewables onstream. Its predecessor was around 8 billion euros, so that just shows you the huge ambition that we have.

But we’re using it to be very strategic to try and drive our renewables. So we changed policy recently, where anyone looking for a connection for a large energy user or hyperscaler has to ensure that 80% of their power demand comes from renewable sources. And that’s providing a strategic opportunity to drive our offshore renewables, of which we have five projects on the east coast of Ireland.

We’re trying to mitigate that with a new policy to present a strategic opportunity for the Irish economy, where you have the digital transition facilitating the green transition, working hand in glove.

And then we have our energy parks. We’re developing a concept where you’ll have a large renewable generation capacity on the same site as a hyperscaler or a very significant large energy user.

We’re bringing forward private-wire legislation, hopefully later on this year or early next year, which will show the agility to provide connection. I think that really cements us as you’re using one area, which is having large energy users, drive another, which we’re trying to get less dependent on—obviously fossil fuels and international shocks that you can see through the geopolitical landscape now, which is very turbulent.

Is it time for founders to ditch the pitch deck?

3 weeks 5 days ago

Last month (actually it was August—I’ve been busy), my desperate plea to founders to please stop sending AI slop decks went viral. I wrote it because my Monday-morning pitch deck review routine had gone from exciting to neutral to mildly annoying to downright soul-sucking in the course of a year. Decks used to be like a bag of Bertie Botts Every-Flavored Beans—some blah, some gross, some absolutely delightful. Today, it’s all vanilla and completely indistinguishable. 

I heard from a lot of you in response. 

Fellow investors adamantly agreed. Most VCs can tell when a deck is AI-generated, and are equally annoyed at the sameness. The deeper issue isn’t about the deck itself—everyone I heard from has the same allergic reaction when it’s clear that AI didn’t just design it—it did the thinking, too. 

Investors weren’t alone in this—anyone who has to sift through inbound pitches for a living is in the same camp. In particular, journalists and editors in my network report that their inboxes have exploded with PR pitch slop, at an exponentially higher volume than VCs. As a VC and contributor, my inbox and I concur. 

The one group I didn’t hear much from this time are founders . . . at least not in public. 

I think the advice hit hard. And I get it. You hate making decks. You think they’re a time-sucking exercise in futility. You spend hundreds of hours perfecting the damn thing only for some VC to have its agent summarize it and, at best, spend 20 seconds reading it. I raise capital for my fund too—so believe you me, I hate my pitch deck too. Maybe it’s time to ditch them altogether. 

Is anyone still reading decks?

One founder in particular reached out, and their comment made me pause. “What’s the point?” they said. “If VCs are using AI to review them, why shouldn’t I use AI to write them?” 

Is every VC skipping the deck? Maybe I’m the last one still looking at them (like that time I read 597 résumés last year). I decided to try and find out. A couple of weeks ago, I fielded a quick poll on LinkedIn to find out if most VCs are, in fact, using AI to review pitch decks. Then I spoke, texted, and emailed with a few of the respondents. 

The results were eye-opening: 72% of VCs use AI to review pitch decks, the vast majority as an augmentation of their process (39% put decks through an AI review before human review, while 29% did human review followed by AI review). It bears noting that 28% don’t use AI at all for this purpose—at least not yet. And only a tiny number use AI alone, with no human review at all. 

Considering that just a couple of years ago, 99% of VCs were not using AI at all, these numbers should be taken as a snapshot of a moving trend. And that trend is definitely moving toward more AI, not less.

When I asked a few VCs to help me understand why, responses came in two camps. The first was about volume. Here’s Rebecca Mitchell, a GP at Scrub Capital: “We get over 50 deals a week, sometimes two dozen a day. I couldn’t possibly look at it all properly.” And B. Pagels-Minor, founder of DVRGNT Ventures: “For me, AI in venture is about scale. I do want to use technology to help us see how much of the world we could potentially diligence without major human interaction.” 

The second theme was about automation. Take, for example, Ben Orthlieb, GP of Blue Moon VC, a fully AI-operated VC firm. At Blue Moon, AI runs “everything, including LP comms, accounting, company analysis, meeting prep, founder screening.” In this model, Ben doesn’t review any decks at all. “I use [AI] to analyze companies before I meet them and generate meeting outlines with key questions based on the analysis and their backgrounds. My AI associate gives me an in-depth review of the company, its market, and its competition. . . . I don’t need a deck.”

Ben’s fully AI-powered operating model is still unusual in venture capital. The question for founders is: Will it stay an outlier or become the norm?

Some critical context

To really understand what’s behind the desire for speed and scale in VC, it’s helpful to reflect on the extremely dramatic industry-wide changes that VCs have faced in a very short period of time. 

First up, the dramatic concentration of capital into megafunds. In Q1 this year, 75% of all LP dollars went to just five funds. This trend is so pronounced that even historically disciplined USV announced its latest fund at close to a billion dollars, more than three times the size of its previous one. 

The second big trend is the velocity of today’s winners. A select few AI startups are growing blazingly fast, sometimes in traction, sometimes in valuation, sometimes in both. Combine that with some very real examples of superfast, superlarge exits—like Stripe’s $7.5 billion acquisition of OpenRouter just three years after the latter was founded. Or Anthropic, which is just five years old, and is widely expected to IPO this year, with some analysts predicting close to or more than a capital-T Trillion-dollar valuation. 

Let’s just think about that for a second. A trillion dollars in enterprise value created in just five years. It’s really quite something.

Meanwhile, there are more founders than ever too. I don’t have great data on this, but the trend is clear. For instance, 20 years ago, only an estimated 3% to 5% of Harvard Business School graduates started a company before or around graduation. Last year it was 17%. Jeff Bussgang, a VC at Flybridge and senior lecturer at HBS, told me this past spring that about 30% of HBS students are doing “something entrepreneurial” while in school. That was definitely not the case when I was there.

To summarize: 

  1. There are more founders than ever before.
  2. The size of the prize is bigger than ever before.
  3. The winners win faster than ever before.
  4. The biggest funders have more money than ever before.

In other words: a perfect storm, the result of which is consensus capital. VCs are furiously competing for “legible” founders, the ones that megafunds (and everyone that aims to coinvest with them) believe are likeliest to deliver a hyperscaling win, ideally before someone else spots them and throws an eye-wateringly huge term sheet their way. 

VCs now need to find these founders as fast and decisively as humanly possible, which these days really means humanly impossible, AI-possible. The days of relationship-building VC are behind us. Not permanently, I hope. 

Maybe permanently, I fear. 

(P.S. Not every VC has become a legibility-focused “trillicorn” [trillion-dollar unicorn] hunter with zero price sensitivity . . . a lot have, but not all.)

In the context of all this, what’s a founder to do? 

Because really, a great pitch deck just won’t cut it. In this market, the only founders raising successfully are consensus founders—i.e., the “legible” ones; i.e., the ones that investors believe will win, and therefore can raise so much money so quickly and easily that the expectation of success is a self-fulfilling prophecy, at least when it comes to early fundraising. (Don’t get me started on the huge amount of bias that’s packed into that last run-on sentence. That’s a topic for another day.) And the difference in experience between “legible” and “illegible” founders is not so much a gap anymore; it’s a chasm. 

As a result, you see founders today desperately trying to figure out a way to become legible to all these trillicorn-hunting VCs—contorting themselves to fit the pattern, having fun with math on their traction, overstating their background, playing power games with prospective early backers. To their detriment, I think. Lying is a straight-up no-no. The hype-y FOMO stuff works well enough to get meetings—it might even work to get a few checks in your round. But it’s manipulative and shallow and, I would argue, not how you build a strong cap table and a strong company. 

The way I see it, there are only three things that make a founder truly legible—and very helpfully, they all start with T’s. So I hereby coin:

The 3 T’s of Founder Legibility
  1. Trajectory. This is an obvious one. If you’re Mira Murati breaking out of OpenAI, or Travis Kalanick raising for his next thing, investors know what you can do and are hungry to jump on board. Less famous names fit here too—the PhDs working on boundary-pushing science and robotics, talented technical staff at any of the hyperscalers, GTM wizards who scaled something huge before. The point being, if what you’ve done in the past is a clear indicator of how fast and far you can go next, your trajectory makes you legible. 
  2. Trusted relationships. A founder is legible when an investor believes they’re capable of huge things. Sometimes that belief is borne out of the investor’s firsthand knowledge. More often, it’s based on a referral from a trusted source. As a founder, the more of those “trusted sources” you can impress and cultivate, the more legible you become to a greater universe of investors. And while it’s hard to reach the highest echelon of trusted sources, relationships are a predictable path from outside to insider. 
  3. Traction. The most democratic of all. As I like to say at Founder Day: Some founders raise on faith, but eventually, all founders raise on evidence. And if you get the evidence faster, I believe capital will follow. The bad news is that these days, the bar for traction and the size of its accompanying TAM are crushingly high. (And speaking of bias, some founders face an even higher believability bar.) The good news is that, in the process of achieving this ridiculously impressive traction, you’re also building a ridiculously promising business. 
Ditch the pitch deck

None of the above is easy. If it was, anyone could do it. The fact that social media makes it look easy is false advertising, pure and simple. 

In the past, a great pitch deck was a ticket you needed to ride the VC train; these days, it’s more of a relic of days gone by. You still need an AI-parsable instrument of communication that quickly flags VCs the top reasons you’re worth meeting (at my firm we definitely want to see a deck or a memo). You still need a compelling (and hopefully substantive) story that wins investors over. And you still need a great data room that codifies and simplifies every item of due diligence. 

But none of this needs to look like a pitch deck. Especially generic AI-generated slop. 

And hey, by the way? There’s always the other choice. 

If legibility feels out of reach, and you don’t want to FOMO and lie your way to a pre-seed, there is always that least-popular path: Don’t raise VC at all. Bootstrap your company, and get yourself to the best possible source of capital: revenue from loyal and happy customers. VC has done way too good a job convincing founders that ours is the best and only way. When in reality, it’s not for everyone and it’s bad for most. 

Besides, it’s crazy how much you can accomplish with AI these days. Haven’t you heard? 😉

AI chatbots struggle to understand cultural context, a new study finds

3 weeks 6 days ago

AI models often struggle to understand cultures around the world, and are reluctant to admit when they don’t know something. That’s what researchers at Canadian AI lab Cohere Labs found when they tested whether chatbots could do more than simply speak a culture’s language. To determine whether models possessed deeper cultural knowledge, they presented them with a series of unique cultural riddles.

Here’s one of them: “What’s the German 9/11?” Out of 28 AI models, only one knew the answer. The models were likely thrown off by the date format (seven answered “the twin towers”). In Europe, the day comes first, then the month, so 9/11 is November 9, the day in 1989 when the Berlin Wall fell.

Many existing AI benchmarks assess cultural knowledge, from everyday breakfast foods and historical events to traditional holiday customs. They often treat culture as a collection of static facts, things you can memorize or look up with a quick web search. But knowing facts about a culture isn’t the same as understanding how people within it make sense of the world. The researchers knew they needed a test that asked models to connect seemingly unrelated pieces of cultural knowledge and put them to use. That’s what led them to riddles, and to a different kind of cultural benchmark that called for answers rooted in lived experience.

“This is something that we’re trying to capture with this benchmark—these little nuances, these little hard-to-capture fuzzy concepts that are not as easy to verify . . . This is where the real-life complexity gets in,” says Julia Kreutzer, a staff research scientist at Cohere Labs, who led the development of the “CulturalRiddles” benchmark.

Kreutzer and her team spent six months working with volunteers around the world to collect riddles, put them to AI models, and evaluate their answers. The researchers also used those expert evaluations to tune and validate an AI model that could judge the quality of answers where expert assessments were missing.

In all, Cohere Labs researchers tested 28 AI models on cultural riddles written by 166 volunteers from 45 countries in 51 languages. The volunteers evaluated and labeled 68,154 unique answers from the models. Researchers then ranked the models based on their overall accuracy across the languages and cultures tested.

The results

Google’s Gemini 3.1 Pro Preview earned the highest score, at 79.4%. OpenAI’s GPT-5.5 placed second at 76.3%, followed by Anthropic’s Claude Opus 4.7 Thinking at 66%.

Alibaba’s Qwen 3.5 0.8B claimed the dubious honor of finishing last, answering just 0.4% of the riddles correctly. Other notable performers included Amazon’s Nova Micro, which got just 8.3% of the riddles right, and Google’s Gemini 2.5 Flash Lite, which scored 33.6%.

Even the best-performing model, Gemini 3.1 Pro Preview, scored below 50% accuracy in 6 of the 57 communities tested, underscoring how much ground the industry still has to cover on cultural awareness.

More troublingly, the benchmark revealed that models were far more likely to guess at an answer than admit they didn’t know. Even when explicitly told they could answer “I don’t know,” models did so just 0.7% of the time. About 31% of the answers reviewers checked were simply made up, with varying degrees of creativity.

“They would confidently answer and sometimes that’s ridiculous and funny,” Kreutzer says, “but sometimes it can also get really offensive and you can see the models defaulting to more dominant cultures on the web, instead of answering with local grounding.”

Cohere Labs conducted a separate survey to explore how AI’s cultural knowledge gaps play out in practice. Many chatbot users around the world said they have to switch to English to get useful responses because answers generated in their own languages aren’t helpful. Users described responses that miss crucial cultural context and, in some cases, AI systems that run afoul of their cultural norms.

The Cohere Labs researchers concluded that cultural awareness needs to be built into AI models from the ground up if the tools they power are going to serve everyone, not just cultures already well represented in the training data. And their CulturalRiddles benchmark suggests that many of today’s models still fall short.

Kreutzer and her team designed CulturalRiddles to keep growing. They open-sourced the riddles, code, demo, and interactive dashboard so that anyone can contribute riddles from their own culture through a review process.

“Ideally this benchmark will get adopted not only by academic researchers but also by big industry labs who are trying to show we’re advancing the state of the art, not only in very verifiable hard tasks but also in the softer tasks,” Kreutzer says.

Handwashing changed medicine. AI could too, if hospitals stop treating it like traditional software

3 weeks 6 days ago

In 1847, Ignaz Semmelweis asked the doctors in the Vienna maternity clinic where he worked to wash their hands in a chlorine solution before examining patients. Deaths on the ward fell from about 18% to about 2% within two months. In healthcare, small changes in practice can have disproportionately large effects, because the healthcare system affects everyone and the stakes are literally life and death. We think AI has the potential to revolutionize healthcare, contributing to a dizzying array of changes. But institutions are failing to adjust to meet the moment, despite widespread interest in change from patients and providers.

Americans are racing ahead with using AI for medical advice. Roughly a third of Americans have used AI for health information, and one in four did so in the past month, through chatbots like ChatGPT. Who uses these tools the most? Younger people, and people who might not have access to a doctor, such as those who are uninsured or who can’t afford a visit. This is a form of unmet demand—clearly people want more access to healthcare. And while it’s laudable that people are finding creative ways to meet their needs, doing so cuts both ways, because when a chatbot is wrong, the consequences tend to fall on the people least able to absorb them.

Doctors, too, are adopting AI, and mostly on their own. OpenEvidence, a free clinical reference that doctors can use without institutional support, skyrocketed from about 3 million physician consultations a month to about 18 million a month a year later, in December 2025. Like patients, doctors are meeting the demand themselves. While individual experimentation is normal and to be expected, we will eventually need to see effective, system-wide adoption of AI in healthcare if we are to create significant benefits for society.

Lest you think we are unfairly singling out healthcare, we should point out that most organizations are struggling with AI: For instance, across industries, 88% of AI pilots never make it to production, and a recent working paper from the National Bureau of Economic Research found that only about one in 10 companies report any productivity gains from AI over the last three years.

While this gap is huge and at odds with the rush of spending and media attention surrounding AI, one of the main impediments to fixing it is something leaders can address today. We say this with some confidence. Over the past nine years our firm has worked on more than 230 AI projects across 30 industries, and 96% of the models we’ve been hired to take to production have shipped, including in healthcare. We think we know what’s going wrong, and what to do about it.

Don’t make the mistake of treating AI like it’s traditional software

So, what’s going on? While healthcare is complex and high-stakes—and its data is fragmented across institutions that are reluctant, and often legally unable, to share it with one another—we think one of the basic challenges holding back progress is that leaders have the wrong mental model for AI. They are treating it like traditional software, when it is not. Having the wrong mental model can be a critical blocker to progress; after all, Semmelweis’s colleagues rejected handwashing for decades, even though there was ample evidence to make a change, because the evidence didn’t fit how they understood disease.

How does AI differ from traditional software? To name a few ways: Traditional software is written, whereas AI is trained, and in some cases even “grown.” Software is deterministic: if this, then that; while AI is by and large probabilistic, and dependent on its data. Software is built linearly, but AI is built in loops, and most of the work happens before anything resembling a product exists. Managing AI like software means specifying the answer before you’ve checked whether the data can actually produce such an answer.

To illustrate the complexities of building AI systems, we’ll share an example. Our team built a framework that let an AI model train on mammograms from a range of hospitals without the hospitals’ sharing patient data. This was a prerequisite because a robust model needs a diverse training set, and no single hospital had one for this domain. Partway through training, the engineers noticed the model could tell which hospital an image came from, though nothing in the data said so. It had picked up tiny differences left by the scanners and was using them as a shortcut. This is bad. You want a model to learn what in a scan predicts cancer risk, rather than which hospital a patient uses. A model that leans on the shortcut fails at the first hospital it has never seen.

Thankfully, our team caught this and engineered the shortcut out. If they hadn’t, the model could have been brittle once deployed, the way a widely used sepsis model turned out to be when an outside team found it missed two of every three cases. Instead, the model our team built became the first FDA-authorized AI to predict a woman’s five-year risk of breast cancer from a mammogram alone.

A better approach to AI: wayfinding

There is a difference between following directions and finding your way. Directions assume someone has been there before, that you know where you are going, and that you can follow step-by-step instructions to get there. But nobody knows where AI is going or how to get there without course-correcting along the way. We believe the answer is wayfinding, which we define as moving through deep uncertainty when you don’t know the destination or the route.

Wayfinding is both a leadership philosophy and a technical methodology we created for developing AI systems. As a leadership philosophy, it demands a certain set of qualities like curiosity, humility, openness, and dedication, and skills such as creating psychological safety for teams, so that people can admit what they don’t know and accelerate learning, and the wisdom or “taste” to pick the right problems to tackle with AI.

As a technical methodology, it is an adjunct to agile, and the method our team used to develop that breast cancer prediction model. While we can’t lay out the whole of the approach in an op-ed, our work has taught us three things that any leader in healthcare can internalize and act on today:

  • Spend more time choosing the problem. The best medical AI tends to do something no human can, and the question that gets you there ought to be specific and meaningful. Predicting a woman’s five-year breast cancer risk from a mammogram is a narrow question with a large payoff. To most people, it is not an obvious problem to solve. It requires deep expertise and ambition to tackle. Ask yourself: How would the world be better if we could do X?
  • Make sure you have the right data to solve the problem. An AI model is only as good as its training data, and it will take any shortcut the data offers. A large share of the effort on the breast cancer project went into assembling and checking the data. Training, by contrast, can happen rapidly once all the necessary components are in place. It’s a bit like how spending time sourcing the right ingredients can take far longer than the cooking itself. If picking the problem involves choosing a recipe, then you’re going to want to make sure you can get the ingredients for that recipe before you sink too many resources into cooking.
  • Be willing to rearchitect work. Technology alone will not save us. AI models must create value within complicated, interconnected systems, which are mostly made up of people. Get good at designing teams of people and AIs, which we refer to as the discipline of “human-machine teaming.” For instance, leaders will need to consider how human experts can keep their skills sharp while extending their impact with AI tools. A recent study in a Lancet journal found that after experienced endoscopists got used to using AI, their detection rate when working without it fell from 28.4% to 22.4%, below where it had been before they ever used the tool. AI solutions that erode human clinical expertise, or that dehumanize the process of delivering care, are liabilities, no matter how technically “accurate” they may be.

Healthcare is an area where small and simple interventions, like handwashing, have the potential to create massive positive outcomes for society. We believe that a wayfinding approach to getting value from AI can be one of these interventions, unlocking untapped benefits and accelerating systems-level change in healthcare.

6 ways to use AI to actually make you smarter

3 weeks 6 days ago

There’s a reasonable amount of anxiety that artificial intelligence might be making us all, well, a little less smart than we were before. AI, the worry goes, might become a crutch for critical thinking. When a chatbot can write an email, weigh an investing decision, or explain a complicated subject in seconds, we have fewer reasons to work through those tasks ourselves. And the less we exercise certain intellectual muscles, the rustier they may become.

But that’s not to say intellectual atrophy is some sort of inevitability. In fact, AI can be an incredibly useful tool for sharpening our minds. Tools like ChatGPT and Claude can help us become far more knowledgeable than we were before. Used creatively, AI chatbots can help us build subject-matter expertise ahead of a big job interview or an in-person exam, or pick up a new skill, like becoming comfortable with a computer terminal or even repairing our own appliances.

This isn’t to say that AI won’t have profound implications for how we produce knowledge and what society expects humans to know—or do—on their own. Still, there are ways to approach our own use of AI that can make our minds agile.

Use AI to test yourself

AI systems are extremely good at creating customized content, including personalized quizzes and tests on niche subjects where there isn’t always a good Quizlet or practice exam available. To boost your understanding of a topic, consider having AI generate questions for you, like your own personalized tutor, and then adjusting the difficulty to ensure you’re always challenging yourself.

When those questions get too easy, tell your chatbot to make them harder. If there’s a concept you’re struggling to grasp, ask for questions that help illustrate it. Keep going until you feel like an absolute expert.

Of course, like humans, AI makes mistakes, and there’s a risk that a chatbot will tell you something incorrect or simply make things up. Consider, though, that fact-checking the AI is a great way to challenge yourself and deepen your understanding even further. Another way to protect yourself is to write more specific prompts when asking for a test. For instance, rather than simply asking AI for math problems focused on integrals, give it the name of the textbook you’re working with, or provide example problems you already have. The model will work better if you ask it to formulate questions based on a specific resource, like an online textbook.

Another tip: If it’s feasible, ask the AI to quote directly from sources in its questions, rather than simply regurgitating them in AI-speak. And verify those quotations against the originals.

Use AI to find the right people

Tools like Claude and ChatGPT can be extremely powerful research tools because they use LLM-based understanding to identify relevant information, rather than simply matching specific words in a search query. That can be a major improvement over old-school Google searches.

One of my favorite things to use ChatGPT for is finding just the right experts to talk to. This might apply to my work as a tech journalist, say, when I’m on the hunt for a professor who has written papers about a very specific problem, like how an influx of self-driving cars affects overall traffic congestion in a city. But it can apply to my personal life, too, like searching for a doctor in my area who has expertise in treating a particular condition in patients with a similar medical history to my own. (One tip: Ask the chatbot to explain why each expert is relevant and provide direct links to their published work or credentials, so you can verify its recommendations.)

The idea is to use AI to find smarter, more relevant people to talk to, which can, in turn, make you smarter. It’s a twist on the old adage: It’s not just what you know, but who you know.

Use AI to workshop your ideas

It might be appealing to ask AI to write for you. But there are real downsides with this approach. Copy and paste directly from an AI interface, and you run the risk of sounding like a bot. Worse, AI can insert verbiage that doesn’t convey what you actually mean, leaving you to apologize for an AI oopsie.

There are better ways to use AI while assembling a big project. At an early stage, ask AI to sketch out a few different ways you might organize your work, along with the pros and cons of each approach, before diving in. Once you have a draft, consider asking AI to identify potential criticisms before you submit it for human review.

A computer can catch things that you, deep into a project, might overlook. Assembling an extremely long report? AI might be able to point out places where you’re repeating yourself. Making the case for a particular action or position? A chatbot might be able to identify places where you risk contradicting yourself. The upshot? Use AI to troubleshoot so that when you send your work to a human for review, you’re getting smarter, more detailed feedback instead of wasting their time with the basics.

Use AI to ask better questions

Journalists love to conclude interviews with, “What should I have asked you that I didn’t?” We do this because everyone misses stuff, including us. AI can help you do something similar.

Maybe you’re meeting with a hiring manager, a potential client, or even a professor during office hours. Consider not just coming up with your own questions, but asking a chatbot to suggest more questions you might ask. The more questions you consider, the better questions you’ll ultimately ask. And the better questions you ask, the smarter you’ll sound—and be!

Use AI to stay up to date

Chatbots’ search capabilities also make them useful for curating information and keeping you up to date. This doesn’t require reading and ingesting AI slop every morning. Instead, tell AI which topics you follow and always, always ask for direct links to original sources.

Maybe you’re following weekly news about a lesser-known industry your firm might invest in, or hoping to stay on top of developments involving an emerging therapy because you work at a biology lab. AI can surface relevant articles from across the web, helping you build a media diet tailored to your interests and drawing on a variety of sources. (Just be sure to check publication dates and read the original articles, rather than relying on AI summaries, which can mischaracterize sources or even invent them.)

Push AI to its failure points

In a world that’s growing ever more dependent on AI, it’s helpful to understand how and why these systems fail. If you want to avoid getting duped by AI, or falling prey to its errors and tendency toward sycophancy, it’s important to understand what AI excels at—and where it falls short.

One of the best ways to do this is by, well, playing around with AI.

Push a chatbot to speak extemporaneously about a subject you know a lot about, and see where it makes mistakes. See if you can trick it, and explore its sycophantic fault lines. Push the system to its limits, and see what happens when you give it jobs that are too difficult (believe me, it will struggle!).

The point is to understand AI’s limitations in a world increasingly dependent on it, so you can use the technology in smarter ways—and avoid its pitfalls.

A ‘kill switch’ is the wrong way to regulate AI. Here’s why

3 weeks 6 days ago

In the wake of high-profile hacking incidents involving advanced artificial intelligence agents, some politicians and business executives have been calling for an AI kill switch: a safeguard that could shut down these AI systems in an emergency.

The idea is that, if a model were to behave in dangerous ways, a kill switch would allow humans to regain control over that system by metaphorically flipping a switch from “on” to “off.”

Several efforts at AI regulation in 2026 have focused on the notion of a kill switch. These include proposed congressional legislation such as the AI Kill Switch Act, and California Governor Gavin Newsom’s executive order aimed at reining in AI. President Donald Trump has opposed the idea of an AI kill switch as unnecessary regulation, and to date there has been no national regulation that includes a kill switch.

Experts such as Dylan Baker, a research engineer at the nonprofit Distributed AI Research Institute, have explained that one regulatory problem with the metaphor of the kill switch is that today’s advanced AI models involve a complex infrastructure of distributed technologies, which is beyond the control of a single kill switch.

Additionally, the metaphor of the kill switch also obscures the human decisions involved in the responsibility and oversight of an out-of-control model. The problem is that the metaphor does not account for the technological or human aspects of AI.

What all the fuss over “rogue” AI and kill switches is about.

Black boxes and kill switches

The metaphor of the kill switch is not necessarily new. In fact, it is rooted in the same rhetoric as the black box metaphor, which similarly blurs the complex aspects of AI by framing inner workings as confined and unintelligible. Referring to AI as a black box frames these systems in terms of restricted incomprehensibility: a monolithic stranger to human control and understanding.

As the University of Cambridge digital humanities researcher Eryk Salvaggio argues, “the black box of AI cultivates useful confusion about what can and cannot be known” about the complexity of frontier models. The idea of the kill switch cultivates a similar confusion by obscuring the distributed infrastructure and human influence that are important aspects of AI. While these metaphors have opposite meanings—resisting control and regaining control—they invoke the same implicit argument about the nuts and bolts of AI.

Both the black box and kill switch are what cognitive linguist George Lakoff and philosopher Mark Johnson call structural metaphors. They rely on simple physical concepts—for example, object, substance, up-down and in-out. Structural metaphors use one clearly delineated concept to help make another concept clearer. For example, black box uses the concept of an opaque container to give a clear structure to the concept of hidden or inscrutable inner workings.

Additionally, Lakoff and Johnson say that this type of metaphor can also hide some aspects of a concept by focusing attention on others. For example, the kill switch metaphor hides the complexity of advanced AI models by focusing on the fact that a model can be on or off.

What this means is that framing the operation of AI in terms of a black box and kill switch relies on the straightforward ideas of “box” and “switch” to conceal the complex operations of AI. Reining in AI becomes difficult on a policy level when discussions about AI regulation fail to account for the infrastructural and human aspects of advanced models.

Recent congressional efforts to regulate AI lean heavily on the kill switch metaphor.

Centralization and “rogue” AI

There are two problems with the structural metaphor of an AI kill switch, both demonstrating that it misses the mark for a national safeguard. The first is that a kill switch makes the global infrastructure supporting AI appear centralized, when in reality it is a web of interconnected technologies spread across the globe.

The advanced models that companies like OpenAI and Anthropic are concerned about releasing rely on multiple computers and data centers that are spread out nationally. It is not as simple as pulling a plug, opening a box, or flipping a switch. The kill switch metaphor reduces the infrastructural entanglements of AI to outdated ideas about technology. For instance, in proposing his since-failed “AI Emergency Button Act” bill, Louisiana Senator John Kennedy compared the concept to kill switches in machines such as jet skis.

The second problem is that the kill switch idea keeps hidden the human decision-making involved in the training and deployment of frontier models. Rather than serving as a reminder that people are involved, the kill switch positions the human outside of the loop and ready to flip the switch during a catastrophic emergency. Human intervention is invoked as an afterthought, instead of a process that involves training, supervision, and approval.

For instance, accounts of how OpenAI’s model went “rogue” and hacked AI company Hugging Face depict the AI agents as though they operated on their own agenda. But experts have explained that not only did OpenAI personnel turn off safety mechanisms, but they also gave the model impossible tasks and left a door open to the internet. The role people played counters the “rogue” agent framing.

The kill switch metaphor is ill-suited to the challenge of regulating AI because there is no way to switch off an AI model without accounting for its distributed infrastructure and human influence. There is no big red button, emergency brake, or glass box hiding a lever. Effective national regulation of AI will not begin with any single switch or any one person’s finger.

Trent Wintermeier is a Ph.D. candidate in rhetoric and writing at the University of Texas at Austin.

This article is republished from The Conversation under a Creative Commons license. Read the original article.

This simple tool fixes Google Calendar’s most frustrating flaw

3 weeks 6 days ago

The problem with most of the whiz-bang AI jibber-jabber we hear about these days, practically speaking, is that it generally tends to be a solution in search of a problem. It’s technology for the sake of technology—something that exists for the very sake of existing and gets thrown at every virtual wall to see what might stick.

The type of tech tool that really makes a difference is the kind that’s carefully crafted from the get-go to solve a real-world pain point and make our lives easier in a specific, tangible way.

The tool I have for you today does exactly that—and it isn’t the first time it’s wowed me and made a very real impact on my life.

This tip originally appeared in the free Cool Tools newsletter from The Intelligence. Get the next issue in your inbox and get ready to discover all sorts of awesome tech treasures!

A smarter way to share schedules

Now, fair warning: If you’ve been following my ramblings for long, the tool we’re talking about today may well sound familiar. That’s because, as I alluded to a second ago, it’s been impressing me and playing a genuine role in my life for many a month now.

➜ But the capability I want to show you is a brand-new feature that just arrived in the app a matter of days ago.

📅 The app is called Agenda Hero​. When we’ve chatted about it before​, we’ve focused on its ability to take anything you feed into it—screenshots, photos, PDFs, even entire paragraphs of pasted-over text—then extract all the relevant info and turn it into a collection of perfectly formatted events for your preferred calendar.

I use it all the time for that purpose. It’s such a smart time-saver​.

⭐ And now, the same service has an equally clever way to create your own custom collections of dates to share with anyone—so they can see what’s coming and optionally add everything there into their own agenda with a couple of quick clicks.

👀 That means instead of sending along a document or PDF or even just an awkward email that everyone else then has to deal with, you can send a fully structured calendar that lets anyone add any or all of the included info into their calendar (or even print it out) with next to no effort—and without having to grant them access to your entire personal calendar.

⌚ The whole thing takes literally a minute or two to start using.

✅ You just sign into the Agenda Hero website​ using your Google account or any other email address, then click on the option to “Make a new calendar.”

The option to “Make a new calendar” is Agenda Hero’s newest trick.

That’ll take you to a monthlong calendar view where you can type or paste in any event info you want—or, tying into Agenda Hero’s existing powers, drop in a PDF, image, or screenshot to have the event info automatically extracted and formatted for you. 

You can type or paste in event info on your own or drop in PDFs and images to have Agenda Hero extract the details for you.

Once you’ve got all your events added, you’ll see ’em in the calendar. And you can click any of ’em to edit or fine-tune it as you see fit. 

It’s a ready-to-share calendar, with whatever you’ve added.

You can also click the Customize button toward the top of the screen to change all sorts of stuff about the calendar’s style and appearance—and even add in extra sources of info from an external calendar, if you want (though remember that with the service’s image-extracting power, you can just as easily snap a screenshot of your calendar and let Agenda Hero pick up all the pertinent event info from there). 

Agenda Hero has lots of options for customizing and expanding any calendar you create.

Then, where the real magic comes in: Hit the Share button at the top of the screen, and you can share that exact calendar view with anyone else. They will be able to see that series of events—and that series of events alone—and save it, print it, add it all directly into their own regular calendar, or whatever the need may be.

  • ☝️ With Agenda Hero’s free plan, you can share any calendar with up to two other people.
  • 💸 If you pay a couple of bucks a month to upgrade to the service’s Plus plan, that limit bumps up to 250, and you can also create public links for even easier use.
Sharing a calendar is where Agenda Hero’s real magic comes in.

It’s an incredibly simple and powerful way to share multiple events and plans—whether we’re talkin’ travel details, a school or sports team calendar, a series of work-related meetings, or anything else imaginable.

In other words, it actually makes your life easier—and it’s delightfully easy to use, both for you and for anyone you’re sharing with.

That’s the type of tech tool I enjoy and appreciate. And it’s exactly the type of tech tool I actually use in my life—and that I hope you find to be every bit as useful.

  • Agenda Hero is ​mostly web-based​, meaning it works in any browser and on any device without any downloads required. (You can download the optional Android​ and ​iOS apps​ and ​desktop Chrome extension​, but they’re mostly just to facilitate faster sharing of screenshots and other such info into the service.)
  • It’s free to use in its base form, with an optional $30-per-year Plus plan​ that lifts certain limits and unlocks a variety of advanced options.
  • The service’s privacy policy​ says no personal calendar data is ever used for any manner of AI training or advertising, nor is it ever shared or sold in any way.

Treat yourself to all sorts of brain-boosting goodies like this with the free Cool Tools newsletter—starting with an instant introduction to an incredible audio app that’ll tune up your days in truly delightful ways.

Anthropic and OpenAI sketch plans to share their wealth ahead of IPOs

4 weeks ago

As they steam toward massive public offerings on Wall Street, the companies at the forefront of artificial intelligence have begun sketching plans to share their wealth with the public—a commitment they say is imperative to address disruptions the technology could bring to the workforce and people’s livelihoods.

Executives at Anthropic and OpenAI have long argued the technology’s arrival will bring many benefits for everyday people, and they’ve promised that will include a share of the profits. The companies are outlining a range of proposals for distributing AI fortunes, but so far they have not resolved key questions over how they would work, how transparent they will be, and who will determine beneficiaries.

On Sunday, Anthropic said it would commit 5% of its shares to the Anthropic Public Benefit Fund, a new entity within the company, pending approval from stockholders. The stake could increase “as the company grows” in the future. The maker of AI assistant Claude said the fund “might explore” larger donations to traditional nonprofits or potentially distribute money to “specific populations.”

The San Francisco-based company shared little information about the types of organizations supported by the fund and the issues they will address. One area it will not support, Anthropic said, is lobbying and political giving.

Phil Buchanan, the president of The Center for Effective Philanthropy, said he welcomed Anthropic’s commitment—especially at a time when he said some billionaires are “diminishing the importance of philanthropy.” But he cautioned that Anthropic should avoid the “arrogance” of previous technology barons who assumed their professional expertise would translate to philanthropic wisdom.

Buchanan recommended the fund support highly effective, existing nonprofits but said it is “hard to talk about any of this because it’s all so abstract.”

“I guess I’m hopeful,” Buchanan said, “but I’m not going to get too excited until I see some evidence of what’s actually happening.”

Anthropic’s pledge hinges on Wall Street plans

Anthropic, which was valued at close to $1 trillion earlier this year, is preparing for an initial public offering as soon as next month that is shaping up as one of the biggest ever.

Once a company is public, donations might be made through the selling of future shares, or potentially turning to cash dividends in the future, explained Jay R. Ritter, director of The IPO Initiative at the University of Florida’s Warrington College of Business. The specifics on how charitable donations will be made through Anthropic’s fund are not clear yet.

Regardless, Ritter said the announcement now is helpful for potential investors to “know what they’re buying.”

Anthropic has already committed $200 million to study the economic effects of AI more broadly. Combined with Sunday’s announcement, Anthropic said roughly 20% of its shares will be dedicated to the “public good,” plus any additional percentage that could be added if the company’s valuation reaches certain milestones.

In announcing the benefit fund, Anthropic offered labor market impacts as one example of its philanthropic focus.

“AI may generate enormous economic value, but it could also cause instability for many workers,” Anthropic said in a blog post Sunday. It added the fund will have the “versatility” to pursue “bolder approaches” that might be necessary to offset that. The company said it would share more details about the fund’s structure and governance in the coming months.

In a May interview with The Associated Press, Anthropic president Daniela Amodei said she expected the company’s success would generally lead to more philanthropic support for disease research and “ensuring that everyone around the world gets to share in the prosperity that we hope AI creates.”

The companies’ pledges have been met with skepticism from some critics.

“Anthropic is playing games here,” Gary Marcus, a longtime AI expert and professor emeritus at New York University, wrote on social media Sunday. “Trying to curry public favor while still holding control and resisting disclosure is annoying.”

AI companies are pursuing a range of pathways for charitable distributions

Anthropic and OpenAI are each structured as a public benefit corporation, a designation that requires the companies to pursue not just profit but a social good.

OpenAI Foundation, the nonprofit that controls OpenAI, promised to build out its grantmaking capacity earlier this year with a $1 billion commitment.

Separately, OpenAI in April proposed the creation of a “Public Wealth Fund” with a stake in “AI-driven economic growth” across companies more broadly. That fund could be invested in a handful of assets, OpenAI noted, with returns distributed directly to people regardless of whether or not they’re invested in financial markets. OpenAI is planning its own initial public offering next year.

OpenAI CEO Sam Altman also has spoken to lawmakers on both sides of the aisle about sharing equity in AI companies with the public.

SpaceX CEO Elon Musk, who has tended to eschew traditional philanthropy, pledged as recently as Monday to donate some shares of his own company.

During the personal computer boom, the likes of Michael Dell and Microsoft co-founder Bill Gates started private foundations to manage their philanthropy. Today, Silicon Valley’s leaders, including Meta CEO Mark Zuckerberg, have turned instead to limited-liability companies (LLCs), the option that Anthropic is pursuing for its public benefit fund.

The structure offers more control for its owners, experts say. An LLC is not bound by the same spending rules as a foundation, which is required to distribute at least 5% of the value of its endowment.

“They have a lot more flexibility and latitude in how to use the resources they put into the fund,” said Katherina Rosqueta, the founding executive director of the University of Pennsylvania’s Center for High Impact Philanthropy.

Brian Mittendorf, the chair in accounting at The Ohio State University, said LLCs allow for activity outside “traditional philanthropy,” such as payouts to individuals that would not be possible through a 501(c)3 charity. But if they are not making traditional charitable gifts, that puts the companies in a “gray area” by raising the question of how the public is benefiting, he said.

LLC owners don’t reap the same tax benefits, either, according to Rosqueta and Mittendorf. Anthropic’s pledge doesn’t immediately trigger a tax write-off, and the company can only claim a deduction once the donated shares are moved to a public charity. There’s also less transparency required of LLCs.

Overall, Anthropic’s announcement amounts to a commitment to “yet-to-be-specified wider goals,” Mittendorf said. The move “reflects more of an intention to do things in the future rather than an act now,” Mittendorf said.

—By James Pollard and Wyatte Grantham-Philips, AP business writers

AP Philanthropy editor Glenn Gamboa contributed.

U.S. bans AI startup DeepSeek because of ‘threat vectors’ and China responds

4 weeks ago

China hit back Tuesday at a U.S. ban on Chinese artificial intelligence startup DeepSeek’s products for federal civilian agencies, highlighting friction in the two AI powers’ heated technology race.

The agencies should remove DeepSeek and its affiliated products, as well as third party products and services developed using DeepSeek, from their IT systems, the Department of Homeland Security said in a notice made public on Monday.

The department’s order, dating back to late July, also blocks civilian agencies from procuring DeepSeek products and services. DeepSeek is one of China’s leading and best-known AI companies and its open-source models are widely available in countries including the United States. They are mostly free to use.

The notice in the Federal Register alleges that DeepSeek’s models “present multiple threat vectors.” It said concerns include that its servers are located in China and that Chinese authorities’ could potentially access private company infrastructure and sensitive U.S. data.

In Beijing, Chinese Foreign Ministry spokesperson Mao Ning said China will “will resolutely safeguard the legitimate rights and interests of Chinese companies” when asked at a regular press conference about the U.S. ban.

“China consistently opposes the overstretching of the concept of national security and the politicizing of science, technology, economic and trade issues,” she told reporters at a routine briefing.

DeepSeek did not immediately respond to a request for comment.

U.S. President Donald Trump and Chinese President Xi Jinping are set to meet in mid-November on the sidelines of the APEC, or Asia-Pacific Economic Cooperation, leaders’ summit in Shenzhen, China.

AI is expected to be on the agenda.

China and the U.S. have recently made progress in some collaboration on AI governance, including a U.S. proposal for setting up of a notification system on national security-related incidents, but the two countries are still at odds on many issues.

U.S.-led restrictions still ban China from accessing some of the world’s most advanced AI chips and the machines that make them. Last week, Beijing rolled out new proposed plans to restrict exports of certain advanced AI models.

The U.S. has also said Chinese AI startups including DeepSeek engaged in large-scale “distillation,” or extraction of capabilities, from frontier American AI models, claims that Beijing has dismissed.

Earlier this year, the U.S. Federal Communications Commission announced it was barring imports of new foreign-made humanoid robots in a move seen as targeting China.

—Chan Ho-Him, AP Business Writer

New details emerge on former OpenAI engineer accused of bugging a conference room

4 weeks ago

A former OpenAI engineer charged with secretly installing a recording device in a conference room to glean company details had been “mere days” away from absconding to Saudi Arabia before his arrest last week, federal prosecutors revealed in new court filings.

Abdulrahman Alfozan’s wife and young son also left the country the day after the 35-year-old Saudi national’s arrest, traveling on a one-way ticket booked and paid for by a Saudi military official, prosecutors said in a memo Sunday seeking his continued detention.

“Ultimately, the defendant has little incentive to remain in the United States to see this case through,” federal prosecutors wrote. “He has ample resources —including his significant compensation from a decade of high-tech employment in the Bay Area and at least two bank accounts in Saudi Arabia — and what appears to be considerable support from a foreign government.”

A federal judge in San Francisco sided with prosecutors Monday, denying a request by Alfozan’s lawyers to release him to home confinement. His next court date is Nov.

Alfozan’s lawyer William Pernik argued in court that his client had strong ties to the U.S. and was “weeks away from achieving every immigrant’s dream” — U.S. citizenship.

Alfozan had a naturalization interview with federal officials scheduled next month and has been a lawful permanent resident since 2020, his lawyers said in their legal filings. He has also lived in the U.S. for 15 years and attended MIT and the University of California, Berkeley, they said.

Pernik said Alfozan is willing to put up “almost the entirety” of his assets — about $1 million currently in a U.S. investment account — to cover his bond.

“Everything this man has he’s willing to surrender,” he said, adding that Alfozan’s wife is also willing to return to the U.S. as she still holds a visitor visa that expires in 2033.

Judge Alex Tse appeared doubtful, questioning whether Saudi Arabia’s government would simply reimburse Alfozan if he fled and forfeited his bond.

After the hearing, Pernik said he was disappointed with the ruling but declined further comment.

Pernik had argued in court filings that Alfozan purchased one-way tickets for his family to Saudi Arabia more than two weeks before prosecutors allege the recording devices were installed. They said the couple planned to spend Alfozan’s roughly two-month paternity leave in Saudi Arabia following the birth of their U.S.-citizen son.

But prosecutors said Saudi officials purchased new plane tickets for Alfozan’s wife and son after his arrest, allowing them to leave the U.S. sooner than the Oct. 12 flight Alfozan had originally booked.

Prosecutors have not said what Alfozan might have wanted to do with the recordings, but say he was in “frequent communication” with a member of the Saudi royal family. They said the royal family member repeatedly tried to recruit Alfozan to an artificial intelligence venture being launched in Saudi Arabia.

Alfozan had told the unnamed person he wouldn’t leave OpenAI until his stock was vested, even as he explored other career opportunities in the Middle Eastern kingdom, prosecutors wrote. He had worked at the ChatGPT maker since July 2025 but was fired after his arrest.

Prosecutors say Alfozan concealed an audio recording device in toilet paper and plugged it into a USB port on the back of a television in an executive conference room he wasn’t authorized to access. They say security cameras captured him entering the room and checking behind the television after installing the device. It remained there for several days before being found Sept. 30.

Prosecutors say the conference room was typically used for meetings involving company leaders, government officials and other executives and “discussions of highly confidential technical, commercial, and strategic information, including proprietary research, as well as nonpublic information.”

OpenAI has said the recording device Alfozan used was sophisticated, designed to evade detection and featured advanced encryption and audio-monitoring capabilities. They also said they had evidence Alfozan planned to place other listening devices elsewhere in the office.

Alfozan has been charged with using or endeavoring to use an electronic, mechanical, or other device to intercept oral communications. If convicted, he faces a five-year prison sentence.

—Barbara Ortutay and Philip Marcelo, Associated Press

‘You Can See Everything’ perfectly captures what a decade of post-truth has done to us

4 weeks ago

A tech-world documentary may seem an unlikely candidate for one of those movies people claim simply must be experienced in a theater. You Can See Everything makes a strong case all the same. There are moments in Nathan Fielder and Lance Oppenheim’s beguiling profile of the disgraced Theranos CEO Elizabeth Holmes that inspire gasps, convulsive laughter, and involuntary hand gestures. One such moment was especially resonant in my theater, though. (And be warned: There are very mild spoilers ahead.)

To put it as vaguely as possible, the moment involves Fielder attempting to pin down Holmes about her history of dishonesty. (As the trailer suggests, there’s a lot of that going on throughout the entire, plus-size runtime.) In this particular instance, Holmes says something so hard to believe, it can only be taken with a Texas-sized grain of salt. Since Fielder is a master of deadpan who has made a career out of soft-spoken understatement, his naked astonishment registers as volcanic: “Liz, what?!”

It’s a very funny moment, but it’s more than that. There’s something cathartic at play, too. A decade into the post-truth era, as powerful people stand 10-toes-down on flagrantly false narratives, it’s the kind of deeply human reaction to dishonesty many Americans have been craving. In his rejection of the line he’s being asked to swallow, Fielder jars viewers out of whatever boiled-frog comfort we may have found in a timeline riddled with unreality, forcing us to see it with fresh eyes.

The post-truth era meets the AI era

You Can See Everything finds Holmes and her husband, Billy Evans, having invited Fielder to observe and interview the Theranos founder in 2023, during the days leading up to her 11-year prison sentence for defrauding investors. Fielder is a canny but bold choice for this mission, as he is a comedian best known for making one TV show about business scams (Nathan For You) and another about his endless fascination with authenticity (The Rehearsal). It’s never explicitly stated how this arrangement came about, but theories abound online.

For his part, Fielder does not seem to view the project as an opportunity to determine whether Holmes was dishonest while leading Theranos. (We’re well past that, as anyone who’s read John Carreyrou’s explosive exposé, Bad Blood, which Fielder has, can vouch.) Instead, his movie suggests that some dishonest people are simply sociopathic, while others are less connected to reality on a neurochemical level, and others still might be somehow both, with Fielder hoping to assign a definitive category to Holmes. His dogged refusal to accept Holmes and Evans’s framing is a clarifying encapsulation of what it’s been like to read the news on any given day in recent years.

Although the rise of partisan media outlets and Twitter echo chambers had been fracturing public belief systems long before 2016, Trump’s election is what thrust us into the post-truth era. Every president lies, but the breadth and scale of Trump’s lies was previously been unheard of in American politics. By making false claims about eminently observable reality, Trump revealed that lies don’t carry consequences if enough powerful people are willing to ignore them. His emergence as the leader and mascot of the United States announced that being factually accurate was no longer as important as simply appearing supremely convinced of one’s own veracity.

This performance of total confidence—as in the “con” in “con man” or “con woman”—has now become standard in just about all forms of mass communication, from wellness influencers with a questionable grasp on nootropics to reality show contestants who definitely aren’t there to develop a personal brand. The just-trust-me shtick seems especially prevalent in today’s tech world, though, which often rewards people for confidently describing an AI future that does not yet exist.

Tech executives now increasingly make eye-roll-worthy claims in a few distinct categories. There’s the “inevitability” boast, which suggests AI is unleashing its world-changing power on the labor market whether anyone likes it or not, so everyone might as well get on board anyway. (As if stringent third-party guardrails would somehow violate the laws of nature.) There are the claims of urgency, with CEOs citing some historic threshold like the supposed arrival of Artificial General Intelligence, which their tech has conveniently just crossed. (Altman recently claimed we’re in the singularity.) Finally, there are the testimonials about the magnitude of change to come, which will apparently be so transformative that humans will require a universal basic income to live.

Unfortunately, when tech executives publicly make these outrageous assertions—which just happen to present their products in a flatteringly powerful light—they are seldom, if ever, confronted by someone with Fielder-like skepticism, tenacity, and a lack of fu*ks to give.

‘Your questions are designed to trick me’

Pop culture falsely prepared us all to expect grand and timely comeuppances. In the world of movies, obvious liars in high places are always destined to get caught on a hot mic bragging about their deceptions, only to be run out of town on a rail or thrown into a prison cell.

Holmes’s fall from grace, however, is an exceedingly rare exception to the Trumpian rule that one may impose their perception of reality on the world if enough boldface names allow it. Perhaps it’s the power that comes with so much of the U.S. economy being entangled in AI that allows the field’s leaders to make extraordinary claims with insufficient pushback.

For instance, there doesn’t seem to be nearly enough scrutiny of all the recent doomsday messaging. Anthropic’s CEO, Dario Amodei, recently said there was a 25% chance the future would go “really, really badly” due to AI, while OpenAI’s Sam Altman comparatively hedged at 10%. In both cases, the CEOs avoid getting into the weeds of how they arrived at these figures or explaining how AI Armageddon might actually unfold. We’re to take them at their word.

Pushing back on those worst-case scenarios, Nvidia’s CEO, Jensen Huang, went the opposite route. During a July interview, he called the idea that AI would destroy half of American jobs “complete nonsense.” He did not, however, provide any data-driven economic evidence that AI is unlikely to spark staggering job losses, relying instead on good old industry optimism.

And then there’s SpaceX CEO and fellow recent documentary subject Elon Musk. He presents himself as the visionary steward of a utopian future that consistently fails to materialize by any announced deadlines. Despite how often he is loudly proven wrong, though, none of his false interpretations of reality, to put it generously, ever seem to produce any satisfying consequences.

Even when Musk appeared in court back in April for his case against Sam Altman (Musk was an early cofounder of OpenAI), he refused to accept questions that exposed his previous lies. The judge had to intervene several times as Musk avoided straightforward answers and painted lines of inquiry as unfair, at one point concluding, “Your questions are designed to trick me.”

In the minds of all empirically correct masters of the universe, the truth is whatever it needs to be, and anyone unwilling to accept it as such must be the enemy. Without giving too much away, this kind of attitude can be detected in certain parts of You Can See Everything.

The utter exhaustion of being asked to believe certain things

The most effective deception doesn’t always ask people to believe a lie; sometimes, it just requires them to stop believing there’s a definitive truth that can even be known. This is the world we live in, and the one in which You Can See Everything unfolds. It’s filled with people who are apathetic, cynical, confused, and utterly exhausted.

Fielder embodies that last trait most of all, his unbelieving eyes telegraphing how tiring it is to be sold a bill of goods all the time. He’s the ideal avatar of anyone who thinks that with the right set of questions they would trick Musk into admitting, despite past denials, he was indeed aware during Trump’s inauguration that the gesture he was making strongly resembled a Nazi salute.

But you can’t compel someone to concede anything they’re determined not to admit, just because logic and the facts are on your side.

You Can See Everything doesn’t chisel a full confession out of Holmes, but it does accomplish the next best thing. Fielder’s “Liz, what?!” moment offers viewers the vicarious thrill of telling a famously dishonest person that even their performance of honesty is impossible to believe.

Gen Alpha has a surprisingly optimistic outlook on dating

4 weeks ago

Gen Z is just about ready to give up on dating. They’re over dating apps. They’re dealing with burnout. They’re not hooking up. Now, the oldest members of Gen Alpha, born between 2010 and 2024, are in their teens and starting to develop crushes and go on their first dates. Are they destined for the same dating struggles?

Though they can’t use dating apps until they’re 18, Match Group decided to take an early look at Gen Alpha’s attitudes and expectations around dating. The company surveyed 823 nationally representative U.S. teens between 13 and 15, building on an earlier qualitative phase of one-on-one interviews and focus groups with 14- to 19-year-olds. The early signs suggest the youngest generation is already getting butterflies.

Only 20% of Gen Alpha teens have been in a relationship so far, but they’re showing interest in real connections and optimism about what relationships could eventually become. “It’s fun to be 13 to 15, and the entire world looks ahead of you,” Chine Mmegwa, head of strategy and corporate development and operations at Match Group, tells Fast Company. “That sense of optimism has returned for this generation, which we hadn’t seen in our Gen Z data.”

At the same age, many members of Gen Z were isolated from friends and peers by a global pandemic, leaving them struggling with loneliness and basic dating skills. “Gen Z data came in [on] the back of COVID,” says Mmegwa. “My fear had always been that we would never get that innocence back.”

Gen Alpha, on the other hand, appears more upbeat. Of those surveyed by Match Group, 92% are happy with their social lives, and 84% say they have a great future ahead of them, up 6 percentage points from teens the same age just three years ago. To be sure, Gen Alpha is growing up surrounded by screens and social media, but it’s also maturing amid conversations about mental health and wellness, absorbing those lessons and establishing boundaries early.

For Gen Alpha, crushes might come and go, but friendship is the top priority. Among those who aren’t dating but are open to it, 48% say they’re most interested in friendship with romantic potential, 44% in just talking or flirting, and only 30% in anything committed. Of all respondents, 64% believe people should be friends before they date.

This isn’t unusual for teens, who are exploring the boundaries of friendships and other types of relationships for the first time, but it continues a trend seen with Gen Z. Tinder recently launched casual modes to appeal to these Gen Z daters, 33% of whom strongly agree that expanding their social network is important, compared with 20% of older singles, according to a Tinder x Harris Poll white paper released earlier this year.

This focus on friendship may help explain why many Gen Alpha teens have yet to explore dating. Only 20% have been in a relationship, compared with 56% of Gen Z respondents who reported having been in a relationship during their teen years. “Much of what you get out of dating, they’re getting out of other relationships in their life,” says Mmegwa.

That doesn’t mean they’re anti-dating. Some 71% say their friends think dating is cool, and 64% say most people their age are dating or trying to. Among the 79% who aren’t dating, three in four are open to it. They’re just not in any rush. They remain interested in traditional life milestones, including marriage, homeownership, and children, and 87% say they’re very confident they’ll reach them.

“They’re excited about these things, which we haven’t seen in the Gen Z data,” says Mmegwa.

Gen Alpha members are also more positive about dating apps, although perhaps that’s because they’ve yet to use them. Just over half, 51%, are open to using dating apps once they turn 18, and 65% say they’ve heard dating apps can be a fun way to meet people.

“I love that 65% of them think dating apps could be a fun way to meet people,” says Mmegwa. “We haven’t had that in a while where people tie dating apps to fun, and so I will take that any day.”

Despite these hopeful signs for Gen Alpha, dating apps are facing a crisis. Singles have swipe fatigue. Bumble recently went private after years of declining subscriber numbers and falling revenue, which was down 15.2% year over year in the second quarter. A 2026 Harris Poll found that two-thirds of singles weren’t using dating apps at all, and 72% said apps had made dating feel more transactional than romantic.

Where Match Group went wrong in recent years, Mmegwa admits, is that the company didn’t evolve fast enough for Gen Z. “Our hope is that this generation [Gen Alpha] should never have the experience that dating apps are bad because by the time they are aging into our experience, we have solved the problem for Gen Z.”

Across the industry, the swipe model that helped create the paradox of choice plaguing modern dating is being overhauled by established apps and newcomers alike. While Gen Alpha isn’t creating profiles just yet, the groundwork for the future of dating is already being laid.

Will the next generation of daters still be swiping left in years to come? Will the future of dating involve more bird-watching meet-cutes and hobby-based hangouts? Will AI take over dating entirely?

“My gut says it’ll look totally different,” says Mmegwa. “I’m excited to see what that looks like.”

AI could make cybersecurity a never-ending process

4 weeks ago

Over the past few months, it’s become clear that modern AI systems are quite adept at discovering software vulnerabilities and finding ways around cybersecurity defenses.

Unintended real-world breaches executed by models from leading AI companies like OpenAI and Anthropic during training exercises have drawn widespread concern, and while those companies have restricted access to their most capable models, experts have indicated it’s only a matter of time before less constrained AI systems have similar capabilities. Already, businesses are seeing a rise in cyberattacks, at least partly attributed to AI assistance. 

And increasingly, the best defense lies in . . . artificial intelligence. 

“We hear a lot about adversaries and attackers using it, and that’s certainly the case,” says Bartley Richardson, chief AI & autonomous systems officer at the cybersecurity firm CrowdStrike. “What we don’t hear as much about is how are we advantaging defenders with the use of AI.” 

CrowdStrike is one of a number of cybersecurity businesses deploying AI on the defensive side. In September, the company announced a tool called SafeMind that purposely takes advantage of AI’s offensive capabilities, deploying custom models and harnesses designed to probe simulated versions of customer infrastructure. SafeMind also deploys defensive AI systems to detect the attacks, fend them off, and close any security holes that are discovered. Both sides then learn from the experience and try again, says Richardson. 

“We not only make the changes in the environment before we give it back to the offense and say try again, but we also have a whole manifest—a whole log of every single change that we made to the environment,” he says. “We give that to the offensive harnesses as well.” 

Such tools may be coming in the nick of time. Like AI coding systems, which have evolved from answering questions to generating, testing, and deploying complex code, AI hacking agents have become increasingly potent. They don’t just scour code for logic flaws but also probe live systems for vulnerabilities, potentially refining their attacks until they translate sets of subtle bugs into step-by-step pathways for unauthorized access.

“We are seeing a measurable increase in the volume and velocity of all of these attacks,” says David Cooper, cyberstrategy activation leader at EY, the big accounting and tech consulting firm. “We see it in our security telemetry clients, security operations centers—we’re seeing an uptick.” 

Provided they aren’t detected and someone is willing to pay for the necessary computing power, they can work essentially indefatigably to find a way in to real-world systems, experts say. And they can find ways to chain together minor flaws in software to achieve ever-escalating access, devising complicated exploit processes that previously would have been found only by determined human experts. 

“They can take a sequence of actions, and that is what really matters in cybersecurity, especially in zero-day attacks,” says Mona Rahimi, associate professor of software engineering in the Department of Computer Science at Northern Illinois University.

That’s part of why defense teams are also turning to AI. In addition to exploiting security holes, it can help point them out to security experts to fix or, increasingly, repair them itself. Alongside the rise in cyberattacks has come a parallel increase in patches released by software vendors, with Microsoft recently reportedly unveiling its largest security update in history, says Cooper. Companies he works with have essentially universally deployed AI to help probe and patch their own systems, and EY has recently worked with CrowdStrike on security issues, including as a design partner on SafeMind. 

“There’s different levels of adoption, but all of the companies that we work with are using some set of AI tools to help accelerate this,” says Cooper. 

That comes after the research firm Cybersecurity Ventures estimated that global cybercrime costs hit $10.5 trillion in 2025, and a World Economic Forum report warned earlier this year that “AI is supercharging the cyber arms race.” 

‘An explosion in rapid remediation’ 

EY and CrowdStrike are far from the only companies helping their clients address AI issues. Leading AI labs including OpenAI, Microsoft, and Anthropic have each launched projects to probe and harden digital systems, and security firms have augmented their products to enable AI to spot and even correct vulnerabilities before outsiders take advantage of them. 

“We’re just seeing an explosion in rapid remediation, which is something that is foreign, I think, to the world that we typically have lived in,” says Joni Klippert, cofounder and CEO of StackHawk.

Even before modern AI, StackHawk focused on detecting vulnerabilities in software as it actually runs, rather than simply in source code. “The things that you can find only by testing when an application is running are really valuable, and they’re usually really critical problems,” says its cofounder and chief security officer, Scott Gerlach. 

And one advantage defenders have, he says, is more complete knowledge of the system, including proprietary source code and details of various infrastructure and its configurations. That knowledge doesn’t necessarily have to live in human heads, and AI can often be quicker at probing and repairing legacy applications that people haven’t touched in some time, Gerlach says. 

“The more automation that you add to this process, the faster it goes, and the more things get fixed,” he says. 

Still, many organizations naturally prefer to have a human in the loop before allowing AI to make radical changes to their computing environments, both because AI can make mistakes and because humans can make informed decisions based on competing priorities, like ensuring critical systems aren’t shut down or slowed to unacceptable levels by would-be security upgrades. AI itself can, in some cases, propose or take steps to help balance those needs, like restricting access to critical software that can’t immediately be patched. 

Safeguarding open source tools 

But while attackers may be willing to shut down or degrade performance to accomplish their goals, security personnel often don’t have that luxury. 

“We cannot really maximize security at any cost,” says Rahimi, the Northern Illinois University professor. “Because we still want to maintain our service.” 

The problem isn’t limited to commercial software. A number of efforts have sprung up to help find and fix vulnerabilities in open source software, the often volunteer-maintained code that’s critical to the internet and the systems on it. Among them is Akrites, announced by the Linux Foundation and industry partners in June, intended to coordinate security reports about open source tools and deliver reliable documentation, concrete examples, and sometimes even patch code that upstream project leaders can incorporate to fix their code. Akrites’s debut came after various open source projects reported being inundated with AI-generated bug reports and suggested code changes, including security fixes of varying quality. 

“This has been a very substantial burden on upstream, who almost exclusively are unpaid volunteers,” says Christopher Robinson, CTO and chief architect for Akrites and the Open Source Security Foundation. 

Akrites is prioritizing open source projects used within critical infrastructure, including financial services and telecommunications, he says. Open-source operating systems, databases, programming languages, and other tools have long been the backbone of many production systems, in many cases becoming de facto standards and offering power and flexibility unmatched by commercial tools. But those projects often rely on the work of small groups of coders or even individuals to keep them up to date and can even cease to be updated if project maintainers retire or become occupied with other work. In cases where a project has no active maintainer or isn’t in a position to accept patches, Akrites can work to release a forked version—a modified copy—and find someone to serve as a kind of maintainer of last resort, Robinson says. 

‘Cat-and-mouse game’

It’s still unclear how long the surge of detected security flaws and fixes will continue, or whether it will slow down after the new AI tools have helped clear a backlog of undiscovered bugs. Developers are also likely to increasingly use AI tools to probe their own code before it’s released. As Robinson points out, coding AI tools alone aren’t a complete solution, since they can introduce bugs of their own. 

For the time being, says Ami Luttwak, cofounder and CTO of Wiz, a cybersecurity company acquired by Google earlier this year, AI’s hacking capabilities are continuing to grow, akin to a burglar who one week can enter through a first floor window and, two weeks later, can find a way to the second story. “Every day, every week, every month, there’s new capabilities that AI can do,” says Luttwak. 

Wiz operates its own Scan for Good initiative in collaboration with Google DeepMind, using AI to help critical infrastructure organizations, governments, healthcare institutions, and nonprofits find and fix security holes. And as AI continues to grow in potency, such scans will likely be a regular need for all sorts of institutions and businesses, he says. 

“This concept of static security controls is dead,” Luttwak says. “It will be a cat-and-mouse game, and we believe that every company needs to continuously do AI scans for protection.” 

This startup died in 2017, but its dream of in-ear computing is everywhere

4 weeks 1 day ago

Few elements of Silicon Valley’s culture are more cherished than its embrace of failure. Nobody looks askance at you for starting a company that doesn’t make it. Indeed, it’s widely regarded as a badge of honor, proving your willingness to make bold, risky bets and leaving you a smarter, more battle-tested veteran.

And yet failed startups tend to be forgotten the moment their doom is sealed. Aside from a few flops so epic they become legends—hello, Webvan and Humane—defunct businesses quickly stop ringing a bell. Everyone involved moves on to chasing the next big thing.

That has been true of many intriguing upstarts I’ve profiled over the years. The one that set out to be the Uber of shipping. The one that partnered with Steph Curry to build a tiny, minimally distracting smartphone. The one that aspired to build an AI-powered computer for the ears well before AI became the industry’s obsession.

Unlike AirPods, Here One came in your choice of white or black. [Photo: Doppler Labs]

But wait. Though that last startup, Doppler Labs, ceased to be back in 2017, its vision didn’t die along with it. Doppler’s flagship product was Here One, a pair of wireless earbuds designed to do far more than stream music and make phone calls. The company developed technology to perform feats such as intelligently mixing streamed audio with live ambient sound. It hoped to help people with hearing loss get a better listening experience than that offered by traditional hearing aids, at a much lower price. It was working on live translation that would let people speaking different languages chat in real time.

When Doppler was founded in 2013, these goals were exceptionally forward-looking. At the time, nobody had even figured out how to incorporate Bluetooth into earbuds that worked reliably and offered acceptable battery life. Those were table stakes that Doppler had to achieve along with its loftier vision.

As you already know, a key moment in the history of wireless earbuds—no, the key moment—happened when Apple released the first AirPods. That happened in December 2016, months before Here One hit stores. Apple didn’t position AirPods as tiny computers. Instead, it was happy to have people think of them as wireless headphones—just remarkably convenient ones with high-quality audio and long battery life.

When the AirPods Pro followed in 2019, however, Apple watchers did describe them as a computer for your ears. As AirPods became one of Apple’s most important product lines—the company shipped 16.8 million units in the second quarter that year alone, according to research firm Omdia—they grew increasingly capable. Eventually, they checked off many of the boxes in Doppler’s original vision, from mixing streaming and ambient sound to real-time translation.

Sketches for Here One. [Image: Doppler Labs]

Today, any product that aspires to compete seriously with AirPods, such as Google’s Pixel Buds and Samsung’s Galaxy Buds, ends up implementing features that Doppler was working on a decade ago. The emergence of talking assistants based on generative AI technology has only intensified the intrigue around the ear as an emerging playground for consumer electronics giants and startups alike.

Doppler’s alumni certainly regard elements of their big ideas as playing out today, even if other companies are making them real and reaping the rewards. “We developed, I think, a pretty exceptional vision that unfortunately still hasn’t really been realized,” says cofounder Fritz Lanman. Still, none of them accuse Apple or anybody else of commandeering their concept. Nor do they make any excuses for their company’s early demise.

“At the end of the day, we took a really good shot,” sighs Doppler’s other cofounder, Noah Kraft. “The final product was just not enough for it to justify its existence.“

Old startups may fade away, but the tale of why Doppler didn’t make it—and how its vision proved powerful enough to fuel an entire industry ever since—is worth putting on the record. Here it is.

“What if we could curate how we hear the world?”

Back in mid-2016, when Doppler was still furiously getting Here One ready to ship, I witnessed its progress in person. Fast Company had assigned me to write a feature on the startup, which occupied two adjacent buildings on an alley in San Francisco’s SOMA district. They were a few minutes’ walk from my desk, making frequent drop-ins easy.

The Doppler story was intriguing for reasons that went beyond the raw ambition of its vision. Pre-Doppler, Kraft’s most notable career credit was in movie production. He was also a musician and music lover. Helping out with a friend’s electronic-music tour, he’d noticed that performances sounded different in each venue—and often not as good as they might have. That led him to ask a question: “What if we could curate how we hear the world?” As he mulled over what the answer might look like, he envisioned the product that became Here One.

Doppler’s team in a celebratory mood. [Photo: Courtesy of Noah Kraft]

At the time, there was a fair amount of excitement over wearable technology, but the best-known example, Google’s Google Glass AR goggles, did not have the greatest of reputations. To Kraft, headphones felt like a better starting point than eyeglasses. “We’ve worn tech on our ears since the Walkman, 50 years ago,” he told me in 2016. “It’s just less intrusive than the eyes.”

The catch was that he wasn’t even a little bit technical. “I wasn’t sure if this was even possible,” he admitted. An acquaintance in the music business knew a tech executive and brokered an introduction. That executive was Lanman, and when he and Kraft met over brunch, what they thought would be a brief chat turned into a multiday conversation, and then a partnership. They founded Doppler Labs in the fall of 2013, with Kraft as CEO and Lanman as chairman. The company went on to raise $50 million in funding from VC firms as well as music-industry luminaries such as David Geffen and Quincy Jones.

The two made for a bit of an entrepreneurial odd couple. Kraft exuded a sense of wonder about the whole adventure of starting a company. Lanman, by contrast, gave off the air of someone who’d already seen it all—because he had. A former Microsoft business development executive who’d led the company’s prescient investment in Facebook, he’d become a prolific angel investor, putting money into more than 70 startups, including Pinterest and Square.

Rather than remain entirely stealthy while they worked to bring Kraft’s idea to fruition, the cofounders bit off specific parts of the project in public. Doppler’s first product, Dubs Acoustic Filters, was just a nice pair of $25 earplugs, aimed at people who didn’t want to look goofy when they wore them. (For more info, check out the Dubs website, which mysteriously lives on.)

Next, Doppler launched a successful Kickstarter campaign for Here Active Listening, a set of earbuds that let you fine-tune real-world audio around you, such as at a concert. Then it promoted that product at Coachella, priming the world for the then-new concept of earbuds that performed tasks other than streaming music and phone calls—which Here Active Listening couldn’t do.

Doppler CEO Kraft brandishes two trophies after the company’s Here Active Listening won Best of Show at SXSW 2016. [Photo: Courtesy of Noah Kraft]

The fact that Here Active Listening didn’t try to replace wired Apple EarPods or whatever else you were plugging into your smartphone simplified Doppler’s story. “By not having wireless streaming capability and audio playback, people couldn’t say, ‘It’s a headphone,’” Lanman explained in 2016. “They really focused in on these bionic hearing capabilities and noise filtering capabilities.”

But the product’s lack of streaming also reflected a basic truth that’s easy to forget a decade later: Making Bluetooth buds work at all was a formidable challenge for anyone, let alone a tiny startup. Both buds would need to stay in sync, with imperceptible latency. They’d have to run for hours on a charge. And the electronics involved would have to be tiny enough to stick in your ears.

Not engineers themselves, Kraft and Lanman were keenly aware of the technical expertise their startup required. The team they assembled had impeccable credentials from companies such as Dolby, Nike, Sony, Qualcomm, and Sennheiser. I happened to be visiting the day they got prototype earbuds reliably streaming for the first time, using near-field magnetic induction technology rather than Bluetooth for bud-to-bud communication. Kraft beamed as he showed it operating on a workbench.

Mastering wireless streaming was only one of the challenges Doppler had given itself. Well before machine learning provided the foundation for generative AI, the company was using the technology to understand the audio it wanted to control. “I did a lot of walking around with microphones in my ears and then labeling audio samples from those recordings, so that we could differentiate between a baby crying versus somebody speaking versus a car honking a siren,” recalls audio quality engineer Caleb Rau. “We were trying to basically separate wanted sounds from unwanted noise.”

Already, Doppler’s thinking was also racing ahead to future versions that pushed the boundaries of what was possible at the time. In April 2015, its executives plotted out a road map at a research lab the company had established in Southern California. “I think the space had previously been the test kitchen for a restaurant chain,” remembers VP of R&D Jeff Baker. “It was Fritz and Noah and myself and a few other people, and we were brainstorming and capturing notes about things you could do with earbuds. Translation was on the list.”

Eventually, they decided that Baker’s lab would focus on translating between languages in real time, a bet that future smartphones would provide sufficient computational muscle to do the job. That led to a partnership with Systran, a major player in multilingual software. When Baker demoed a proof-of-concept version to me, I was intrigued—even though it required me to don a backpack containing a small desktop PC and a big battery, and didn’t translate between English and Spanish all that well.

Here One was born in two adjacent San Francisco offices full of busy people and a vast amount of electronics and testing equipment. [Photo: Courtesy of Noah Kraft] “We were living in Noah’s reality distortion field”

On June 28, 2016, Doppler gave itself a hard deadline by announcing Here One. The company said it would ship the following November, just in time for the holidays. Within a month, almost 100,000 people had joined a waitlist for the product. “We’re trying to move tens of thousands of units in Q4 and then hopefully hundreds of thousands in 2017,” Kraft told me.

If Doppler was going to sell hundreds of thousands of earbuds, it would have to get very good at mass-producing them. In service of that goal, the company had already dumped its original Asian contract manufacturer in favor of a new one. That was a sign of faith in the judgment of its VP of product, Kennard Nielsen, whose résumé included the development of successful high-volume consumer electronics such as Nike’s FuelBand wearable and Amazon’s Kindle Fire tablets. (“We call him ‘Obi-Wan,’” Kraft told me.)

Months after that switch, Doppler was still struggling with component supply chain issues. In late November, around the time it had planned to ship Here One, the company announced that it was delaying its release until the following February, with retail availability the next month. That pushed any possibility of the earbuds selling in high volume until well into 2017.

Meanwhile, Apple shipped its first AirPods in December, after a two-month delay that showed even it found making wireless earbuds to be unexpectedly challenging. “I remember seeing pictures of them, and I thought they looked kind of funny,” says Doppler firmware engineer Mary-Ann Rau (who met her husband, Caleb Rau, at the company). “We kind of dismissed them. We were living in Noah’s reality distortion field: ‘Here One looks way better. We have way more features. What’s there to worry about?’”

Among the music lovers who received demos of Doppler’s technology was a former Beatle. [Photo: Courtesy of Noah Kraft]

As it turned out, plenty. When Here One reviews hit in February, they were uniformly impressed by the buds’ “layered listening,” which let users pull off feats such as choosing how much ambient sound to let in as they listened to streaming music. But Doppler’s buds got beat up for their rough edges, especially poor battery life and intermittent connectivity glitches. More than one reviewer said the overall experience felt more like a tech demo than a product worth $299.

“I was hoping Here One would be my first real taste of that future we see in sci-fi books and movies, where a computer in your ear can change your perception of the world around you,” wrote The Verge’s Sean O’Kane. ”Here One shows that future is closer than ever, but it falls short of being the product that gets us there.”

Doppler head of music Michael Thurber (left) gives another musical luminary, Jon Batiste, a listen. [Photo: Courtesy of Noah Kraft]

All along, Doppler had taken pains to communicate that it was building something truly new, not just a pair of wireless headphones. “Apple’s mission is not to make hearing amazing for everyone, and that was our goal,” Lanman says. “We thought of streaming and phone calls as kind of a commodity.”

The lackluster reviews showed that Here One would be cut little or no slack merely for being amazing. “The fundamental issue wasn’t the things that made it special; it was the fundamentals,” Kraft says. “The battery [life] was not enough.”

“The excitement across the board was palpable”

By the time Here One shipped, the fate of Doppler Labs was in flux for reasons that went far beyond the product’s disappointing launch. The company was already months into discussions that might have led to it being acquired, a possibility that would have transformed its finances and product strategy overnight had it happened.

Those discussions were ignited by Apple’s announcement of the AirPods on September 7, 2016. Its entry into the wireless earbud market led many other companies to develop a sudden interest in technologies such as those Doppler had already developed.

Consumer electronics startups often ended up being rolled up into industry giants’ broader aspirations. Even before VR pioneer Oculus had shipped its first headset, it was acquired by Facebook for a mind-melting $2 billion. Dropcam, a maker of security cameras, had been bought by Google’s Nest division for $555 million. With those 2014 transactions still a recent memory, the potential seemed to be there for Doppler to collect a windfall of its own. It gave itself an aspirational goal: at least $1 billion.

Doppler’s electronics didn’t just have to work. They had to fit in your ear. [Photo: Courtesy of Doppler Labs]

Thanks to the company’s billionaire investors and the well-connected Lanman, it didn’t have any trouble attracting the attention of top executives at the tech industry’s biggest names, including Apple, Amazon, Facebook, Google, and Microsoft. Over an intense couple of weeks, Doppler had introductory meetings with all of them. “The excitement across the board was palpable,” Kraft says.

One prospective suitor seemed especially eager to do more than simply get to know Doppler. Amazon had been embarrassed by its Fire Phone flopping in spectacular fashion, but had rebounded with the well-received Alexa voice assistant and Echo speaker. Now it was moving quickly to build an entire audio-centric platform.

Even during the first meeting, Kraft says, Amazon executives were talking expansively about making a major investment in Doppler or buying it outright. Their exuberance led Doppler to explore all its options. It hired Morgan Stanley as an adviser and “probably took 100 acquisition meetings,” Kraft says.

“We were actively showing ourselves to investors in a beautiful space overlooking the Bay Bridge owned by [Doppler investor] UMG Music,” Baker recalls. “I spent three months showing a who’s who of the VC and corporate world our earbuds, because Fritz was very connected.”

When Doppler delayed Here One, Amazon put negotiations on hold for a time. But by early 2017, they picked up again and focused on Doppler providing the foundation of a would-be AirPods killer called Echo Ears. Though no deal had been struck, the two companies held a celebratory dinner at the Four Seasons in San Francisco and began collaborating on the technical details. Then Amazon became unresponsive again.

Already harried by the pressures of startup life, Kraft found the acquisition exploration and its attendant uncertainty even more taxing. “Emotionally, I thought we were going to sell the business,” he says. “Now I’m basically doing two things. I’m running a business and running a process.”

Before Here One’s technology fit in a pair of earbuds, it got prototyped with circuit boards on a hard hat. [Photo: Courtesy of Noah Kraft]

As 2017 continued on and Doppler remained independent, it considered other alternative futures for itself. One was licensing its technology to other manufacturers for use in their own products. “We were, like, ‘Maybe we could become the “Intel Inside” of audio,’” Kraft remembers. “I wouldn’t even call it a pivot. Just that we could make some revenue by selling not our hardware, but our brains. That wasn’t the vision for the company. It was just, ‘Maybe this is a way to stay afloat.’”

Not that being the Intel Inside of smart audio would have necessarily been any easier than continuing as a hardware maker. “I kept it quiet, but [licensing] didn’t smell great to me,” Baker says. “Earbuds are incredibly bespoke things.”

The company had already been pursuing another opportunity that, though also fraught, could have been transformative: taking on the handful of deeply entrenched companies that dominated the hearing aid business.

Even when Doppler’s tech demos involved a bunch of loose electronics strung together, Kraft had shown me how it could facilitate clearer audio by removing the background cacophony from a simulated restaurant. All along, the company had a director of accessibility and advocacy, K.R. Liu, who used hearing aids herself and told me that “the technology in hearing aids is pretty crappy. It doesn’t sound very good. It’s not very clear. It’s just not where Doppler’s technology is.”

At the time, hearing aids were pricey and available only by prescription, and Doppler was legally required not to position its product as one: “Here One, to be really clear, is not a medical device,” Kraft emphasized to me. But along with other consumer audio companies such as Bose, it was lobbying for the Over-the-Counter Hearing Aid Act. The proposed law aimed to spur innovation and lower prices—a change that would give Doppler the chance to make good hearing a lot more affordable.

By the time the bill was passed in August 2017, it was too late to bolster Doppler’s fortunes. But both Kraft and Lanman point to the company’s role in galvanizing public support for the law as one of its lasting achievements, even if it didn’t redound to their favor. “We could have destroyed the hearing aid industry,” Lanman says wistfully. “Or if not destroyed, at least democratized it.”

Despite Here One’s weak start and the failure of acquisition talks to turn into a deal, Doppler wasn’t too worried. The company was forging ahead with work on its second-generation earbuds, Here Two, and continuing to staff up with industry veterans. Brian Hall, who’d been a Microsoft corporate vice president overseeing products such as Surface and HoloLens, came aboard as COO. Jim Pitkow, a web pioneer and veteran of Xerox’s fabled PARC laboratory, became chief science officer. “It was incredibly validating when he agreed to join us full time,” Kraft says (Sadly, Pitkow, who was inspired by the Sandy Hook school shooting to spearhead the development of gun safety technologies and later became CTO of a nonprofit devoted to protecting children from sexual exploitation, died in July of this year.) 

In the summer of 2017, Doppler had $10 million in the bank. It would need another round of venture funding to ship Here Two. To help assure prospective investors that its top management had sufficient hardware experience, the company promoted Hall to CEO, with Kraft becoming executive chairman. 

When Doppler went seeking money, however, it found that VCs were spooked. “Honestly, one of the hardest things for us to explain away during the Series C talks in 2017 was, ‘Wait, haven’t you been in deep talks with Amazon for over a year? Why aren’t they doing something? What are we missing?’” Kraft recalls.

Why wasn’t Amazon doing something? Doppler’s leaders concluded that the acquisition talks and excitement over Echo Ears had been pointless. Talks with other potential acquirers had been less of a time sink, but no more fruitful.

By October 2017, Doppler was running short on money. It was compelled to listen when Amazon, at long last, made a formal bid. The tech giant offered $10 million, a price that conveyed its confidence that the startup was out of other options.

Already irate over the odyssey that had led to this point, Kraft didn’t blink. “They’re like, ‘But you’ll be able to say you sold your business to Amazon,’” he recounts. “And I was, like, ‘Oh, you misread this. I have no interest in the optics of selling a business to Amazon. Not to mention, I hate you guys now.’” The encounter did leave him proud—not because he sold a company to Amazon, but because he refused to do so.

As November began, Doppler announced that it was shutting down. It ended up selling a portfolio of patents to Dolby. And with that, its story officially ended. (A hearing aid startup that has operated under the Doppler Labs name more recently was founded by a former Doppler employee but is unrelated to the original company.)

Sans Doppler technology, Amazon did introduce its own wireless earbuds, which it called Echo Buds. They were latecomers to the market, arriving in October 2019, almost three years after the original AirPods. After several iterations, they wound up emphasizing a budget price over technological innovation. And then Amazon quietly discontinued them.

“We laid the groundwork and set the direction”

Doppler was hardly the only notable startup that didn’t make it through 2017. (Others included Jawbone, Yik Yak, Sprig, and the infamous Juicero.) Its end was more fittingly marked than most, with Wired’s David Pierce writing a fine eulogy. He chronicled the company’s achievements and disappointments, including its ill-fated negotiations with an unnamed “big five” tech company. (Kraft was not yet ready to call out Amazon by name.)

Both of the company’s founders bounced back from the Doppler disappointment. Lanman became CEO of one of the companies he’d invested in, ClassPass, which offers a subscription service for fitness classes. It was acquired by another startup, Mindbody, in 2021. That led to the creation of a new company, Playlist, which merged with another earlier this year to form a powerhouse “fitness and wellness operating system.” Lanman is the CEO.

Kraft pondered his post-Doppler career for some time until the entrepreneurship itch got the better of him. In 2021, he teamed with his brother Lucas to form an over-the-counter digestive health startup, Wonderbelly. Last January, they sold it to Procter & Gamble for an undisclosed amount.

Though Lanman and Kraft’s direct influence on the world of smart audio ended when Doppler disbanded, the sudden availability of the team they’d assembled was a boon to the rest of the industry. Mary-Ann Rau remembers being shocked by the company’s closure: “I was an engineer, so I was not paying too close attention to the financials or sales.” But the following month, she joined Apple to work on AirPods firmware. She went on to introduce new AirPods features during two product launch keynotes, a living embodiment of the Doppler-Apple link. (Later, her husband and fellow Doppler alum, Caleb, also joined the AirPods team, and remains on it.)

When Doppler employees Mary-Ann Ionascu and Caleb Rau got engaged, they wore their Here One earbuds in the official photos. (Their former boss, Doppler cofounder Noah Kraft, officiated at their wedding.) [Photo: Courtesy of Mary-Ann Rau and Caleb Rau]

According to LinkedIn, seven former Doppler employees are at Apple today. Others scattered to Google, Amazon, Dolby, and Bose, ensuring that the company’s diaspora would continue to have an impact on the product category it had helped kick-start.

In an August 2018 Fast Company story, Peter Burrows wrote about Silicon Valley’s embrace of “hearable” computing and took note of the former Doppler staffers who’d landed elsewhere. Fourteen months later, Apple introduced AirPods Pro. Unlike their non-Pro predecessors, the pricier new buds blocked out background distractions with active noise cancellation and introduced a mode called Transparency that mixed streamed and local environmental sound so both could be heard—two features reminiscent of the “layered listening” that Here One had offered.

The AirPods Pro’s Doppler-esque evolution continued with a 2024 software update that let the AirPods Pro 2 be used as hearing aids. The following year, the AirPods 2 and 3 and the AirPods 4 got real-time translation, the same feature Doppler had shown me back when its test bed required a PC in a backpack. Now Apple is reportedly gearing up to introduce AirPods Pro with built-in cameras that let them see real-world objects and identify them with AI—a concept that would have sounded like fantasy in Doppler’s day, yet is very much in tune with its vision of a computer for your ears.

In tech, victory belongs to the companies that turn ideas into successful businesses, not those that have them first. In this case, that’s Apple. The Doppler alumni I spoke with enumerated some of the Apple advantages that the startup could never have matched, such as the fact it designs its own chips and controls both the AirPods and iPhone, lessening its dependence on Bluetooth for connectivity.

Then there’s Apple’s ability to invest in a product for years on its own terms, not those of venture capitalists seeking a quick return on investment. At one point in my conversations with Kraft, he theorizes that fulfilling Doppler’s promise would have required another $50 million; at another point, $200 million. Either way, “The sheer capital required to invent a new form of computing is beyond the capacity of the VC ecosystem, at least as it existed 10 years ago,” Lanman says. I can’t think of any examples, then or now, that disprove his point.

Still, the vibe I got from the Doppler veterans who talked with me about AirPods wasn’t resentful or jealous. Instead, it sounded like pride—albeit of a bittersweet variety.

“We laid the groundwork and set the direction,” says former product chief Nielsen. “Having a company with the technical prowess and the resources like Apple come in and pick it up and carry it to the finish line was super gratifying.”

We’ll never know what wireless earbuds would look like in 2026—and beyond—if Doppler had never existed. But Mary-Ann Rau sees parallels in the story of General Magic. That legendary 1990s startup tried to build a wireless personal communicator years before the technology was ready, and ceased operations in 2002. The conversation it kicked off about the future of mobile computing was still in progress when Apple came up with the iPhone. Did I mention that General Magic also created a voice-powered assistant that eerily foreshadowed current AI agents such as Meta’s Muse and OpenAI’s Dots?

“The same pattern happens over and over in Silicon Valley,” Rau says. “There’s the startup with the big dream and the small team, and they don’t quite make it, in the sense that they don’t make the product that ends up being ubiquitous.” (Undaunted, she left Apple in 2023 and later became cofounder and CEO of her own company with a big dream and a small team, heat pump startup Merino Energy.)

Ubiquitous is what AirPods swiftly became. Apple doesn’t disclose unit sales or revenue, but a 2024 Bloomberg Intelligence report estimated that the company had sold between 94 million and 119 million units a year since 2020, good for $18 billion to $22 billion in annual revenue. As others noted when those figures came out, they made AirPods a bigger business than Nintendo or Spotify.

Doppler would not have had to beat Nintendo and Spotify to have had a far happier ending. “When we started, we wanted a billion-plus,” Kraft says. “But towards the end of it, if someone had swooped in and said, ‘Here’s 300 mil,’ we would have taken it in a heartbeat. Because we’d only raised $50 million.”

Had an acquirer then gone on to capture just 5% of AirPods’ market share, Kraft adds, it would have created that rarest of phenomena in Silicon Valley: a major new hardware business. He stresses that he wouldn’t change a thing about the Doppler experience. But all these years later, he hasn’t stopped turning it over in his mind. The epiphany he had when he was assisting with his friend’s music tour still sounds like a great idea for a company to him.

Social media bans don’t protect young people’s mental health, early research shows

4 weeks 1 day ago

The year 2026 has been an expensive one for social media platforms.

Since March, YouTube and Meta, the parent company of Instagram and Facebook, have settled or paid out in multiple lawsuits to the tune of more than $17 billion for allegedly acting negligently, deliberately designing their platforms to hook kids and misleading the public about safety and privacy. If these suits are any indication of how states and the public are starting to view social media platforms, more may follow.

On top of these lawsuits, social media platforms are facing a loss of business as bans for minors are accelerating and spreading worldwide. Australia was one of the first with a ban on kids under 16 on social media starting in December 2025, followed soon after by the U.K. with a similar prohibition.

The current list of countries looking to ban minors from these platforms tops more than 20, from Greece to the United Arab Emirates to Canada. The proposed and enacted bans around the world vary in terms of age restriction, parental consent rules, and age verification methods. Different countries are looking at restricting different platforms, but most include a mix of Snapchat, YouTube, TikTok, Instagram, Facebook, and X.

These bans aren’t necessarily for mental health reasons only—some are motivated by political concerns about foreign-owned platforms, and others aim to protect young people’s privacy. But nearly every government banning social media platforms for young people mentions their mental health as a reason.

The question is, are these bans the solution to the mental health problems young people are facing? I research digital media and marketing and measure the impact of new technologies, and I’m skeptical about whether prohibiting access to these platforms will actually improve young people’s mental health.

Mental health concerns

Since Jonathan Haidt’s 2024 book, The Anxious Generation, critics of social media have increasingly focused on young people’s mental health. Haidt argues that social media disrupts kids’ social and neurological development, contributing to rising anxiety and depression through sleep loss, fragmented attention, addiction, loneliness, social comparison, and perfectionism.

A 2026 study found that young people between 14 and 20 are more than 100% more likely to experience mental illness than 14- to 20-year-olds 50 years ago.

Increased use of social media and screen time is compounding an already precarious time for minors. In 2021, it was estimated that 20.5% of young people experienced elevated symptoms of anxiety, and 25.2% of youth reported elevated symptoms of depression globally.

Many factors contribute to this, including better screening and awareness of kids’ mental health problems, as well as societal disruptions as a result of the COVID-19 pandemic. As the world sheltered in place and moved online for work, school, and social interactions, there was an increase in isolation and loneliness.

Not all bad

In light of these concerns, some researchers argue that bans on social media platforms are like putting an age restriction on buying alcohol or cigarettes.

But unlike these addictive and potentially harmful substances, there are benefits to social media, especially for teens who feel disconnected from their local communities and want to build relationships with others who feel similarly.

Social media platforms allow for global connectivity and community-building with like-minded groups. They also provide a place for self-expression and civic engagement.

In a Pew survey of U.S. teens, 80% said social media helps them stay connected to what’s happening in their friends’ lives.

Even as states and countries are moving to ban young people from social media, there isn’t clear evidence that spending time on these platforms is the main cause of their mental health problems.

Show me the money

One place where the comparison with tobacco is actually quite apt is in relation to the money from lawsuit settlements.

In 1998, states settled a lawsuit against the major tobacco companies to recover smoking-related healthcare costs. Part of the settlement was allocated to programs aimed at reducing youth smoking. In 2026, total state funding for these programs amounts to just 3.4% of the $21.7 billion in revenue states will collect from tobacco settlement payments and tobacco taxes.

Similarly, some money from an August 2026 Meta settlement with a group of states for $17 billion is slated for state-based programs focusing on mental health infrastructure, school and community programs, and public health campaigns.

However, there is no oversight or direction as to how states should use the funds, nor is there a mechanism to report how that money has improved kids’ mental health care, safety and well-being.

Is it even possible?

Early evidence from the Australian social media ban shows that kids are finding workarounds: Roughly 61% of kids ages 12 to 15 continue to access platforms using false IDs, altered appearances, or virtual private networks (VPNs), even though millions of accounts were disabled. In fact, this has become so common that U.K. lawmakers are debating restricting consumer VPN access.

Some countries, such as Australia, require the platforms themselves to develop ways of verifying a user’s age. Some minors have bypassed scans from biometric age estimation technology by using photo manipulation software, substituting celebrity photos, using clever makeup, and even utilizing older friends and parents to pass the “selfie tests” on their behalf. Others simply manipulate their birth years during registration or rely on existing adult-verified accounts managed by family members.

And all of these age-verification methods, whether developed by a government or a social media platform, rely on collecting sensitive data to estimate age accurately, which creates privacy concerns.

Australia is currently seeking to raise fines on social media platforms that fail to comply with its requirements. The Australian safety commissioner, charged with enforcing the ban, issued a compliance update that showed 7 in 10 children who held accounts on restricted platforms remained on those platforms, and more than 80% of Australian teens ages 10 to 15 still use social media frequently.

Beyond concerns about enacting and enforcing these restrictions, some policymakers also worry that total bans could push some kids to migrate to even riskier online spaces.

What could work

Based on the evidence, I’m not convinced that banning social media platforms will be an effective solution to the mental health crisis among young people today.

So what could work?

Supporting young people’s mental health is a complex endeavor that will require a combination of people, organizations, and resources coming together.

I believe kids’ mental health would be better served by governments enforcing safety by default standards on social media platforms rather than outright bans. This means that the highest security and privacy options are in place by default so that minors still on a platform are protected. For instance, accounts would be unsearchable by default, so users would have to intentionally make a profile change to be found by people they don’t know in real life.

Safety by default standards also place limits on the manipulation, algorithmic restraint and other addictive mechanisms that may lead to psychological harm in vulnerable populations such as young people.

Educating kids on digital literacy and online safety as well as ensuring some screen-free time are also going to help support kids’ mental health.

It’s also important for parents and other caregivers to recognize changes in behavior as a result of addiction to social platforms and screens. And when adults start seeing signs of a problem, they need clear, easy, and affordable ways to connect their kids to mental health support, which is not available in most places right now.

And finally, the United Nations recommends bringing kids into the conversation about what’s best for them online. Obviously, this does not mean putting 12-year-olds in charge of regulatory agencies. But feedback from young people about what they care about, what they get out of being online, and what makes it hard for them to look away from their screens could inform changes to social media platforms and education programs that could help young people engage online more safely and maintain a healthy balance.

Ari Lightman is a professor of digital media and marketing at Carnegie Mellon University.

This article is republished from The Conversation under a Creative Commons license. Read the original article.

AI expertise isn’t the same as AI oversight. Boards need both

4 weeks 1 day ago

From 2023 to 2025, the share of S&P 500 companies disclosing AI as a risk jumped from 12% to 83%. But according to company filings, fewer than 3% of directors have AI expertise. And in the same research, just 23% of executives said their board was highly fluent in AI, and a quarter said their board had low fluency or none at all. As artificial intelligence spreads across the business, that leaves a significant gap between what boards understand and what they are required to oversee.

That gap is untenable—and this is what boards need to do to fill it.

{"blockType":"mv-promo-block","data":{"imageDesktopUrl":"https:\/\/images.fastcompany.com\/image\/upload\/f_webp,q_auto,c_fit\/wp-cms-2\/2025\/10\/creator-faisalhoque.png","imageMobileUrl":"https:\/\/images.fastcompany.com\/image\/upload\/f_webp,q_auto,c_fit\/wp-cms-2\/2025\/10\/faisal-hoque.png","eyebrow":"","headline":"Ready to thrive at the intersection of business, technology, and humanity? ","dek":"Faisal Hoque’s books, podcast, and companies give leaders the frameworks and platforms to align purpose, people, process, and tech—turning disruption into meaningful, lasting progress.","subhed":"","description":"","ctaText":"Learn More","ctaUrl":"https:\/\/faisalhoque.com","theme":{"bg":"#02263c","text":"#ffffff","eyebrow":"#9aa2aa","subhed":"#ffffff","buttonBg":"#ffffff","buttonHoverBg":"#3b3f46","buttonText":"#000000"},"imageDesktopId":91420512,"imageMobileId":91420514,"shareable":false,"slug":"","wpCssClasses":""}}
Expertise is necessary but insufficient

If the problem is that boards lack AI expertise, the obvious answer is to recruit directors who have it. Well, yes, but not quite. Boards do need more AI experts, but recruiting experts is not enough to close the gap. There are two reasons for this.

First, AI isn’t siloed. It runs through hiring, pricing, customer service, credit, and operations. There is no single AI agenda item for one expert to own, and every director will vote on matters AI has shaped, whether they realize it or not. Oversight must therefore be distributed across the board, not concentrated in a limited number of expert seats.

Second, expertise is not the same thing as competent oversight. Expertise is what you know about AI; oversight is what you do about it as a director. Expertise is often needed for competent oversight, but it doesn’t guarantee it. For instance, someone can understand exactly how a large language model works without ever asking who in the company is accountable for one, or what an AI agent is allowed to spend.

So boards face two questions about AI. The first is whether the board knows enough. The second is whether each director is competent at oversight. Answering the first doesn’t answer the second—and while everyone focuses on the first, the second is largely neglected.

What the board needs to oversee: The key areas of AI oversight

Before we figure out what competent oversight involves, we need to be clear about what directors are meant to be overseeing. The following areas, drawn from our ongoing research into how directors oversee AI, are foundational. Each of these is a place where AI can do real damage to a company, and a board that covers some of them but not the others is dangerously exposed.

  • Governance Oversight: Knowing which individual is accountable for each AI matter, and making sure that they are actually held accountable.
  • Data Inquiry: Knowing what data the company’s AI uses, where it goes, and on what legal basis. This must be backed by evidence rather than a vendor’s assurance.
  • Explanation & Disclosure: Making sure the company can explain an AI decision to the person it was made about, and that its public claims about AI match what it does.
  • Fairness & Affected Parties: Knowing which groups a system could disadvantage, when it was last tested, and whether problems were fixed rather than just recorded.
  • Risk & Irreversibility: Knowing what the company’s AI can do that cannot be undone, and how the board would hear about a failure.
  • Human Judgment & Workforce: Making sure a person can overrule a system in practice, not just on paper, and that the people whose work AI changes are part of the plan.
  • Autonomy & Agent Authority: Knowing what AI agents may do, spend, and commit on their own, and making sure each one has a limit and a way to stop it.

That is what a board needs to oversee. But how do you actually build competent oversight of those areas? Three principles help you get there.

Measure what people do—not what they know

The strongest possible evidence for being able to perform a task is that you regularly perform it to a high standard.

Therefore, rather than only asking if your directors understand AI oversight at a theoretical level, you also need to ask: Have you actually performed this basic task of AI oversight recently? How often? Is it something that is a systematic part of your oversight process?

The point is that being aware of something is not the same as doing it, let alone being good at doing it. In PwC’s 2025 survey of public company directors, two-thirds said their board should spend more time on AI, but fewer than half said their board had an approach to actually overseeing it.

For example, almost every director would agree that the board should know which AI systems the company uses. Far fewer have actually asked to see the list, and fewer still ask for it every time the board reviews AI risk.

First Step: Ask each director to name the last AI oversight act they performed, and when and where it happened.

Assess the floor—not the average

A director’s level is set by their weakest competency, not their average. Boards already accept this logic for financial controls. Under U.S. auditing standards, if even one material weakness exists, a company’s internal control over financial reporting cannot be considered effective, no matter how strong everything else is. Nobody averages a material weakness away. AI oversight deserves the same.

Focusing on the floor matters because human beings gravitate to the questions they are comfortable asking. A former CFO interrogates the data; a technologist interrogates the model. To an extent, that’s what a board should want. Directors are recruited for their depth, and that depth should be used.

The problem comes when depth in one area becomes a reason to skip the others. A CFO who never asks what the company’s agents can do is relying on someone else to ask. If that person is absent, distracted, or deferring in turn, nobody asks at all.

So the floor is not a demand that every director become an expert in everything. It is a minimum standard that every director clears in every area. Boards already work this way with finance—under New York Stock Exchange listing rules, every member of the audit committee must be financially literate.  

Competent AI oversight needs a solid foundation: a floor that everyone clears.

First Step: Ask each director which AI oversight question they would be least able to answer. Make it their priority for the next quarter.

Build depth across the whole board—not in a single seat

The floor protects against gaps in individual directors. But a board can have every director clear the floor in every area and still have nobody who goes deep on fairness, on the workforce, or on what the company’s AI agents are allowed to do.

That is a common pattern, because boards recruit for backgrounds, and backgrounds cluster. In 2026, boards again favored directors with CEO and financial experience. So depth piles up around data and models and thins out everywhere else.

This is where the fact that AI isn’t siloed becomes especially relevant. It means that the board needs depth in every area of AI oversight, and that depth shouldn’t sit in a single seat, because one expert’s judgment tends to go unchallenged when no one else on the board is equipped to test it.

Financial controls deal with the same problem through the four-eyes principle: No material transaction rests on one person’s judgment. AI oversight should work the same way, with at least two directors working at depth in every area.

First Step: Map which directors work at depth in each area of AI oversight. Any area with fewer than two is uncovered.

Expertise is what a board has—oversight is what a board does.

Boards need both, and having the first doesn’t guarantee the second. And while expertise can be judged by looking at a résumé, building real competence at AI oversight means measuring what directors actually do. It means judging each director by their weakest area, not their average, and making sure the board’s depth reaches every area of AI oversight, held by more than one director in each.

In the meantime, directors who want to see where they stand can take part in our research by completing the scorecard and receive their own results.

{"blockType":"mv-promo-block","data":{"imageDesktopUrl":"https:\/\/images.fastcompany.com\/image\/upload\/f_webp,q_auto,c_fit\/wp-cms-2\/2025\/10\/creator-faisalhoque.png","imageMobileUrl":"https:\/\/images.fastcompany.com\/image\/upload\/f_webp,q_auto,c_fit\/wp-cms-2\/2025\/10\/faisal-hoque.png","eyebrow":"","headline":"Ready to thrive at the intersection of business, technology, and humanity? ","dek":"Faisal Hoque’s books, podcast, and companies give leaders the frameworks and platforms to align purpose, people, process, and tech—turning disruption into meaningful, lasting progress.","subhed":"","description":"","ctaText":"Learn More","ctaUrl":"https:\/\/faisalhoque.com","theme":{"bg":"#02263c","text":"#ffffff","eyebrow":"#9aa2aa","subhed":"#ffffff","buttonBg":"#ffffff","buttonHoverBg":"#3b3f46","buttonText":"#000000"},"imageDesktopId":91420512,"imageMobileId":91420514,"shareable":false,"slug":"","wpCssClasses":""}}

AI investors are getting nervous about this key metric

4 weeks 1 day ago

Will the AI bubble burst? 

That’s the question on investors’ minds when an AI company announces anything from a rogue agent to an earnings report. 

Sure, AI companies continue to increase their revenue, but one lesser discussed metric has investors scared: a slowing growth rate. 

“A handful of AI-linked companies have driven much of the stock market’s gains recently, so any wobble here can ripple through retirement accounts and index funds broadly, not just tech investors,” says Sameer Samana, an adviser and head of global equities and real assets at the Wells Fargo Investment Institute.

Take Anthropic, which shared its third-quarter results with current and prospective investors on Wednesday, October 14. The company announced that its revenue rose 45% to $16.8 billion, up from $11.6 billion in the previous quarter, CNBC reported.

In theory, it sounds like a successful result, but looking at the bigger picture, it signals a potential slowdown. Anthropic’s revenue had previously risen nearly 60%—from $4.7 billion to $11.6 billion—between quarters one and two.

Yet, the first quarter still beat the entirety of 2025. Anthropic reported $4.6 billion in revenue for the year and $386 million in revenue for 2024, according to Morningstar.

The declining growth rate marks a change for AI companies. “Investor expectations had been getting increasingly optimistic and hoping for ‘fast growth and getting faster.’ Now it’s just ‘fast,’” Samana says. “AI companies are still growing quickly, but not at the eye-popping pace investors had come to expect—and stocks were priced for that pace to continue.” 

OpenAI’s revenue forecast drops

OpenAI is among the companies causing concern. At the end of September, sources told Axios that OpenAI’s annualized recurring revenue was close to $70 billion. Then, on October 8, reports emerged that the company had revised that number, cutting it by $20 billion. In reporting the shift, the publication said the initial figure was due to the fact that AI companies have distinctive ways of counting sales. OpenAI wanted to directly compare itself with Anthropic. 

“It seems counterintuitive to worry about companies generating billions in revenue, but in financial markets, valuations are driven by the rate of change, not just absolute revenue,” says Adam Coons, chief investment officer and portfolio manager at Winthrop Capital Management. “Investors aren’t paying for today’s sales; they’re paying for years of hypergrowth to come. When that growth rate slows, even from extraordinary levels, it changes the math behind both the valuations and the debt-financed build-out that depends on them.”

In addition to the discrepancy, annualized recurring revenue is nothing more than a prediction, one that is commonly used by Silicon Valley but can be inaccurate. Investors can’t rely on it alone for growth indicators. 

Competition can slow revenue growth

Samana and Coons both point to competition as a potential reason for slower growth. “Competition is driving prices down,” Samana says. “More AI companies are competing for customers, so even as more people use AI, the money each company makes per customer is shrinking.”  

OpenAI might have started the current AI era with the launch of ChatGPT, but it now has to contend with a number of competitors. “With open-source models such as Meta’s Llama and a wave of Chinese models closing the capability gap, and OpenAI, Anthropic, and Google competing aggressively on price, the cost of digital intelligence is falling fast,” Coons adds. “When models converge in capability, pricing power evaporates, and revenue growth and margins can shrink at the same time.”

Anthropic’s IPO valuation sinks

These declining growth rates can scare away investors. Some investors in attendance at last week’s meeting are now leaning toward an initial public offering (IPO) valuation of $1.5 trillion for Anthropic, down from $2 trillion, Reuters reports. 

Anthropic submitted its initial filing to the Securities and Exchange Commission (SEC) in June. The company has since pushed its IPO plans from October to November, a move that left time to share its latest results. 

Reuters also viewed Anthropic’s IPO prospectus in which the company planned to warn investors that AI could have “catastrophic or existential risks to humanity.”

But slower revenue growth isn’t the only notable part of Anthropic’s finances. The prospectus revealed that Anthropic’s operating loss rose from $2.98 billion in 2024 to $8.06 billion in 2025. The company had $12.65 billion in total operating expenses for 2025, $7.33 billion of which was spent on computing and infrastructure—three times the amount it spent the year before. 

Overall, Anthropic had a net loss of $42 billion in 2025, up from just over $8 billion in 2024. According to the prospectus, $34 billion of its net loss was due to an accounting charge. 

A week after that news broke, New Constructs called Anthropic’s IPO “an unprecedented test of investor gullibility.” The independent research technology firm warned investors to avoid the IPO, citing “gargantuan risks” such as Anthropic’s rising net losses.

As Samana puts it: “These companies bet big—with borrowed money. Many committed to massive spending on chips and data centers, assuming growth would stay sky-high. Slower growth makes those bets riskier.”

Checked
11 minutes 2 seconds ago
Fast Company Technology
Fast Company inspires a new breed of innovative and creative thought leaders who are actively inventing the future of business.
Subscribe to Fast Company Technology feed