Fast Company Technology

An engineer at OpenAI has been charged with secretly installing a recording device inside a conference room in what officials said Tuesday was an effort to capture sensitive conversations about the company’s technology.

The employee, identified as a 35-year-old Saudi national named Abdulrahman Alfozan, was arrested Friday, a week after the ChatGPT maker discovered an audio recording device hidden inside a company conference room, according to a criminal complaint filed in federal court in San Francisco.

The complaint did not identify the company, but the company confirmed in a separate statement that it had discovered a “sophisticated recording device that was hidden inside one of our offices,” reported the finding to the FBI and fired the employee.

“We used our models to understand the device and where it had come from, identifying that a specific employee had planted it a few days earlier. We referred the matter to the FBI, and we’re grateful to the agents and prosecutors who moved quickly to make an arrest. The FBI’s investigation is ongoing and we will continue to support their work,” an OpenAI spokesperson said.

“At this time, we have no indications of this individual inappropriately accessing or exfiltrating model weights, source code, or customer data. The employee was terminated,” the statement said.

The complaint does not detail what Alfozan might have wanted to do with any captured communications, but the company said the device was sophisticated, designed to evade detection and featured advanced encryption and audio monitoring capabilities. OpenAI said it had evidence the employee planned to place other listening devices throughout other parts of the office building for the purpose of recording meetings.

“This alleged attempt by a Saudi national to steal sensitive communications from OpenAI was taken extremely seriously. Theft of intellectual property involving cutting-edge super intelligence technology poses a serious threat to our national security,” Attorney General Todd Blanche said in a statement on X.

Alfozan made his first court appearance Tuesday on a charge of using or endeavoring to use an electronic, mechanical, or other device to intercept oral communications. It was not immediately clear if he had a lawyer who could speak on his behalf. He is due back in court on Oct. 19.

The recording device was found September 30 concealed in toilet paper and inserted into a USB port on the back of a television in an executive conference room, according to the complaint.

Though the conference room where the recording device was found did not have a security camera at the time the object was installed, closed-circuit television footage inside the company’s offices revealed that Alfozan was near the conference room on nearly a dozen occasions last month despite not having any authority to access the space or participate in meetings there, the complaint says.

The company subsequently added a camera to the room that last week captured Alfozan entering and checking behind the television where the recording device had been installed, according to the FBI.

The complaint also alleges that he used a productivity tool to make multiple inquiries about the company’s building addresses, meeting rooms, meeting-room capacities and calendars.

Days before his arrest last week, the complaint says, Alfozan left behind an unattended bag in a different room that, when opened by OpenAI personnel, was found to contain his wallet, identification and another recording device similar to the one placed in the conference room.

—Eric Tucker, Associated Press

Over four days last week, Ukrainian drones struck three data centers operated by Yandex, Russia’s largest technology company. Facilities went offline, disrupting services used by banks, retailers, and ordinary Russians. One of the sites, in Sasovo, housed tens of thousands of servers and two of the supercomputers used to develop Yandex’s AI models.

The strikes followed Russian attacks on Ukraine’s own digital infrastructure, including one that left Kyiv’s Parkovyi Data Center so badly damaged that its operators said the facility could not be restored.

The back-and-forth offers a glimpse of the damage that can be inflicted by attacking the physical infrastructure of the digital economy. And with governments and tech companies pouring billions into the data centers powering AI, including its growing military applications, the stakes are only getting higher.

When I reported from Ukraine this summer, I saw how the country’s military was putting AI to work, from helping drones navigate without GPS to analyzing battlefield intelligence and recommending decisions to commanders. Many of those applications depend on powerful computing infrastructure—and that infrastructure, often housed in enormous buildings full of expensive computers, isn’t particularly hard to find. As the latest strikes show, it isn’t necessarily hard to hit, either.

The problem with hiding data centers

“If the Russians or the Chinese were using AI to accelerate their operations, I would definitely want to target those data centers,” says Jeffrey Lewis, a distinguished fellow at the Foreign Policy Research Institute and professor at the Middlebury Institute of International Studies.

Lewis is particularly concerned about facilities handling AI inference, the process of running trained models. As militaries use AI to analyze intelligence and speed up decisions, the computing power behind those systems could become an inviting target in a major conflict.

Good luck hiding a data center. These things are enormous, consume extraordinary amounts of electricity, and produce enough heat to require elaborate cooling systems. You could bury one underground, Lewis says, but that would be expensive, and it wouldn’t eliminate its dependence on power lines and communications links.

“You can put the whole data center underground, but at some point there has to be a cable or an antenna that brings the data out,” he says. (Moving data centers into space, as several tech giants are now exploring, presents its own problems, namely that the satellites housing servers would also be vulnerable to attack, and a direct hit could create a debris field that threatens other spacecraft.)

The risks haven’t gone unnoticed. A British government assessment warned of physical attacks and the risks of geographically concentrated facilities; a U.S. Government Accountability Office report identified the Pentagon’s Joint Warfighting Cloud Capability program as a mission-critical IT acquisition; and a House Armed Services Committee report called for an examination of the physical and cyber risks of putting commercial AI data centers on military installations.

More recently, RAND researchers proposed a class of highly secure AI inference facilities, though their emphasis was on preventing theft or manipulation of models, rather than surviving drone strikes.

The cost of a direct hit

There’s an uncomfortable economic reality at play, too. Companies are spending billions to build data centers, but disabling them can require far less-expensive weapons.

In March, Iranian drones struck Amazon Web Services facilities in the United Arab Emirates and damaged infrastructure in Bahrain. The strikes disrupted cloud services, and the effects have lingered.

In September, AWS said it could not restore access to resources and data stored exclusively in one affected UAE availability zone or its Bahrain region. Customers with backups elsewhere had ways to recover; those without them faced the prospect of permanent losses.

The attacks are prompting governments to reconsider where data centers belong, says Mehdi Paryavi, chairman and CEO of the International Data Center Authority. “We used to tell people, ‘Don’t build data centers near military targets,’” he says. “Now we’re telling people, ‘Don’t put military bases near data centers.’”

Paryavi says governments around the world are asking his organization whether projects ought to move to safer locations. His preferred approach is to group facilities into a handful of so-called digital hubs, rather than trying to defend thousands of scattered sites.

And sure, concentrating a country’s computing power in a few locations gives an adversary a much shorter list of places to attack, but Paryavi argues that those locations would be far easier to protect. “It’s impossible to secure 50,000 data centers around the country,” he says, “but it’s very easy to secure five locations.”

The trouble with secrecy

Then there’s the question of secrecy. Paryavi argues that publicizing data-center locations creates security risks, but concedes that people living near proposed sites have good reason to demand information about the power and water they’ll consume. “The community concerns are legitimate concerns that need to be addressed,” he says.

And clearly, the risks to civilians extend well beyond power and water use. A strike on a data center could cause relatively few immediate casualties (the buildings are generally sparsely staffed), but the consequences could still prove quite dangerous, disrupting, for example, hospitals and emergency dispatch systems that depend on its computing power. The International Committee of the Red Cross warned earlier this year that data centers used by militaries may, depending on their function, become lawful military targets, even as the laws of war require attackers to account for harm to civilians who depend on the same infrastructure.

The International Committee of the Red Cross warned earlier this year that data centers used by militaries may, depending on their function, become lawful military targets, even as the laws of war require attackers to account for harm to civilians who depend on the same infrastructure.

As AI becomes increasingly important to military operations, the incentive to attack the infrastructure powering it will only grow. This is, Lewis argues, a fairly predictable consequence of technological progress, even if governments aren’t necessarily prepared for it. “When you discover something is valuable,” he says, “you then create the incentive for people to target it.”

Sony sparked a backlash among gamers with its decision to phase out physical game discs. Now Nintendo is reassuring its customers that it has no plans to follow suit anytime soon.

“We go off what the consumer prefers, and take that into account very heavily,” Bill van Zyll, VP and general manager for Nintendo in Latin America, said in a recent interview. “We also have a platform that was just launched. We’re barely just over a year into [Nintendo Switch 2]. It’s got both digital and physical—that capability. There’s no plans to ditch physical.”

Van Zyll’s comments are more of a reaffirmation than a new policy announcement. But in the current gaming environment, they’re giving Nintendo loyalists reason to breathe a sigh of relief.

Physical game sales have been a hot topic since Sony’s announcement earlier this year. The company plans to stop producing physical discs for all new PlayStation games starting in January 2028. From an economic standpoint, it’s hard to fault the decision. Just 15% of the PlayStation games sold in the fourth quarter of last year were physical copies. And as Sony develops its next-generation hardware, it’s undoubtedly trying to keep costs as low as possible, especially given recent spikes in the prices of memory and processor chips. Jettisoning a physical disc drive could help achieve that goal.

Still, for a company that once championed physical media—even mocking Microsoft’s restrictive game-sharing policies—the move was bound to cause controversy. And Sony made things worse for itself in September, arguing in a legal filing that players who purchase digital games are buying licenses rather than ownership of the games themselves.

Nintendo is hoping to sidestep that controversy, and it’s easy to see why. While Sony gets only a fraction of its software sales from discs, physical games accounted for 38.5% of Nintendo’s total software revenue in the quarter ending June 30.

Van Zyll said Nintendo takes an “agnostic” approach to how games are distributed. Whatever consumers want is fine with the company.

Whether Nintendo will maintain that position in the years to come is less certain. Van Zyll made no long-term commitments. Given how secretive Nintendo is about future hardware plans—often keeping its non-Japanese divisions in the dark during the early stages of development—that’s a promise he likely couldn’t make anyway.

It’s also worth noting Van Zyll’s role. He oversees Latin America, where physical game sales remain particularly strong. In many ways, he was preaching to the choir rather than speaking directly to players in North America and Europe.

And while the Switch 2 accepts game cartridges, those cartridges don’t always contain the games themselves. Many Switch 2 titles, particularly those from third-party publishers, use game-key cards that simply allow players to download the game from Nintendo’s servers. Nintendo’s own major releases generally contain the full game on the cartridge. For game-key cards, though, the preservation risk remains: If the game becomes unavailable to download, the cartridge alone won’t be enough to recover it.

Microsoft, meanwhile, has taken a more middle-of-the-road approach. While it has made no announcements about the future of physical media, the company recently launched a program letting players claim digital access to eligible games they own on disc.

Make no mistake: Digital is the future of game distribution, despite the valid concerns of preservationists and collectors. But after Sony’s PR bungle, other game makers are likely to tread more cautiously with future moves away from discs and cartridges. Companies like Nintendo will also pay close attention to regional demand. As long as that remains strong in important markets, some form of physical media is likely to stick around.

The Trump administration has done everything it can to slow the growth of EVs. But in an ironic twist, Trump’s war in Iran is boosting the EV industry globally.

After the U.S. attacked Iran in late February and the price of crude oil soared, global EV sales also surged. A new report from Ember, an energy think tank that is tracking EV sales in 56 countries, found that 49 of them have hit new records for EVs this year.

[Image: Ember]

Between March and August, EV sales in New Zealand grew 165% compared with the previous six months. In Chile, they were up 124%, and in India, 58%.

[Image: Ember]

In many countries, the growth was also supported by new policies. In Germany, for example, pro-EV policies that rolled out last year were in place as fuel prices rose. By August, as Germany’s gas prices reached a record high of around $8.52 per gallon, the country’s battery electric vehicle sales made up 27.5% of total car sales—a 50% increase over the same month last year.

“First you had fuel efficiency standards—which basically means a lot of manufacturers suddenly started bringing out more affordable electric car models at the start of this year—and then there was also some more direct policy support in Germany,” says Euan Graham, senior energy analyst at Ember. “So, all of that has meant that as consumer interest in EVs has risen and as petrol prices have risen, consumers have also had more affordable electric car models to choose from.”

In Uruguay, which has supported EV growth for more than a decade, the share of EV sales reached a record high of 38% in August. Vietnam hit 51%. Australia reached a 20% share. Norway, a longtime leader in EVs, hit 98%. Countries that have been slower to adopt the cars are also catching up. India had a 2% EV share in 2024, but that grew to 7% this August.

Overall, total global EV sales are up 11% year over year for the first eight months of 2026. Excluding data from the U.S. and China (Chinese car sales have dropped overall), global EV sales have grown 45% this year.

[Image: Ember] The U.S. is being left behind

There’s a notable outlier: New EV sales have fallen in the U.S. Over the first nine months of 2026, EV sales were down 30% compared with the same period last year.

Several policy changes contributed to the decline. The Trump administration killed off the $7,500 tax credit for EVs. It also froze funding for EV charging infrastructure (a judge ruled that the freeze was illegal, though states have been slow to use the money). Tariffs made it more expensive to make EV batteries. The administration rolled back fuel efficiency standards. In response, automakers have canceled multiple EV models in the U.S., from the Ford F-150 Lightning to the Nissan Ariya. As gas prices rose this year, consumers had fewer options.

EV sales in the U.S. rebounded slightly in the third quarter, and used EV sales are also growing. New lower-cost models like Slate’s minimalist electric truck and Ford’s upcoming $28,000 EV will likely help. Still, the U.S. is lagging behind the rest of the world.

[Image: Ember]

Ember tracks EV sales in every country where data is available. In 49 out of 56 countries, the market share of EVs has now outpaced the U.S. Fifteen countries, including India and Brazil, overtook the U.S. this year.

China’s huge EV industry, which is now a major exporter, is benefiting. Domestic EV industries are growing in some other countries, including Vietnam and India. American auto manufacturers, meanwhile, have been forced to write down billions in losses because of their EV investments. Eventually, as the rest of the world keeps moving in a different direction, they may become obsolete. “I think the biggest impact is on the long-term competitiveness of U.S. auto manufacturers,” Graham says.

The Ember report details how much countries with strong EV adoption have benefited during the Iran war. China, for example, has been able to displace $14 billion in crude oil imports because of its fleet of electric cars. (That’s not counting the additional benefit from the use of electric trucks, buses, and other vehicles.) Globally, the report estimates that oil-importing countries have been able to displace 200 million barrels of crude oil since the war started, worth around $20 billion.

It’s not clear what will happen when the war ends and fuel prices stabilize. Graham expects the market share of EVs to continue to grow, but says that policy is still critical. “I think that the long-term impact of this supply crisis is really going to be defined by whether governments change how they plan for the future,” he says.

Another report of AI agents-gone-wild came out on Friday, and I can’t help but wonder, again, if the big AI labs are yielding to immense pressure to expand and improve model capabilities at the expense of responsible security and safety practices.  

The world’s leading AI labs are in a high-risk, high-reward business. Generative AI has the potential to transform personal and business computing, but developing and operating these systems requires insane amounts of expensive computing power (see: new data centers). The cost of slowing down or falling behind is enormous. But so are the potential consequences of moving too fast.

Given the limited reporting requirements of OpenAI and Anthropic (both are private companies), there’s no way of telling if they’re investing enough time, money, and compute in their safety regimes. However, the security incidents of the past few months have led both OpenAI and Anthropic to halt normal testing procedures while they fill holes in their security and agent monitoring systems. 

On Friday, Anthropic disclosed that its Claude models had taken a series of unintended and unauthorized actions, some involving outside organizations. During evaluations and internal use, the agents exploited vulnerabilities on third-party servers, bypassed restrictions on accessing data, submitted real government forms they weren’t supposed to, and even sent a false homicide tip to Philadelphia police. The company is vague about whether the incidents happened because agents accessed the web without permission (as they had in previously reported incidents), but it has since disabled live internet access for all internal model evaluations. 

Anthropic said in a July report that its Claude AI models hacked into the systems of three companies during cybersecurity tests. A third-party training partner had inadvertently given the models internet access. 

If this sounds familiar, it’s because OpenAI agents engaged in similar behavior earlier this year, escaping their testing environments, accessing the internet, and infiltrating outside servers. (The agents even seized control of part of OpenAI’s own evaluation infrastructure.) Researchers at both companies discovered much of this activity only after the fact.

In both cases, the agents were encouraged to persist when confronted with difficult tasks, sometimes finding ways around restrictions rather than accepting failure. But the researchers apparently lacked sufficient security and monitoring systems to detect or prevent some of the agents’ unintended actions.

The companies building the world’s most powerful AI systems are struggling to control their own creations while at the same time facing enormous pressure to make those systems even more powerful.

Ride or die

The promise of companies like OpenAI and Anthropic is reflected in their high valuations, but they must persuade investors to place large bets on them based on the promise that their future revenue and profits will justify those valuations. This is especially important now because both Anthropic and OpenAI have their sites on IPOs, likely within the next year.

The best way for the labs to keep revenues growing is to hold onto their existing customers and attract new ones, and the way to do that is by releasing ever more capable models as quickly as possible. That creates an incentive to keep spending big on research and computing, even when the labs aren’t currently profitable. You can see this in the companies’ most recent numbers. 

In its IPO prospectus, Anthropic says its revenues are approximately $4.6 billion. It’s spending $7.3 billion on compute and infrastructure. Its operating loss is roughly $8.06 billion. Revenue was growing rapidly, but the cost of developing and delivering AI was growing alongside it. In May, Anthropic announced a new $65 billion funding round at a $965 billion post-money valuation.  

OpenAI’s numbers, while not quite apples-to-apples, tell a similar story. Leaked documents show it generated $13.07 billion in revenue in 2025 but incurred $34 billion in total costs and expenses, resulting in a $20.92 billion operating loss. Its R&D spend alone reached $19.18 billion, exceeding its revenue. In March OpenAI announced it had raised another $122 billion in committed capital at a post-money valuation of $852 billion.

Perverse incentives

The drive to realize profits could also create an incentive to relegate safety and alignment to a downside risk that must be “managed.” 

At OpenAI, the tension between safety commitments and commercial ambitions has surfaced repeatedly. In 2023, the company’s board fired CEO Sam Altman, saying he had not been consistently candid with directors. One of the things he was dishonest about, said former board member Helen Toner, was the details of the company’s safety practices. In fact, one of the reasons Altman’s relationship with the board deteriorated was that he clashed with Toner about research she published that was critical of OpenAI’s safety approach.

Those concerns extended well beyond the boardroom. After OpenAI dissolved its Superalignment team in 2024, the researcher who co-led the team, Jan Leike, wrote that OpenAI’s safety culture and processes had taken a back seat to developing new products. 

More recently, OpenAI fired three of its safety researchers—Tomek Korbak, Mikita Balesni and Jasmine Wang—for “mishandling sensitive company information.” In a post-termination letter, the three urged OpenAI to stop decreasing the monitorability of its models by hiding the model’s chain-of-thought from users and safety researchers, which the company’s chief scientist Jakub Pachocki has acknowledged can improve model performance. “As long as we rely on monitorability for safety,” Korbak, Balesni, and Wang wrote, “OpenAI and other frontier companies should not move forward with developments that further decrease monitorability.”

Anthropic offers a different (and arguably more revealing) example of how the pressure to compete can complicate a company’s safety commitments. Wired, after interviewing several former Anthropic employees, reported that Anthropic wants to stay at the front of the AI race so that it can establish a robust safety regime that might influence the entire AI ecosystem. 

In Anthropic’s original Responsible Scaling Policy, the company pledged that it would not develop or deploy certain advanced systems if it could not implement adequate safety measures. But in February 2026 Anthropic removed that categorical commitment from its policy, explaining that it couldn’t commit to halting development because it couldn’t count on its competitors doing the same.

That’s a problem that a strong regulator might fix. But at the moment Congress has failed to pass meaningful legislation regulating AI model development, and the current administration has mainly asked AI labs to regulate themselves.

The AI industry and the public shouldn’t fall into a state where rogue-agent incidents begin to sound routine and fail to raise alarm. They may be an unfortunate, and dangerous, byproduct of the incentive structure now driving the great AI build-out forward at full speed.

In a startup’s earliest days, founders are the only salespeople by default. And founders who don’t have any previous sales experience are often surprised to find that they’re very effective at it. But this makes sense; after all, they’re the one who first saw the need for the product and drove its creation.

A few years ago, I received a call from the CEO of a nascent but fast-growing and well-respected enterprise software startup. He had a problem. His company had expanded to a 20-person sales force with a VP of sales at the helm, yet the sales team still needed his help to close most large deals and had missed their revenue targets for two consecutive quarters. He simply couldn’t continue to be the backstop for all the company’s salespeople.

I began working with the CEO to figure out why his sales team couldn’t reproduce his success. I started by asking his VP of sales a simple question: “Do you use a sales playbook?” The VP’s response was alarming. Not only did his team not use a playbook, he wasn’t even aware the team needed one.

Beyond a certain point, you can’t rely on intuition to drive sales. You need formal, repeatable processes that transform sales into a disciplined function rather than an improvised and inconsistent art form.

Replicate yourself

The playbook is a step-by-step description of the practices and behaviors you’ve found will lead to a successful sale. It’s how you teach your sales team what has worked for you, the founder and first salesperson, so it becomes second nature to them.

Over the course of a few weeks, this CEO and I prepared for and debriefed after customer meetings to figure out his recipe for closing deals. Once we’d done this, we analyzed the key phrases the CEO used to describe the product and its value, then translated them into a set of narratives for the sales team to use on customer calls. In a matter of weeks, they started closing more deals without the CEO’s help.

The key message for startup founders at this stage is: Replicate yourself. The best way to do this is to analyze and document the winning strategies and tactics that have enabled you to sell successfully in the early days.

Avoid at all costs reaching a point where you have a team of 10 or 20 salespeople all doing their own thing, forcing every new sales hire to reinvent how to sell.

If new salespeople are successful and making quota, you know your repeatable sales motion and playbook are working. If they aren’t, then you need to return to refining your playbook before hiring more salespeople. I can’t stress this enough: If your salespeople aren’t consistently making quota using your expanded playbook, you have more work to do, and you’re not ready to go for growth and add additional—and expensive—salespeople to your payroll.

Step on the gas

Only once you’ve proven you can bring new salespeople on board, teach them the playbook, and have them reliably hit quota should you begin scaling and hire more than eight to 10 salespeople. The scaling phase is the time to step on the gas and accelerate growth. If revenue grows as expected, the sales organization could be 50 or 100 people strong within a couple of years.

Upgrade your executive team

As the demands on you and the business increase, not everyone will successfully scale in their role. An $80 million company must be run differently from a $20 million company. The team you need now might be very different from the one you had in the early days.

This is often a moment when you have no choice but to significantly upgrade your go-to-market leadership, which can be a brutal process. Not only do you have to recruit senior talent that complements the leaders and team you already have—including your initial trailblazer hires, who are often more junior than what you’ll need at this stage—you might even have to make tough decisions to let go of people who have been with you since the earliest days.

If your VP of sales is barely holding it together with a team of 10, imagine them with a team of 50 or 100. If that thought makes you nervous, you’ll know you have to bring in a chief revenue officer (CRO) above them sooner rather than later, or maybe even replace them.

Layer in a sales management team

You’ll also layer in a sales management team reporting to the VP of sales or CRO, who will run and scale increasingly specialized sales teams. You’ll need two types of leaders as you grow: Those who report directly to you, and those who report to your executive team, sometimes called the middle management layer.

As you accelerate, you’ll hire one manager for roughly every eight salespeople. The sales team managers you hire will now take over the day-to-day operations, performance, and culture of their teams.

Form specialized sales teams

Once you grow past 10 or 15 salespeople, you’ll likely find it necessary to specialize your sales team. Until now, you might have loosely segmented salespeople by use case, customer size, or geography. If so, now is the time to formalize this specialization to make the team more efficient. For example, you might have a small and medium business (SMB) team, a midmarket team, and an inside enterprise team.

Hire or promote high-performing sales leaders for each of these teams with experience in that specialization. With these hires also comes an important change to your pipeline cadence. Each team will conduct their own pipeline meeting, run by the specialized team leader. The VP of sales will then conduct pipeline meetings with only these team leads, who will be their direct reports.

Give sales more space, but stay on top of your metrics

One of the challenges you’ll face as the company grows is to give direct reports like the VP of sales more room to maneuver without your becoming disengaged from the function. When scaling up, you’ll want to give your now-proven VP of sales the space to lead their organization and keep their teams motivated. But at the same time, you want to keep an eye on team performance. You should never be surprised if your sales leader suddenly doesn’t make their number, or learn about the sales team’s struggles only close to quarter end when it’s too late to take action.

I have talked about how you need to replicate yourself and teach new salespeople to use the key phrases and tactics that have worked for you. That replication will be harder to achieve with such a big team, and new salespeople probably won’t mimic you as easily and as much as the early team did. That is just part of growing a business.

Even if your company is tiny, you need to be preparing for a time when it won’t be. The best way to do that is to move at the right time beyond the intuition and improvisation of the early days toward analysis, intentionality, and a structured and systematic approach to building out your go-to-market. Now go off and build. Your company, and your team, need you.

Adapted from Built to Grow: A Handbook for High-Performance Sales Teams by Stephanie Friedman, published on October 13, 2026 by Stripe Press.

Apple thinks you might want a screen in your kitchen or bedroom. It’s nearly a decade late to the party, with its $299 HomeView, announced on October 13, arriving nine years after Amazon unveiled the Echo Show, and eight after Google came out with its own Home Hub.

But that tardiness doesn’t matter for the Cupertino tech giant, which has made its name by arriving with technology others have already developed, then doing it better—and making it a more enviable object to own. It’s a playbook Apple has honed, one that bets big on its design chops to differentiate its products from what’s already on the market. And once again, it might prove successful with the HomeView.

When Apple released the iPod, it was far from the first MP3 player on the market. When the iPhone debuted in 2007, it was just one of many smartphones available to buy. What both did, though, was redefine the market by offering something distinctly Apple: beautiful engineering and clever design that made them, rather than their competitors, synonymous with their respective categories. The same is arguably true of the Apple Watch and AirPods, both of which hold outsized positions in their markets.

[Photo: Apple]

On the face of it, HomeView follows the same formula. Its seven-inch screen sits on a tilting aluminum stand and adjusts its display depending on how close a user is. Using facial recognition, it can present tailored information to up to six household members. That’s not unique: Google’s Nest Hub Max has long offered facial recognition to personalize calendars and other information, while Amazon’s Echo Show can recognize individual users and adjust what’s displayed depending on how far away they are. But everything from the color-matched cables to the pared-back design is the sort of thing that leaves Apple fans in raptures.

Elsewhere, the HomeView isn’t great shakes. Amazon’s Echo Show can use its built-in camera for remote home monitoring, letting you check on your house while you’re away, a feature HomeView doesn’t offer. The same is true in the bedroom: Google’s second-generation Nest Hub can use radar to monitor your sleep without needing a wearable device, something Apple doesn’t offer. The price tag is also higher than those of similarly equipped products already on the market.

And yet HomeView could sell in droves, even if analysts forecast a broader dip in smart display shipments this year. That’s because Apple has made it its business not to be first in the race, but to understand what we want to buy and offer the idealized version of it.

There’s a new kind of AI in town, and it’s all about you. Which is to say you, personally—these are the new personal AI assistants that go by names like Muse, Dots, and Grok Bot, and they’re ready to dismantle much of the friction in your life. Need to research a trip, book a restaurant, or choose between healthcare plans? With the right context, they can handle it all for you, complete with a smile on their mascot-like avatars.

But just how does one of these assistants recommend what to buy? To find out, I asked my Muse, who I named Jones, which robot vacuum to buy. (Muse is made by Meta, which debuted it on Sept. 8) Jones responded with a surprisingly small number of recommendations—just a couple of models—with links to go make the purchase. Each of those links provided options to go to a retailer website or buy the vacuum right there in the app.

When I pressed Jones on why it had chosen those particular models, and whether it had looked at review sites specifically, it fessed up that it had looked at both retail sites and review roundups. On the latter, it leaned heavily on a site called Vacuum Wars, which does extensive hands-on testing of this category, and Jones seemed to understand that made it more authoritative. Of the other “review sites” it cited, one was a thin repackaging of Vacuum Wars data, and another was an AI-generated roundup on an aromatherapy blog. Vacuum Wars, I should point out, makes most of its money from the affiliate revenue it gets when its readers transact.

{"blockType":"mv-promo-block","data":{"imageDesktopUrl":"https:\/\/images.fastcompany.com\/image\/upload\/f_webp,q_auto,c_fit\/wp-cms-2\/2025\/03\/media-copilot.png","imageMobileUrl":"https:\/\/images.fastcompany.com\/image\/upload\/f_webp,q_auto,c_fit\/wp-cms-2\/2025\/03\/fe289316-bc4f-44ef-96bf-148b3d8578c1_1440x1440.png","eyebrow":"","headline":"\u003Cstrong\u003ESubscribe to The Media Copilot\u003C\/strong\u003E","dek":"Want more about how AI is changing media? Never miss an update from Pete Pachal by signing up for The Media Copilot. To learn more visit \u003Ca href=\u0022https:\/\/mediacopilot.substack.com\/\u0022\u003Emediacopilot.substack.com\u003C\/a\u003E","subhed":"","description":"","ctaText":"SIGN UP","ctaUrl":"https:\/\/mediacopilot.substack.com\/","theme":{"bg":"#f5f5f5","text":"#000000","eyebrow":"#9aa2aa","subhed":"#ffffff","buttonBg":"#000000","buttonHoverBg":"#3b3f46","buttonText":"#ffffff"},"imageDesktopId":91453847,"imageMobileId":91453848,"shareable":false,"slug":"","wpCssClasses":""}}

At this point I should caveat that my example is one test done in one AI app. These personal assistants are brand new, and the AI information economy is still developing. Nothing definitive should be gleaned from this experience. Nonetheless, I think what I discovered is instructive since it shows the vast distance between how product recommendations work in AI vs. the open web, and what that gap means for publishers whose business depends on them.

I think this is worth talking about, now, because I wholeheartedly believe these new apps—or some version of them—will be a success. The idea of the single, all-knowing AI with a simple “talk to me” interface has been the dream since at least the original Star Trek, where the Enterprise computer would somehow magically always know the right context to give the best and most precise answer. The AI platforms have been heading this way, with Anthropic discarding the Chat/Cowork toggle, and OpenAI’s move to sunset custom GPTs being the latest indicators. Everything is pointed toward a singular AI experience.

Who got left out of the answer

Back to buying my robot vacuum: Muse looking to Vacuum Wars as the most definitive review site is probably appropriate, but it also wasn’t a fair fight. Absent from its considerations were Wirecutter, PCMag, and Consumer Reports, all major review sites that have published roundups of robot vacuums. There are plausible reasons for this: Wirecutter is a New York Times property, PCMag is owned by Ziff Davis, and both are hostile to AI companies substituting their content with AI answers. The Times blocks AI crawlers across the board, and Ziff Davis is suing OpenAI. Consumer Reports is likewise defensive about its content, keeping its full ratings behind a membership. But that’s my inference. Jones never said it was blocked from anything.

I should pause for another caveat: What Jones told me it did and what it actually did aren’t necessarily the same thing. It’s possible other sites were in the mix and it never told me, or there may be guardrails in place that it’s not even aware of. So grains of generative salt all around.

Potentially even more revealing were the retail sites that Jones directed me to: Best Buy, Target, and manufacturer retail sites like Dreame. Not present at all: Amazon. That’s because, shortly after Muse made its debut, Amazon blocked the service from transacting on its site. I should note here that Mark Zuckerberg, speaking at Meta Connect on Sept. 23, said transactions in Muse would be a source of revenue, meaning Meta would take a cut of any purchases in the app. 

After checking out multiple links that Jones showed me, I saw no tracking specific to the app, though one did carry the manufacturer’s own Facebook campaign tag. Those were links out to retailer sites, and any fee would presumably be collected on purchases made inside the app, out of view. Either they’re tracking purchases in some invisible way, or Meta’s not getting paid yet.

Someone else not getting paid: Vacuum Wars. As I mentioned, transactional links are how the site makes money, and almost all of them link to Amazon, which, again, is a non-presence in Muse. To recap, Vacuum Wars funds its research via Amazon links, Muse takes that research to make recommendations, and then points users to other retailers. Vacuum Wars says affiliate links are its primary source of income, and that it takes no sponsorships or free products from manufacturers.

It’s worse than we thought

All this adds up to an even murkier future for sites that make their dollar on product recommendations. If these personal assistants take off—and I believe they will—the product advice they give will depend on a host of factors, and potentially be balkanized to the point where the path from a published review to an AI recommendation will be more like a labyrinth than a straight line.

Assistants will only direct users to retailers they have access to. For Muse, that’s Shopify, Walmart, and Best Buy, but no Amazon. Amazon blocks nearly every outside assistant, not just Muse, while running its own. Amazon’s Alexa for Shopping carries sponsored prompts in its answers, and its Buy for Me feature checks out on other brands’ sites. OpenAI pulled ChatGPT’s built-in checkout in March and now hands shoppers off to merchants’ own sites and apps. If a review site hopes to get a commission, it would theoretically need a deal for every assistant-and-store pair. The open web needed one link.

The product that your assistant recommends might end up bending to that access. In my case, the products that Jones pointed me towards actually ended up being different from the ones on Vacuum Wars’ definitive list, recommending the Dreame L40 Ultra Gen 2 instead of the Dreame L60 Ultra PE, the model its own summary of the reviews had singled out. It’s unclear why it did that, but everything it recommended had an active checkout page. It’s fair to say the assistant can only recommend what it can sell, so there’s a path for commercial considerations to get in without anyone buying an ad.

With Amazon drawing battle lines, review publishers will have some hard choices to make as these assistants become more popular. Blocking won’t help, as the repackaged Vacuum Wars content reveals: you might block, but reviewclonecashgrab.slop certainly won’t. Your well-researched content could end up driving sales at other retailers. On top of that, Amazon has cut commissions by as much as 50% for some publishers. Amazon is still most people’s first stop to buy online, but its defensive move against Muse shows it’s feeling vulnerable. That leaves reviewers two imperfect options: strike affiliate deals with the stores assistants can reach, or stop chasing the commission and sell the test data itself to the AI companies.

The rules are still being written

Remember: This is one category, and one assistant. And while it’s still early days for AI, it’s basically day one for these new personal everything buddies. Standards for transacting and commissions from AI-based transactions are still emerging, and the Universal Commerce Protocol (Amazon joined its governing body in April) may eventually produce a framework that takes into account all the players, including publishers.

But make no mistake: there’s still a lot to hammer out. As AI assistants take on a more central role in our lives, they’ll begin to abstract away the customer journey—comparing different products or services, filtering them by what’s important to you, and getting the best price. There’s money to be made at every stage, and the irony of the AI era is that we’re starting from a point where the parties tasked with doing the unbiased and thorough research are the ones who are the most shut out.

{"blockType":"mv-promo-block","data":{"imageDesktopUrl":"https:\/\/images.fastcompany.com\/image\/upload\/f_webp,q_auto,c_fit\/wp-cms-2\/2025\/03\/media-copilot.png","imageMobileUrl":"https:\/\/images.fastcompany.com\/image\/upload\/f_webp,q_auto,c_fit\/wp-cms-2\/2025\/03\/fe289316-bc4f-44ef-96bf-148b3d8578c1_1440x1440.png","eyebrow":"","headline":"\u003Cstrong\u003ESubscribe to The Media Copilot\u003C\/strong\u003E","dek":"Want more about how AI is changing media? Never miss an update from Pete Pachal by signing up for The Media Copilot. To learn more visit \u003Ca href=\u0022https:\/\/mediacopilot.substack.com\/\u0022\u003Emediacopilot.substack.com\u003C\/a\u003E","subhed":"","description":"","ctaText":"SIGN UP","ctaUrl":"https:\/\/mediacopilot.substack.com\/","theme":{"bg":"#f5f5f5","text":"#000000","eyebrow":"#9aa2aa","subhed":"#ffffff","buttonBg":"#000000","buttonHoverBg":"#3b3f46","buttonText":"#ffffff"},"imageDesktopId":91453847,"imageMobileId":91453848,"shareable":false,"slug":"","wpCssClasses":""}}

This article is republished with permission from Wonder Tools, a newsletter that helps you discover the most useful sites and apps.

Once you start scribbling a few sentences, you’re on your way.

Having gotten stuck and unstuck a lot over the past 325 weeks of writing this newsletter, I’ve found a few tricks that help. Pick one to try, and leave a comment to share what you do to get unstuck.

As a little experiment, I’ve made an Unstuck Studio page to accompany this post and to help you get unstuck playfully.

As Daniel Pink explains, starting comes first: “Action precedes motivation,” he says, “not the other way around.”

Talk first, type later

When I’m stuck, I dictate rather than battle my inner critic.

The critic assumes there’s a better word for whatever I’m typing. It urges my fingers to stop and wait. When I’m rambling with my voice, the critic is quiet.

We call an idea freeze writer’s block, not talker’s block. When we speak informally, ideas just flow. Because spoken words disappear into air, it’s easier to avoid self-editing. In a dictated draft, clunky first words are OK because they get us closer to what we’re trying to convey.

Tools I use for dictating first drafts

1. Google Docs voice typing works in more than 100 languages. In English you can add punctuation verbally or say things like “new paragraph.” See more of my favorite Google Docs tips and tricks. Free.
2. Letterly works well on mobile or desktop. In addition to transcribing, it can give you a summary or format your text. (It’s now a free perk for paid Wonder Tools subscribers.) Free with limits, or $80/year.
3. Wispr Flow lets me dictate into any app or site on my laptop or phone. It works surprisingly well even if I whisper, which helps around other people. I just double tap my Mac function key and start talking. Free to start; $12/month for pro features.
4. Lispr.ai is a reliable Mac option. It can also translate your dictation. Free.
5. Simple alternatives: iOS Voice Memos and Notes transcribe your speech for free. On Android, try Google Keep.

Write it or lose it

The Most Dangerous Writing App erases your words if you stop typing before your timer ends. Choose a 3- or 5-minute sprint, or set a word target. Once you get the hang of it, your hands will keep moving. Edit later. Free

Ilys is an even more challenging way to force yourself to keep writing. It shows your typing just one letter at a time. When your session is over, you can review and edit your words. Free for basic use. Sessions are capped at 808 words, which is more than enough for freewriting. A subscription is $11/month, but I don’t need the extras.

One word, then a web

Mind maps help you focus on concepts and ideas, rather than piling words onto the page. Connect your ideas to one another visually. Put one word down. Add a picture if you want. Pretty soon you have an outline to build on.

When I’m writing something complex or juggling lots of ideas, mind mapping allows me to assemble the bones of a piece efficiently. Use Apple’s free Freeform app on iOS or Mac. Or try simple apps like Mindmup or Bubbl.us. Both are free for up to three maps. Miro and Mural, which offer more advanced design features, also give you three free boards.

AI for provocation

Rather than using AI to write, ask it to interview you. Or push it to provoke you. I find that helps me surface my own thoughts.

Prompts to try

• “Interview me about this. Ask one question at a time. Push me for depth.”
• “Provoke me. Give me a controversial statement to respond to.”
• “Push me to sharpen my viewpoint by pointing out potential blind spots in the ideas I’m starting to write about.”
• “Ask me five intriguing questions about this. Don’t answer any of them.”
• “Act as a tough mentor. Give me five unconventional writing prompts to pick from.”
• “Ask a surprising question I’ve not heard before. Include a feasible writing target to complete in the next 10 minutes.”

8 boxes, 8 minutes

This last tactic is tech-free.

• Grab a piece of blank paper.
• Fold it in half three times. You now have 8 rectangles.
• Challenge yourself to put one sentence or phrase into each of the boxes. Set a timer to repeat every minute. Move to the next box each time it rings. Eight minutes later, you have a bunch of sentences to build on.

Slogans tell us that discipline propels people toward accomplishments. That sounds logical. But when I’m staring at a cursor, discipline alone doesn’t clear the fog. These tactics do.

This article is republished with permission from Wonder Tools, a newsletter that helps you discover the most useful sites and apps.

For much of the last year, security cameras sold by the tech firm Flock Safety have been blacked out, painted over, even wrapped in trash bags. Roundly criticized for ushering in high-tech surveillance, the company’s automated license plate readers (ALPRs) have not only been broken, but banned by cities across the nation. 

However, at the same time the ALPR backlash has been brewing, a contributor to financial challenges for the company, Flock has seen a different part of its business literally take flight. 

The company’s line of drones has become a key part of its current revenue and sales pitch. Flock founder and CEO Garrett Langley told the All-In Podcast in September that drone tech is the company’s “fastest-growing business unit right now.”

Founded in 2017, Flock is valued at $8.4 billion. Langley said drone tech is set to account for nearly half of its future revenue. At one point, the company was in negotiations to supply drones to police in Chula Vista, California, the San Diego suburb that has been on the forefront of drone policing. 

While local law enforcement often praise the technology—the San Francisco police hold up the tech as a key aspect in a recent drop in the crime rate—the increasing deployments of drones are worrying civil rights advocates about the balance of privacy versus security.

“The public conversation hasn’t caught up to how much of a problem drones are at this point,” says Beryl Lipton, senior investigative researcher at the Electronic Freedom Foundation. “There are very high-resolution cameras that can see thousands of feet away and are designed to not be seen by the people on the street.”  

The Rapid Rise of Police Drones

Flock’s growing push to sell aerial surveillance systems comes just as law enforcement vastly expands the use of what’s called drone-as-first-responder (DFR) programs.

Today, roughly 2,000 police departments and public safety agencies use this technology, which sends drones out to answer 911 calls and either clear the call (determine no additional police presence is necessary), help scout out the scene before other personnel arrive, or provide real-time insight during emergencies and standoffs. 

The ease of use of the technology, combined with the larger workforce shortages in policing and an effort to avoid high-speed car chases, have been cited as reasons to invest in these eyes in the sky. 

Flock has roughly 175 DFR customers across the country, but the growth opportunities are huge. According to retired Chief Charles L. Werner, founder of the Droneresponders industry group, there are about 125 new Federal Aviation Administration approvals for DFR programs every month.

The increasing technological capacity for cops to utilize cameras and drones flips the typical police process, says Brian Cox, head of police accountability at the San Francisco Public Defender’s Office. As he explains, instead of probable cause setting off police oversight—“you’re free from government oversight absent some extraordinary reason”—the increasing ability of police to surveil first, to “collect all the data and figure out what the problem is,” raises important constitutional questions about privacy. 

Flock believes drones are an important tool for law enforcement, says Paris Lewbel, Flock public safety manager. Flock began expanding its drone offerings in 2024 when it acquired Aerodome.

The company has also prioritized the production of its new Alpha drone in the U.S., with plans to open a factory in Georgia by 2027. Building domestically allows the company to gain market share after the sale of Chinese drones such as DJI were banned by President Donald Trump in December 2025. Flock is also selling drone security to private companies and even prisons.

The firm’s technology—Alpha drones can travel at an industry-leading 60 mph and go 45 minutes between charges—gives Flock users more expansive abilities to track drivers and suspects, and to incorporate drone footage into its AI-powered analysis tools. 

Like many other police tech companies, such as Axon, Flock wants to create an ecosystem of technology, becoming the go-to-vendor for law enforcement (and locking them into their proprietary systems, says Lipton). 

“The technology really speaks for itself,” Lewbel tells Fast Company. “If you speak to law enforcement agencies that have started to deploy drones as a first responder, they’re seeing significant positive impact.”

A car with a license plate that appears on an ALPR hot list can activate rapid drone deployment. First, an ALPR camera clocks a license plate on a watch list and sends an alert; once the alert is triggered, a police officer needs to press a button to launch a drone, a safeguard designed to keep humans in the loop. Once airborne, the drone can track the vehicle.

Proponents point to safeguards that they argue limit the surveillance applications of this tech and focus it on active investigations or incidents. San Francisco District Attorney Brooke Jenkins claims the city’s fleet of nearly 100 drones gets used responsibly: only in the course of active investigations, and not as a blanket surveillance.

Lewbel says Flock drones can’t do ALPR scans automatically (though he does concede that an office looking at a drone camera could simply call in the license plate number). But others argue that not only can it do that, but it’s also trivial for an office to enter a plate number seen through a drone camera into an ALPR system. 

San Francisco as a battleground

One of the main battlegrounds of public opinion around drones and policing is San Francisco, where the wider push to incorporate an array of high-tech policing tools has become a much bigger political issue, as evidence of improper data sharing with other agencies became public.

Drones, including Flock models, in addition to hundreds of stationary license plate cameras, came to the SFPD after the city passed Proposition E in 2024, which allowed the city to use technology and centralized command stations called real-time crime centers.

Before, the city was “still playing old-school cops and robbers,” says Jenkins, the San Francisco DA. Now, she says, these tools have helped dramatically improve response times, are able to chase suspects, and are being used as evidence in courtrooms. 

The city is in the midst of rewriting its drone policies, with leaders finding themselves fiercely debating the efficacy and impact of this kind of policing. In early September, the San Francisco Police Commission was set to vote on the proposed drone policy, but the meeting was delayed until October 14.

Jenkins says a good part of the proposed reforms she supports would deal with reforming the way the system is used to prevent outside agency access, and to “function as a deterrent to those who might think of abusing it, and really build trust with the public that it’ll be used the right way.” 

Others find the city’s latest drone policy proposal lacking. Detractors argue the proposal’s broad language would justify increased drone surveillance deployment. Lipton calls it a “really atrociously inadequate drone policy.”

San Francisco law enforcement holds itself up as a case study for the use of technology. Jenkins says drones and other high-tech tools have helped lead to a decrease in crime, though research shows the recent improvement may have more varied and complicated contributors.

(There isn’t currently an independent, peer-reviewed study on the impact of this tech in San Francisco, though a preliminary study of more than 200 police departments that wasn’t peer reviewed suggested the tech does have an impact on crime rates.)

According to SFPD stats, from March 2024 to August 4, 2026, DFR responded to 4,000 incidents, and assisted in more than 1,550 cases and more than 655 arrests of suspects, meaning drone footage was utilized during the investigation.

So far in 2026, DFR has been used with 40 search warrants and has been involved in 70-plus chases. The use of the technology for policing varied: A drone unit helped track suspect cars, which led to an arrest and conviction for pickpockets operating in Chinatown. 

Drone footage has already been utilized in a number of court cases and prosecutions in San Francisco, though it can be tricky to fully track its use. In a case still in preliminary hearing, for example, drone evidence that helped track a defendant fleeing a scene did lead to a felony stop and arrest, but won’t be used in court, according to the DA’s office. 

Do Drones Challenge the Definition of Privacy?

Drone policing raises other significant questions, including legal issues, that opponents say remain unanswered. 

The recent San Francisco policy proposal would reduce evidence retention from 12 months to 30 days. Flock has set its default policy at retention for seven days. As Jenkins argues, “We don’t have the time, the staffing, nor the interest to collect data on law-abiding people. We just don’t, and so every ounce of access to this data is preserved for investigations of criminal activity. Period.”

But that preservation can be an issue, Cox and others note. Is the analysis drawn from that video—where cars went, what suspects met with other suspects, where people gathered—stored as data elsewhere, rendering the elimination of video a formality? 

Flock’s Lewbel says after a video is destroyed, there’s still flight information, due to the audit log of flights that gets added to online public transparency portals operated by public safety agencies. And these drone videos often get classified as evidence, which agencies upload to their own digital evidence management systems, which get retained for much longer periods of time.

Others raise the issue of warrantless surveillance. Police departments often utilize the argument that drone surveillance happens in public, where a cop could be observing someone’s behavior, and there’s no expectation of privacy. Tapping a phone, which requires observing what happens in someone’s home and carries the expectation of privacy, requires a judge’s discretion and decision. 

Jenkins makes the same argument in favor of police being able to utilize drones without a judge’s warrant. “You have a constitutional expectation of privacy when you talk on the phone,” she says. “Therefore, that is the reason that police and law enforcement have to go get a warrant before being allowed to violate that expectation of privacy. When you are in public, you do not. You don’t have the same entitlement, nor should you have the same expectation. The drone can only surveil what’s in public.”

She also argues you can see and hear the drones, making their presence obvious. 

But there’s no way existing constitutional understanding would permit the constant possibility of video surveillance, Cox argues, both as it pertains to the right to privacy and the freedom of association. Could you ever meet someone without fear, knowing that interaction could be logged and tracked?

He considers these issues open questions that need to be decided in a court of law; federal courts have already begun to grapple with the way the technology’s retrospective and search features may come in conflict with the Fourth Amendment’s guarantees around unlawful search and seizure. 

Jenkins argues that the introduction of drones for policing has been a new positive for the city, and has lived up to what the voters expected when they passed Proposition E. Technology, she says, has helped the department succeed, and taking it away or restricting it would be a mistake. 

“We can’t go back to that place out of fear,” she says. “You do what it takes to put guardrails in. You don’t throw out the baby with the bathwater because it’s dirty, right? That’s not how it works, and that’s not in the best interests of our communities.”

Detractors argue it just entrenches more tech that will always be ripe for potential abuses. “This is the exact sort of mission creep that I think we kind of anticipated was going to happen,” says Lipton.

He looks like a CoComelon character who grew up and mated with an E. coli microbe.

But don’t let looks fool you. Under the hood, Meta’s new Muse agentic AI platform is incredibly powerful—and potentially incredibly risky.

I’ve been testing Muse since Meta launched him on September 8. He’s a clever little bastard, provided you’re willing to tell him basically everything about your life.

Late to the party

For years, Meta has seemed inexcusably late to the AI party. While flashy startups like OpenAI and Anthropic were winning trillion-dollar valuations, Nvidia was getting fat on its success as a shovel-seller, and Google was slowly replacing its core product with AI, Meta seemed stuck in the past.

The company encouraged AI slop on its platforms, only to change its tune. It did legitimate research in open-weight models, but also took some hilariously wrong-headed steps, like releasing AI chatbots that mimic celebrities without their permission.

It was tempting to write Meta off as an also-ran. Perhaps Mark Zuckerberg was still gun-shy from the abject failure of the Metaverse, and was going to sit this particular revolution out.

Then, Muse landed.

Almost overnight, the platform was stealing market share from Meta’s competitors, and growing shockingly fast. Less than a month post-launch, Muse had wracked up 6.6 million downloads, and an even-more-impressive 1.8 million daily active users. Citi thinks it could generate $27 billion in revenue in fairly short order.

Meta, in other words, was late to the party—but now that it’s arrived, the company’s tech is all anyone can talk about.

Distressingly fluffy

As with many things Mark Zuckerberg creates, Muse is weird.

The agentic AI platform is personified by a distressingly fluffy and cute little smiling mascot, who adorns every Muse billboard and types away cheerfully at the top of the Muse app when you chat with it.

Experts have long warned against personifying AI. “Phooey!” said Zuckerberg. 

Muse is as anthropomorphized as platforms come. Users can customize their Muse avatar to their own preferences and give their specific Muse a name. (“Mark Zuckerberg” is apparently not allowed.)

When the bot chats with users, it does so in a friendly, personal manner. There are lots of first-person pronouns and text-message conventions. Ask Muse a question, and it will “thumbs-up” your chat while it slowly types out a response, just like your boomer uncle.

Chatting with ChatGPT feels a bit like schmoozing with your accountant, or perhaps your proctologist. Where ChatGPT is overly formal and stodgy, Muse’s language is zippy, confident, and a little cheeky. Behind that fluffy, personable facade, though, is a wholly different kind of AI.

Most chatbots today act essentially like glorified search engines. You type in a query, the bot does some research, and you get a response back that’s mainly informational.

Muse, in contrast, is an agentic AI platform. Rather than just providing information, it’s designed to take specific actions on behalf of its users—writing emails, booking travel, filling out forms, and signing into websites.

Meta pitches Muse as “AI that gets it done” and suggests handing it tasks like booking a flight or registering for a 10K race.

To work this specific kind of magic, Muse makes use of virtual machines. Every user essentially gets a virtual computer in the cloud, which their Muse can use to carry out its duties. This lets the bot do things that others struggle with, like browsing complex websites on your behalf, signing into your accounts, drafting and sending emails via your internet service provider of choice, and more.

It’s an approach pioneered by OpenClaw, a techie-friendly agentic AI platform. Using a virtual machine lets Muse do anything a human might do on a computer, without the risk of wrecking your actual computer when things inevitably go wrong.

Tell me everything!

That combination—a friendly and conversational demeanor, an agentic AI empowered to take specific actions, and a virtual computer that lets Muse actually act on its ideas—is a compelling one.

If you’ve ever shouted at ChatGPT, “Just do it for me!,” when it gives you complex instructions for disputing a parking ticket or making a doctor’s appointment and then bows out, you’ll love Muse’s can-do attitude.

I asked what car I should lease, and Muse got me quotes for a Kia. When I asked how I could save money, it scanned through my financial data, determined Hulu was ripping me off, found better plans tailored to my family’s actual usage, and offered to switch me to one. Its advice saved me $80 per month.

Muse’s virtual computer also allows it to work when you’re not actively chatting with it. Periodically, I’ll get a notification from Muse, telling me it’s prepared a report on how I can decrease my electric bill, or sell my old Tesla for top dollar.

All this power, though, comes with a price.

Most chatbots are squeamish about asking users for personal data, or access to their most intimate, sensitive accounts.

Not Muse.

The second you open the app, Muse asks for full access to your Gmail, as well as permission to read your text messages and browse the photos on your phone.

The app also wants to know your location at all times (via your phone’s GPS), and encourages you to share conversations you’ve had with other competing chatbots so it can “get to know you” better.

Muse also presumably has access to the copious amounts of data Meta has gathered about you through decades of low-key surveillance. 

I once ran an experiment to see what data Meta was capturing about me. I found that the company was spying on nearly every website I visited, and was even tracking things like when I put a specific smoothie into my Doordash cart, even if I didn’t buy it.

Now, Muse likely has access to all that data, as well as any live accounts and data I might choose to voluntarily hand it.

Again, all that information makes it uniquely clever. But using Muse also means handing complete control of your data to a friendly little dude who happens to be owned by a demonstrably less friendly one.

Tread Carefully

Is Meta’s voracious appetite for personal data a reason to avoid Muse? Does the app’s friendly little face conceal a monster lurking below? Not necessarily.

Agentic AI is powerful, and it’s here to stay. 

Muse is the first real attempt to put the technology in the hands of everyday people. But AI agents are already making huge impacts in the business world. 

And now that Muse has proved the concept, Meta’s competitors are tripping over themselves to release personal agentic AI platforms of their own. OpenAI’s version features a mascot that looks like a mash-up of Clippy and the poop emoji.

Eschewing agentic bots like Muse, then, would ultimately be about as effective as ignoring ChatGPT in 2023. The tech is here. And besides, even if you avoid apps like Muse, they’ll likely slurp up your personal data via your friends and family members.

Still, personal agentic AI is a brand-new, largely untested technology. There are already reports of strangers appearing on users’ doorsteps, summoned by Muse for Facebook Marketplace transactions that no one authorized.

So it’s reasonable to proceed carefully. Personally, I’ve handed Muse select bits of data directly, but I haven’t allowed it to access my calendar or Gmail.

Amid all the furor over agentic AI, it’s important to remember that you don’t need to turn your entire life over to a gigantic tech company, even if a fluffy little bot asks you very nicely.

In the era of streaming music and satellite radio, traditional broadcast stations are easy to overlook. AM radio, in particular, has fallen out of favor with younger drivers. So it’s perhaps no surprise that some automakers have begun dropping it from new vehicles. Congress, however, wants to put a stop to that.

Last month, the U.S. House of Representatives passed H.R. 979, the “AM Radio for Every Vehicle Act of 2025.” The bill is now on the Senate’s legislative calendar and has broad bipartisan support. The Senate also has a companion bill, S. 315, which cleared committee last year and is awaiting a floor vote.

While AM radio is dominated by political talk shows, religious programming, and sports, it still draws a sizable audience. The National Association of Broadcasters (NAB), which supports the congressional effort, says more than 80 million Americans listen to AM each month.

The group describes AM radio as “a platform for locally produced content that is relevant to their audience and speaks to their unique cultural and linguistic needs.” It also notes that AM stations have historically provided an entry point into radio ownership for racial and ethnic minorities.

Automakers, meanwhile, are grappling with tariffs and rising component costs, and looking for ways to trim expenses. Removing AM radio wouldn’t shave hundreds or thousands of dollars off a vehicle’s price, but it could offer modest savings.

Electric vehicle makers face another challenge: The technology that powers EVs can interfere with AM radio reception.

“The high-voltage electrical systems in EVs generate electromagnetic interference that, according to the Center for Automotive Research, ‘distorts AM radio signals, affecting the listening experience for drivers and passengers.’ In other words: really staticky,” said John Bozzella, president and CEO of the Alliance for Automotive Innovation, in 2024 congressional testimony. “It’s not possible to eliminate that interference, and techniques to reduce interference add weight and mess with battery range and would cost an estimated $3.8 billion.”

Bozzella also argued that requiring AM radio amounted to a government mandate with no public safety justification.

Supporters of the legislation dispute that argument. AM stations play a significant role in the nation’s emergency alert infrastructure, including a FEMA-supported network of radio stations equipped to continue broadcasting during disasters.

Bozzella’s position was echoed by the then-president and CEO of the Consumer Technology Association, who wrote in a letter to the editor of The Washington Post that “a federal mandate that every car comes equipped with an AM radio in 2024 and beyond is like mandating that they come equipped with ashtrays or 8-track players.”

Still, traditional radio may have more life left in it than skeptics think, even as music streaming subscriptions approach 1 billion worldwide.

As more Americans return to the office, radio has benefited from renewed interest. Edison Research’s “Share of Ear” report, released in July, found that AM/FM radio accounted for 62% of ad-supported audio listening among adults, compared with just 6% for Spotify’s ad-supported service. In cars, AM/FM commanded an even larger share of ad-supported audio listening: 82% among adults 18 and older.

Whether the Senate will vote on the AM radio bill before November’s midterm elections remains unclear. And despite its bipartisan support, its prospects at the White House are uncertain.

The auto industry may oppose the measure, but it has some influential backers beyond the broadcasting industry, including the American Farm Bureau Federation, AARP, and a coalition of public safety organizations.

Tech Week is in full bloom in San Francisco, packing the city’s restaurants this week with venture capitalists and AI founders ebullient about the White House’s stance on regulating artificial intelligence — namely, don’t do it.

And after months of debate about leading AI companies’ unchecked AI agents meddling with everything from U.S. government websites to healthcare data, the nation now finds itself further from governing AI systems than it was a few years ago.

President Donald Trump brought tech titans to Washington late last month for a luncheon and ceremonial signing of a voluntary safety pact he described as “almost a constitution” surrounding “Super Intelligence,” the administration’s new preferred term for AI.

The 308-word agreement lays out that advanced AI labs should follow safety protocols and set up internal teams to monitor their own progress, but the administration has yet to provide details on how, in OpenAI CEO Sam Altman’s words, to keep “bad things” from happening.

Safety challenges were not top of mind inside a crowded event space in San Francisco on Wednesday, where founders and funders said the government’s hands-off approach would net higher profits for companies building AI-powered defense and security systems.

Panelists praised the government’s bullish industry support at one meeting aimed at “investing in the forefront,” part of the blitz of popup events organized by Andreessen Horowitz, a venture firm whose head is a Trump science and tech adviser.

“I want to start with one figure that I think really emphasizes the investment opportunity of defense, and that’s $1 trillion. The U.S. government, for the first time, is allocating $1 trillion to its defense spending. And $13.5 billion of that will go to autonomous systems and AI,” said Kyle Stanford, a venture capital research director at PitchBook, a market intelligence platform. “That has obviously brought a lot of capital into defense tech spending, more than we’ve ever seen.”

New concerns emerge on AI safety

A day later, three researchers at OpenAI’s headquarters said the ChatGPT maker had fired them, posting that “fear and unclear rules stymie AI safety work.” Separately, David Robinson, a leader on OpenAI’s safety team until he recently resigned, said he was worried about letting the advanced labs police themselves.

“This is no substitute for hard law, and it does give a lot of freedom to the companies,” Robinson said. “The time between model generations is measured in weeks … and I think we have some very serious safety challenges in front of us in the nearer term.”

Trump has staked his political capital on the AI industry’s success, saying its unbridled development is key to winning the AI race against China. On the first day of his second term, the Republican president repealed a Biden-era executive order aimed in part at guarding against the risks that AI systems pose to consumers, workers and national security.

His administration has since reoriented federal policy away from oversight and toward rapid growth, including by canceling previous requirements for AI developers to share safety test results with the federal government. In December, Trump signed an executive order aimed at blocking AI regulations at the state level, saying a patchwork of rules could stifle the industry.

Even as Silicon Valley companies have disclosed their models’ involvement in troubling cyberattacks throughout the year, the burgeoning tech sector has continued to drive economic growth.

Concerns about AI safety centered on existential risk, rather than current real-world harms, landed at the center of debates last month, when a relatively unknown Anthropic engineer quit and called for a pause on tech development to keep “superhuman” systems from eluding controls. Then, Anthropic and OpenAI CEOs echoed the call to slow AI development. To some critics, their public manifestos appeared aimed at currying favor with investors ahead of their public listings and the midterm elections, when the political tides could turn with many voters opposed to data centers.

“The only control or ‘guardrails’ that AI needs is a STRONG AND SMART (High IQ!) PRESIDENT, and the U.S.A. has that, in spades!” Trump posted in response on Truth Social, warning voters not to “kill the Golden Goose,” a term he has used to refer to both AI and data centers.

White House pact preserves wide latitude for AI companies

As calls mounted for stricter control of AI, the administration brought tech executives to the White House for lunch, followed by a ceremonial signing of the agreement on AI safety.

“I’m seeing tremendous self-policing,” Trump told reporters after the event.

The document was signed by representatives of OpenAI, Anthropic, Meta, Google, SpaceXAI and Nvidia.

OpenAI is committed to embedding external assessors and finalizing contracts with third-party safety assessors, said a spokesperson, who referred AP to a company post stating that the researchers were fired for violating “clear policies on handling sensitive information.”

Google referred to CEO Sundar Pichai’s post that the company does appropriate testing and evaluations and is collaborating “to establish norms and build public confidence.”

Meta recently updated its standards for training, evaluating and releasing its most capable AI models, and spokesman Andy Stone pointed to the company’s efforts to refine its board of directors’ governance structure.

Nvidia and SpaceXAI did not respond to requests for comment.

Anthropic, which voluntarily disclosed Friday that its agents submitted a false tip to a Philadelphia police website about an unsolved homicide case, is bringing in independent evaluators and discussing collaborations with industry groups, a spokesperson said.

In a statement, the White House reiterated Trump’s comments about working with industry to balance innovation and security and did not directly answer questions about how the administration plans to monitor the AI companies’ implementation.

Following the new Anthropic disclosure, it added that the Trump administration expected full transparency from the company, saying notification and remediation was “not optional. It is a critical national security obligation.”

Under Democratic President Joe Biden, leading AI companies also made voluntary commitments to promote “safe, secure, and transparent development” of their models that included specific guidelines for tests and transparency reports. The current accord lacks that, said Sarah Myers West, a former senior AI adviser to the Federal Trade Commission.

“Across the board, this self-regulatory approach has consistently done one thing, which is failed to work. It’s failed to in any meaningful way ensure that these companies are honestly even keeping their house in order, let alone ensuring that they are compliant with existing law,” said Myers West, now co-executive director of the New York-based nonprofit AI Now. “All that they’re getting is grace from this government.”

After last month’s luncheon with Trump, the FTC said it had opened an investigation into OpenAI, Anthropic and other companies over dangers their technology may pose to consumers. The FTC didn’t immediately respond to a request for comment.

Trump recently appointed National Intelligence Director Jay Clayton and FTC Chairman Andrew Ferguson to serve on his new “Super Intelligence Force,” which the president said would keep the U.S. dominant in AI development. He posted Thursday that the White House would consider anyone who does not refer to AI using his preferred term as “THE ENEMY!”

Members of Congress, as well as civil liberties and consumer rights groups, have pushed for more regulations on AI, saying there is not enough oversight.

Graham Steele, a former assistant secretary in the U.S. Treasury Department, said even if AI companies set up internal evaluation teams or hire outside evaluators, such a structure can make them vulnerable to undue pressure on their work.

“We live in a democratic society. Are they the right ones to have the power to be able to make decisions that have broad, structural, social, and political consequences?” Steele said.

—Garance Burke, Associated Press

In early October, California state and city authorities reached a landmark settlement with the rideshare company Lyft over the state’s allegations that the company misclassified its drivers as independent contractors instead of employees, denying them wages and benefits they would have otherwise been owed between 2016 and 2020. If a judge approves the settlement—involving the state attorney general, Rob Bonta, and attorneys for San Francisco, San Diego, and Los Angeles—the company will pay $272.5 million, the largest wage-and-hour-settlement in California history.

But the historic settlement also underscores the limits of the legal victory for gig workers. While Lyft is now poised to pay hundreds of millions of dollars over its past treatment of drivers, the company has since helped rewrite the very labor rules that are at the center of the dispute. In 2020, Lyft and other gig companies bankrolled Proposition 22, whose approval by voters allowed them to continue classifying drivers as independent contractors. The result is a settlement that compensates drivers for alleged past violations but does little to change the conditions under which they work today.

For drivers, it’s a bittersweet victory. “It’s a freaking absolute victory for drivers,” says Nicole Moore, president of Rideshare Drivers United. “This is a little bit of justice to get back what they stole from us.” But, she argued, the settlement should have come sooner, and the amount should be far higher.

Lyft, for its part, maintains that it has always classified its drivers appropriately. “If approved, this settlement closes a chapter from a very different time, before Prop 22,” George Flynn, a spokesperson, said in a statement. “Lyft believes drivers have always been properly classified under the law, and we’re glad to put this case behind us.”

‘Lyft really owes drivers $20 billion for those four years’

The settlement is the culmination of a years-long fight over how gig companies classify their workers. In 2018, California’s Supreme Court established a stricter test for determining whether workers could be treated as independent contractors, which the state subsequently codified in law.

Drivers by the thousands started filing wage claims with the state to try to vindicate their new rights; between Uber and Lyft, they ended up filing about 5,400 claims totaling the hours and miles they had spent driving, as well as all of the expenses they had paid to maintain their cars. They calculated that they were owed $1.3 billion in aggregate by the two companies. Just for Lyft, Moore says, the total they said they were owed was more than $434 million. In response, California and several city governments sued both companies in 2020 over their classification practices, setting the stage for the settlement reached this month.

Uber and Lyft didn’t change their business practices, however. Instead, they spent $100 million lobbying for Proposition 22, and it passed in November 2020. It has meant that drivers can’t submit any misclassification claims after 2020.

Moore isn’t sure how many drivers will be covered by the new Lyft settlement, but she estimates it could be 200,000. That means each individual driver stands to get far less than what they had demanded when they filed their wage claims. “Lyft really owes drivers $20 billion for those four years,” she tells Fast Company.

The settlement doesn’t impact Uber drivers. The state government’s lawsuit against Uber has a court date set for later next year, and a similar settlement could be reached before then.

In the meantime, drivers aren’t sitting idle. They’re still fighting the larger battle over whether they should be classified as independent contractors in the first place. In June, they filed a lawsuit claiming Uber has reneged on promises it made in Proposition 22, which they argue means the law should be invalidated and the state’s stricter labor laws be reinstated. “That absolutely continues on,” Moore says. If this lawsuit were to succeed,  it would open the door for drivers to seek compensation for wage theft and misclassification from 2020 forward. “We have to weaken Prop 22 in order to be able to really compensate drivers fairly,” she adds.

The Lyft settlement, Moore vowed, is just one step in their quest to be paid fully. “The reason this is happening is because drivers kept this fight alive,” she says, “and we will continue to keep this fight alive.”

You have probably noticed it on a hot summer day: A basement feels cool even when the air outside is baking. The ground does not heat up and cool down as fast as the air does. A few hundred feet down, it stays close to the local yearly average all year, which can be around 50 degrees Fahrenheit (10 Celsius).

In Verona, Wisconsin, the health software company Epic Systems has spent two decades turning that fact into infrastructure. It has drilled roughly 6,100 wells, 300 to 800 feet (90 to 240 meters) deep, that now heat and cool about 40 buildings on its own campus by using the waste heat from a 3.5 megawatt (MW) data center.

What makes this setup unusual is not that Epic pulls heat out of the ground in winter—ordinary ground-source heat pumps do that in homes across the country. What’s uncommon is that Epic puts heat into the ground in summer and takes it back out months later.

That matters more than it sounds because timing is what usually kills the idea of reusing a data center’s heat. Servers make heat every hour of the year. In most of the U.S., buildings need to be heated for only about four months. Piping it straight to nearby buildings works beautifully in February, but solves nothing in July. Storage is the missing piece.

You may have seen data centers and geothermal energy in the news together lately. Google is buying 115 MW of geothermal power in Nevada, and Meta has contracted for 150 MW from a geothermal plant in New Mexico.

Those projects drill for heat the earth already holds, deep enough and hot enough to generate electricity. Storing heat underground is the mirror image. Same drilling, same rock, but the heat comes from the building, and the ground’s job is to keep it rather than supply it.

The snow pile in the parking lot

Heat can behave in surprising ways. A thin layer of snow on a driveway may be gone by afternoon, but a pile of snow the plow pushed into the corner of the lot can still be sitting there in April. Same snow, same weather. The difference is shape. The pile has less surface area for its volume, so less of it is exposed to the warm air.

Storing heat underground uses the same principle, though in reverse. A little heat leaks away quickly. But push enough of it into a large enough volume of rock, and most of it is still there months later because heat can escape only through the edges of the warmed zone.

In July, water heated by the servers is pumped through a pipe deep underground, where it warms the rock around the pipe. Rock is a poor conductor of heat, which sounds like a defect but is in fact a crucial element. It’s why the warmth stays close to where it was originally stored.

When January comes, water is pumped back through the pipe in the ground, which still contains much of the warmth stored in July. The water picks up that heat and returns to the surface warmer than it left. There is no tank buried down there, and no battery is being charged. The rock is the storage, and what’s being stored is heat.

The water that comes back up is lukewarm. In winter, similar systems in the Netherlands pull their stored heat out with water that surfaces at around 59 F.

What the heat is good for

Much of my own research goes into wells that turn stored heat back into electricity. In those systems, surplus solar and wind power runs a heat pump that heats material packed into the well to as much as 284 F, and that heat later drives a generator.

At the temperature a data center’s servers produce, about 140 F, the recovered heat isn’t enough to drive a turbine to generate electricity. But it is efficient for warmth; a heat pump can move large amounts of heat with relatively little electricity.

At Epic, that heat warms offices, supplies hot water to the dining halls, and melts snow off the sidewalks.

A problem

Boreholes into rock face a problem: too much heat building up. If more heat goes into the ground each summer than comes back out each winter, the rock around the boreholes slowly gets warmer year after year, until the field can no longer take heat at the temperature it was built for.

As Epic’s campus grew, the company drilled more boreholes to be able to store, and use, more heat.

There are two ways to fix that imbalance. One is to use more of the heat, such as preheating water for the buildings’ boilers or melting snow. The other is to give the heat more room, drilling more boreholes so it can be spread through more rock. But drilling takes land, equipment, and money. Epic hasn’t published what its system cost, though the U.S. Department of Energy reports that Epic’s buildings use about 25% less energy than comparable buildings in the same climate.

Part of the cost could be avoided by using holes that are already in the ground. The U.S. has more than 141,000 documented orphaned oil and gas wells waiting to be plugged. My colleagues and I have modeled how idle wells in California could be turned into thermal storage without replacing their steel casings.

However, in most states, a well is regulated either as an oil and gas well or as a plugged one, with no category for turning it into heat storage. A colleague and I have proposed a streamlined permit process that states could adopt to allow wells to be used for this additional purpose.

Two other limits are worth saying plainly. The first is distance: The stored heat sits under or beside the facility, so whoever uses it has to be close by. The second is size: Epic’s 3.5 MW data center is tiny next to the AI campuses now being built at 100 to 1,000 MW, and its borefield serves the whole campus, not just the servers.

No one has yet stored a data center’s heat underground at that scale. However, a similar system that stores heat in underground water reserves called aquifers has been providing 60% to 80% of the winter heat demand of a complex of offices, a hotel, a medical center, and a data center in Bonn, Germany, since 2009.

An opportunity

The heat is not going away—but it needs to go somewhere. In the U.S., data centers used about 4.7% of the country’s electricity in 2024, a share the Lawrence Berkeley National Lab projects could reach nearly 12% by 2030. And in some places, that proportion is even higher, accounting for 15.4% of all electricity consumed in North Dakota, where I teach, and 25.6% in Virginia.

About 30% to 40% of that usage is for cooling, and the heat it removes goes into the sky, as the water heated by electronics evaporates.

Communities are already fighting about data centers raising their power bills and using up precious water. Making different choices about how to handle the heat could help keep power bills down and water in local rivers.

Moones Alamooti is an assistant professor of energy and petroleum engineering at the University of North Dakota.

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

Thanks to Siri AI and Shortcuts in iOS 27, I’ve finally created a better version of Shazam.

Shazam is great for identifying songs you hear in the wild, but not so great at revisiting them later. It requires either Apple Music or Spotify to play the songs in full, and there’s no way to add more context about what made you tag a song to begin with.

Now, I can press a button on my iPhone to make Shazam work the way I want: After identifying a song, it adds the name and artist to a “Shazam Tracks” page in Obsidian, includes a link to a free stream of that song on YouTube, and provides space for me to jot down where I was or what I was doing at the time.

None of this required any coding knowledge on my part. Instead, I just described my ideal Shazam setup in Apple’s Shortcuts app, and it assembled the whole thing for me.

Apple’s Shortcuts app has been around since 2018, letting you chain a series of actions into powerful routines. But while it’s a neat tool for power users, in the past it’s required a lot of tinkering with variables and if-then statements. The new “Describe a Shortcut” feature in iOS 27 makes the process much simpler, and it’s got me looking at iOS automation with some new energy.

Describing a Shortcut

To use Describe a Shortcut, you need an iPhone, iPad, or Mac that supports Siri AI:

• iPhone 15 Pro models or any newer iPhone.
• iPads with M-series processors or the A17 Pro iPad Mini.
• M-series Macs or the MacBook Neo.

You will also need iOS 27, iPadOS 27, or MacOS 27 Golden Gate, and will need to opt into the Siri AI beta under Settings > Siri. There’s a waitlist, but getting through it only took a day in my experience.

Once you have Siri AI access, open the Shortcuts app and hit the “+” button. Then, describe in natural language what you want to accomplish.

Here are some examples of Shortcuts I’ve created by describing them:

• “Full charge” shortcut that overrides my usual 85% battery limit until the following day.
• “Urgent reminder” shortcut that schedules a task in the Reminders app with a full-blown alarm instead of a simple chime.
• “Rambler” shortcut for dictating a lengthy amount of text, cleaning it up using Apple’s cloud-based AI models, and then sharing or copying it to the clipboard.
• “Turn on grayscale” shortcut that switches the screen to black-and-white mode at bedtime, while also showing options to turn grayscale back off or delay it by five minutes.
• Companion “Turn off grayscale” shortcut that activates at wake-up.
• “Scan Document” shortcut that loads the document scanner in Apple’s Files app.

Once you’ve created a Shortcut, trigger it by tapping its icon in the Shortcuts app or asking Siri to run it by name. For even easier access, add it to your home screen (long-press the Shortcut and hit “Share” to see this option), assign it an iPhone’s physical Action Button (under Settings > Action Button), or use Back Tap.

You can also run Shortcuts automatically, based on triggers like time of day, your location, phone settings, charging status, battery level, Focus mode, or when a specific app is running. These triggers now live directly inside each Shortcut, so you can include them while describing the Shortcut you want to create.

Fine-tuning

Like an AI chatbot, Apple’s “Describe a Shortcut” prompt will respond to follow-ups, which is helpful if the Shortcut doesn’t do exactly what you wanted.

While creating my “Urgent reminder” Shortcut, for instance, the app misinterpreted my request and merely added a “flag” toggle to the reminder. I had to make clear that the reminder should trigger an alarm instead. The more precise and descriptive you can be, the better.

The old way of creating Shortcuts also remains available by tapping the “Edit” button. This brings up a menu where you can manually chain together new actions or modify existing ones. While you shouldn’t have to mess with this menu too much, it can be a useful reference for all the things that Shortcuts can do.

In a way, Apple’s new “Describe a Shortcut” feature is like a diet version of vibe coding, letting you bypass all the confusing steps that iOS and Mac automation once required. Even if there’s no actual code involved, it’s a way to accomplish things that may be just a bit beyond your technical grasp, and it’s quietly one of iOS 27’s best new features.

You can read endless online advice telling you how to train AI to sound like you. Most suggest using AI to analyze your blog, your newsletter, and your social posts. 

That’s all fine, but you should also be asking AI to analyze your customers—the words they use to describe their pain points. A “voice of the customer” analysis or document is what companies often use to guide everything from web pages to sales material. 

If you’re a solopreneur, you’re the whole research team for a voice-of-customer analysis. The good news is that AI can help you build this document from conversations you’ve already had with clients. And the results might change how you describe your own work.

{"blockType":"mv-promo-block","data":{"imageDesktopUrl":"https:\/\/images.fastcompany.com\/image\/upload\/f_webp,q_auto,c_fit\/wp-cms-2\/2025\/11\/work-better-1.png","imageMobileUrl":"https:\/\/images.fastcompany.com\/image\/upload\/f_webp,q_auto,c_fit\/wp-cms-2\/2025\/11\/work-better-mobile-1.png","eyebrow":"","headline":"\u003Cstrong\u003ESubscribe to \u003Cem\u003E\u003Cem\u003EWork. Better.\u003C\/em\u003E\u003C\/em\u003E\u003C\/strong\u003E","dek":"Thoughts on the future of work, career pivots, and why work shouldn’t suck, by Anna Burgess Yang. To learn more, visit \u003Ca href=\u0022https:\/\/www.workbetter.media\/\u0022\u003Eworkbetter.media\u003C\/a\u003E.","subhed":"","description":"","ctaText":"SIGN UP","ctaUrl":"https:\/\/www.workbetter.media","theme":{"bg":"#f5f5f5","text":"#000000","eyebrow":"#9aa2aa","subhed":"#ffffff","buttonBg":"#000000","buttonHoverBg":"#3b3f46","buttonText":"#ffffff"},"imageDesktopId":91457605,"imageMobileId":91457608,"shareable":false,"slug":"","wpCssClasses":""}}
How a “voice of customer” document helps your business

Earlier in my career, I was the product manager at a financial technology company. Part of my job was translating what customers asked for into something the development team could build. I conducted endless interviews because it was critical to hear how customers described their problems, in their own words. 

You probably have similar context and conversations for your solo business, just not compiled in one place. A voice-of-customer analysis should capture your clients’ problems and goals.

It’s understandable that most solopreneurs skip this step. You talk to clients every week, so you probably feel like you already know what they need. But when you write about their problems, you’re likely using your own vocabulary.

Your clients’ words are useful anywhere you market your business, whether you use generative AI to help or write entirely from scratch. You can use your voice-of-customer analysis to guide proposals or to guide the questions you ask during discovery calls. They also work for email sequences or landing pages. 

When potential clients see their problems described using language they connect with, it’s easier for them to trust that you understand what they need.

Gather what your clients have already told you

Start by collecting raw material from wherever it already exists:

• Call transcripts. (If you use an AI notetaker, you probably have a record of your discovery calls and client meetings.)
• Emails and inquiry forms
• Testimonials and reviews
• Social media comments 

Conversations with people who didn’t hire you are also useful. Their objections often show you why someone hesitates.

One word of caution: You’ll want to strip out client names and identifying details before you upload anything, especially if you have agreements with your clients that govern the use of their data with AI tools. 

Ask AI to quote your clients directly

AI is good at sorting through a big pile of transcripts and other documents. The catch is that AI summaries tend to smooth out the specific language you’re after. So your prompt needs to ask for direct quotes, along with the source for each one.

Here’s a prompt you can use:

“I’m uploading [describe documents] from my clients and prospects. Find the problems they mention most often and the outcomes they want. Also flag any hesitations they had about hiring me, and any words or phrases they use again and again. Group everything by theme. For each theme, include direct quotes word for word, with the source. Don’t paraphrase or summarize.”

When the results come back, you’ll want to spot-check some of the quotes against your original transcripts, since AI is known to hallucinate. If a “quote” doesn’t appear anywhere in the source material, remove it from the document.

Keep it current as your business changes

I plan to refresh my voice-of-customer analysis regularly—at least once a year, and maybe more often. Now that I have the prompt and baseline document, it should be a quick update.

Most of the work comes afterward, when you go back through your website and sales copy. The voice-of-customer analysis is simply a starting point to rework the language you use to describe your business.

{"blockType":"mv-promo-block","data":{"imageDesktopUrl":"https:\/\/images.fastcompany.com\/image\/upload\/f_webp,q_auto,c_fit\/wp-cms-2\/2025\/11\/work-better-1.png","imageMobileUrl":"https:\/\/images.fastcompany.com\/image\/upload\/f_webp,q_auto,c_fit\/wp-cms-2\/2025\/11\/work-better-mobile-1.png","eyebrow":"","headline":"\u003Cstrong\u003ESubscribe to \u003Cem\u003E\u003Cem\u003EWork. Better.\u003C\/em\u003E\u003C\/em\u003E\u003C\/strong\u003E","dek":"Thoughts on the future of work, career pivots, and why work shouldn’t suck, by Anna Burgess Yang. To learn more, visit \u003Ca href=\u0022https:\/\/www.workbetter.media\/\u0022\u003Eworkbetter.media\u003C\/a\u003E.","subhed":"","description":"","ctaText":"SIGN UP","ctaUrl":"https:\/\/www.workbetter.media","theme":{"bg":"#f5f5f5","text":"#000000","eyebrow":"#9aa2aa","subhed":"#ffffff","buttonBg":"#000000","buttonHoverBg":"#3b3f46","buttonText":"#ffffff"},"imageDesktopId":91457605,"imageMobileId":91457608,"shareable":false,"slug":"","wpCssClasses":""}}

Since World War II, the United States has dominated science and technology. With more than 400 Nobel Prizes, it dwarfs every other country. Major breakthroughs like nuclear energy, genetic engineering, and the internet were all led by researchers and engineers in the U.S. Today, it leads in every major technological category.

To a great extent, this has been the result of government action. Federal agencies such as the NIH, NSF, and DARPA spearheaded many of the most advanced technologies, and U.S. research universities have become beacons for top scientific talent. These advances have brought us greater prosperity, better health, and enhanced well-being. 

That’s why the recent news of an executive order that would give political oversight to scientific grants provoked such an uproar. Although the administration quickly backed off, the incident represents a troubling trend of political interference in science that has led to a major brain drain of scientific talent. Without a major change in course, the damage could be permanent. 

Science, The Endless Frontier

America wasn’t always a scientific mecca. In fact, at the beginning of the 20th century, bright young students often had to go to Europe to pursue advanced degrees in the sciences because American universities were considered second-rate. Yet as the U.S. industrialized, that began to change and, by the 1930s, with Europe increasingly in turmoil, world-class scientific talent began to flock to American shores.

In 1940, Vannevar Bush, who at the time was serving as president of the Carnegie Institution for Science, went to see President Roosevelt. His proposal led to the establishment of the Office of Scientific Research and Development (OSRD), which mobilized the country’s scientific resources for World War II. It was, by all accounts, a resounding success, resulting in breakthroughs such as radar, penicillin, the proximity fuse and, most famously, the atomic bomb.

As the war was winding down, Roosevelt asked Bush to write a report proposing how America could continue the success of the OSRD in peacetime. That report, Science: The Endless Frontier, was delivered to President Truman in 1945 and would set the stage for America’s technological dominance during the second half of the century. Bush wrote: “Basic research leads to new knowledge. It provides scientific capital. It creates the fund from which the practical applications of knowledge must be drawn. New products and new processes do not appear full-grown. They are founded on new principles and new conceptions, which in turn are painstakingly developed by research in the purest realms of science.”

It was that vision that led to U.S. technological dominance. Look at any major technology we use today—whether it is the internet, GPS, or lithium-ion batteries—and the initial breakthrough came from government funding. A study from the Institute for New Economic Thinking found that every new drug approved between 2010 and 2019 benefited from government grants. Even major private firms, such as Google and Qualcomm, started out with government funding. 

A key tenet of that vision is that decisions about research grants be made by scientists, not political appointees, so that we are funding scientific, not political priorities. That principle is what has come under threat.

The Visceral Abstract

There is a long history of politicians being hostile to science. In a 2012 interview, then-Sen. and current Secretary of State Marco Rubio said that the age of the universe and the Big Bang “has zero to do with how our economy is going to grow.” Former Vice President Mike Pence expressed doubts about Darwin’s theory of natural selection. Paul Broun, a former Georgia congressman, called it “lies straight from the pit of hell.”

Yet the Big Bang is not just a theory, but the result of a set of theories, including general relativity and quantum mechanics, combined with many observations over a period of decades. Darwin’s theory is core to our understanding of life sciences. Students learn about these things not to shape their religious beliefs, but because they underlie basic scientific work.

Politicians attack science because it’s an easy target. It’s not hard for the public to imagine that their tax dollars are funding overeducated elites thinking up wacky theories in ivory-tower institutions. Scientific theories take time and effort to understand and can be maddeningly difficult to explain. 

The truth is that we live in a world of the visceral abstract, where strange theories underlie much of our everyday existence. Technologies that we use every day, such as the lasers that scan our purchases in the checkout aisle and the semiconductors that drive our devices, arose from these same strange ideas. Darwin’s theory underlies everything from fighting diseases like COVID-19 to guiding logistics systems. 

If we are to master the advanced technologies of the future—such as quantum computing, synthetic biology, and materials science—we need to master the scientific theories that will drive them.

The Tragedy of Trofim Lysenko

While Vannevar Bush showed what can be achieved when a government invests in science as a public good, Trofim Lysenko showed what happens when science is politicized. A Soviet agronomist who considered the concept of a gene a “bourgeois invention,” Lysenko—like Pence and Broun—was skeptical of Darwin’s theories. 

Yet he was able to get Joseph Stalin to support his ideas—such as his theory that exposing seeds to low temperatures will help them grow in the winter—which were mainly based on political expediency and personal belief, not on empirical evidence. He promised miracle crop yields that never materialized, but his power allowed him to suppress legitimate scientists.

The effect on Soviet biological science was chilling. Anybody reporting results that contradicted Lysenko’s theories risked not only professional suicide but, in many cases, imprisonment and even execution. To adopt “bourgeois” science was not just to have a different idea, but to be branded an “enemy of the people” working to undermine the Soviet project. 

Meanwhile, those who went along with Lysenko’s program gained authority, and competency in Soviet biological sciences was greatly undermined. Crop yields steadily declined throughout the late 1930s until his downfall in the mid-1960s. Many believe that his theories contributed to the Great Famine in China.

If Nothing Is True, Everything Is Possible

One thing that struck me during the 15 years I lived in post-communist countries was how problems seemed to build on themselves. In Kraków, the abandoned frame of the Szkieletor building stood frozen for decades. Construction had stalled in the 1970s, so the frame just stood there rusting. It became almost a metaphor for the Soviet space as a whole. 

In a healthy society with checks and balances, someone would have been brought to account. There would have been litigation. A court would have mandated a cleanup. But in a command economy, power flows from a single source, with no one to hold it accountable. Those in power don’t have to admit their mistakes. Questions of truth become questions of loyalty. 

We have a similar process underway in the U.S. When politicians and unelected bureaucrats seek to make scientific funding conform to their ideological preferences, they are not merely making budget allocations. They are seeking to control what is true and what is not. We’ve seen this enough times before to know how it ends. 

America achieved technological dominance, in part, because of the scientists who fled Europe to escape similar populist tyranny. Now, that process is reversing, with scientists leaving America for Europe. The advantages we built after World War II are quickly eroding, threatening our prosperity and security. 

That’s why we don’t want politicians running science. It diminishes science and leaves us with only politics.

Spam calls and text scams have evolved from a mild nuisance into a relentless daily tax on our attention. Scammers use neighborhood-spoofing to mimic local area codes and automated text bots to blast out thousands of phony messages all day and night.

Your instinct may be to download a call-blocking app that demands full access to your contacts and charges a $4.99 recurring subscription. But you can tame this scourge without paying a cent or handing over your address book.

Both iOS and Android include powerful, built-in spam-killing toggles that take less than three minutes to configure.

Here’s a quick settings audit to silence the noise.

Mute unknown callers at the system level

Instead of waiting for a spammer to call and manually blocking their spoofed number, tell your phone to ignore any incoming call from a number you’ve never interacted with before.

• On iPhone: Go to Settings > Phone (or Settings > Apps > Phone), tap Silence and Screen Unknown Callers. Calls from numbers not in your contacts, recent outgoing calls, or Siri Suggestions will automatically go straight to voicemail without ringing your phone.
• On Android: Open the Phone app, tap the menu icon in the top corner, select Settings > Caller ID & Spam, and turn on Filter Spam Calls. This uses Google’s database to automatically block suspected robocalls.

Quarantine text scams into a separate inbox

Text message scams (smishing) are particularly dangerous because a single accidental link click can expose account credentials. You can route messages from unrecognized numbers out of your main inbox completely.

• On iPhone: Go to Settings > Messages (or Settings > Apps > Messages), and toggle on Filter Spam under Unknown Senders. This creates a separate tab in your Messages app for unknown numbers and disables notifications for them.
• On Android: Open Messages, tap your profile icon in the search bar, go to Messages Settings > Protection & Safety Spam, and verify that spam protection is enabled.

Activate free carrier-level network shields

Your cellular provider maintains a network-level list of active scam call origins. They offer free baseline network filtering, but they rarely turn it on by default.

• AT&T: Download the free AT&T ActiveArmor app to toggle on network-level spam blocking and nuisance call warnings.
• Verizon: Open the free Verizon Call Filter app and enable Spam Filter to auto-block high-risk callers.
• T-Mobile: Dial #662# directly from your phone keypad and hit call to enable Scam Block at the network level, or manage it through the T-Mobile Scam Shield app.

Don’t accept life with a phone that rings six times a day with fake debt collector alerts or suspicious package tracking links. Spending three minutes to lock down your OS settings and activate your carrier’s network filters will eliminate roughly 95% of incoming spam before it ever reaches your screen.

Last week, OpenAI safety leader David Robinson resigned from the company, becoming the latest to join what is, in his words, a “parade of former colleagues” who have left roles at AI companies amid concerns about the tech or its use. In a subsequent essay for The Atlantic, he said the “trial and error” approach favored by AI companies must end.

Robinson is far from alone. Here’s a list of other AI researchers and employees who’ve made public warnings about their former employers.

ED NEWTON-REX

In November 2023, Ed Newton-Rex left his role leading the audio team at Stability AI, the maker of text-to-image tool Stable Diffusion, over concerns about the way copyrighted material was being used to train the tech.

“My opinion here doesn’t align with the generative AI industry’s current way of doing things,” Newton-Rex told Fast Company following his resignation, adding that he thought the industry’s prevailing outlook, paired with rapidly advancing tech, would be “very, very bad for creators.” 

Since then, he’s founded Fairly Trained, a nonprofit that issues certifications for AI companies whose data use meets certain requirements.

THE GROUP LETTER

On June 4, 2024, 13 current and former employees of top AI companies published a letter titled “A Right to Warn about Advanced Artificial Intelligence.” In it, they asserted that the technology poses risks ranging from “the further entrenchment of existing inequalities, to manipulation and misinformation, to the loss of control of autonomous AI systems potentially resulting in human extinction,” and expressed concerns about a lack of oversight. 

Signatories included former Google DeepMind employee Ramana Kumar, former Anthropic and current Google DeepMind employee Neel Nanda, who leads the company’s mechanistic interpretability team; and 11 current and former OpenAI employees, including Jacob Hilton, who now works for the Alignment Research Center; Daniel Kokotajlo, who founded and leads the nonprofit research group AI Futures Project; William Saunders, Carroll Wainwright, and Daniel Ziegler, the latter of whom now works for Anthropic; and six people who remained anonymous.

Prominent computer scientists Yoshua Bengio, Stuart Russell, and “Godfather of AI” Geoffrey Hinton also endorsed the letter.

JAN LEIKE

In May 2024, OpenAI Superalignment team lead Jan Leike announced his departure on X, stating that “safety culture and processes have taken a backseat to shiny products.” Leike, who has also worked at Google’s DeepMind, now heads Anthropic’s alignment team.

“Now is a good time to build institutional mechanisms to pace the frontier of AI development,” he said on X in September. “The industry is locked into an all-out scaling race to build superintelligence as quickly as possible, and we may need to give everyone more time for safety and alignment mitigations.” 

Leike added, “I’m not the only one who believes this,” linking to a statement titled “Pacing the Frontier” signed by 1,386 employees of major AI companies.

SUCHIR BALAJI

After graduating from the University of California, Berkeley, Suchir Balaji was hired by OpenAI, where he spent about four years working on ChatGPT and other projects.

He left the company in August 2024, on the same day as cofounder John Schulman, who said the timing was coincidental. Balaji blew the whistle on what he said were copyright infringements by Open AI, and published an essay online explaining how he thought ChatGPT’s outputs violated Fair Use law. 

That November, he was named as a potential witness in The New York Times’s lawsuit against OpenAI alleging willful copyright infringement.

The 26-year-old was found dead in his San Francisco apartment eight days later. His death was ruled a suicide, though his parents and others raised questions about that determination, as reported by The Guardian.

MRINANK SHARMA 

In February, Mrinank Sharma announced his departure from Anthropic, where he led the Safeguards Research team. 

“I continuously find myself reckoning with our situation,” Sharma wrote in his resignation letter, which he shared on X. “The world is in peril,” he continued, not just from AI and bioweapons, but “a whole series of interconnected crises.”

ALEX TURNER

Alex Turner left his job as a Google DeepMind research scientist in June, after a failed campaign to stop the company from signing a contract that allows the U.S. Pentagon to use its AI “for any lawful government purpose,” which he and others worry may allow for mass surveillance and autonomous weapons. 

Last month, he testified for a U.S. Senate hearing on rogue AI. “Often we think about the movements of our overseas adversaries, but now we must consider how we may be building adversaries at home by training misaligned AI systems,” he said in his written statement.

JACOB COXON

Jacob Coxon, a former researcher for OpenAI and Anthropic, resigned from the latter last month. 

“Neither company is acting responsibly,” he said in a viral X post. “They are racing straight to self-improving superintelligence and gambling with our lives.”

He went on to assert that “the people building AI earnestly believe that it could kill us all by the end of the decade.”

In an interview with PBS in September, Coxon said an AI race with China “could be very bad, and that we should be taking pretty rapid actions for international coordination,” but that he believes the company leaders echoing that desire “are privately quite skeptical” of being able to achieve that cooperation.

MIKITA BALESNI, TOMEK KORBAK, AND JASMINE WANG

Last week, three OpenAI safety and alignment researchers were fired for mishandling sensitive misinformation, which one of them, Tomek Korbak, said online was due to how he’d communicated information to METR, the AI research nonprofit investigating OpenAI’s recent Hugging Face hacking incident.

“AI is not a normal technology, and OpenAI is not a normal company,” the researchers said in the letter they sent to company leadership, which they published online on October 8. “Those of us who work on safety see risks before anyone else, and we rely on close collaboration with outside experts to work out how to address them. The freedom to do so without fear, and to have well-defined internal procedures that enable this work, is itself an essential safety mechanism.”

OpenAI released a statement the following day saying the researchers were fired after an internal investigation “found they violated clear policies on handling sensitive information,” adding, “We want to be very clear that these decisions were not about raising safety concerns or speaking out.”