Skip to main content

Fast Company

What not to do as a CEO pushing AI? Call your workers ‘losers’

2 weeks 4 days ago

Plenty of CEOs are bullish on AI and have urged their employees to embrace it enthusiastically. A more unusual approach is to brand your workers as “losers” for not using AI enough. 

A new investigation this week from 404 Media shows how one CEO is reportedly not only pushing AI, but practically bullying his employees into using it to work more quickly. At Proton—the company behind the encrypted email service Proton Mail—the leadership team has become singularly focused on AI adoption, 404 Media reports, putting pressure on employees to write code with AI and move faster. 

In a recent Slack exchange reported by 404 Media, CEO Andy Yen suggested it was “bullshit and loser thinking” to spend a few days testing new code for security issues before going live. “Of course it’s possible to be fast and secure,” he wrote. “Stop thinking always of how to be slower, and focus more on how to be faster.” Another message emphasized, again, that “only losers think like that.”

Yen’s attitude toward his employees has made many of them fearful of expressing concerns about security vulnerabilities with AI-generated code. As one worker told 404 Media: “Whatever persona Andy [Yen] has out of Proton, he’s a straight-up bully inside. He routinely breaks his little facade in Slack and calls people losers and stupid, often for stuff he’s paying them for.” 

To the outside world, Yen may appear more cautious about AI adoption, in line with what the public might expect from a privacy-first company like Proton. But the 404 Media report paints a picture of a company that, in the interest of meeting revenue targets, is both forcing AI on its employees and being punitive if they don’t eagerly adopt it. 

When reached by Fast Company, Proton did not immediately respond to a request for comment. 

In a statement to 404 Media, the company did not deny the reporting: “It’s clear that some people at Proton are unhappy with our direction, or misunderstand it, and many people are anxious. In our determination to move fast, we have sometimes been too blunt or too slow in explaining ourselves to our people.”

The company’s current position on artificial intelligence seems to be a major departure from its prior stance. (AI use was previously banned at Proton.) Now, however, Proton is tracking the volume of AI-generated code used by its employees through software installed on their computers. Yen has set up an email address that employees can use to alert leadership to anything that might stand in the way of AI adoption, enabling the company to “ruthlessly smash everything that stands in the way of AI velocity.” According to 404 Media, he has said that after this year, Proton will no longer hire engineers, and will instead employ “what we used to call product managers, to build things with AI.” 

Perhaps most notably, previous reporting from 404 Media revealed that the company planned to conduct “re-interviews” with its employees about their AI usage, allegedly with the understanding that their jobs were on the line. (Those interviews have since been canceled, in response to 404 Media’s investigation.) 

In a Slack conversation about the re-interviews, one employee dared to voice their concerns to Yen about the culture at Proton. “People shouldn’t be scared to talk about this in public and to push back on you and leadership. A fear-based culture feels like a cancer to me,” the employee said. Later in their exchange, Yen suggested that “this might not be the place for you.”

As companies race to keep up with AI advances and strive to boost productivity, some leaders may believe this kind of culture and leadership is not just necessary but acceptable. Yen, for his part, seems to be doubling down. An internal document obtained by 404 Media hints at his plans for the company, which involve creating an “elite fast team” that will work longer hours in exchange for higher pay and generous perks. 

Also on his wish list? Becoming a “996 company” (9 a.m. to 9 p.m., six days a week), a reference to the grueling work culture first popularized in China that has now been endorsed by some AI startups. But from the looks of it, Yen may have a difficult time getting employees on board with the idea.

Mozilla has a plan to rescue us from the zombie internet

2 weeks 4 days ago

There is an uneasy feeling these days when surfing the web that the internet has irrevocably changed. Log on to social media and you’re wading through a sea of influencers psy-opping you into buying stuff you don’t need. Scroll through LinkedIn and you can barely tell whether a post was written by a human or generated by Claude to rack up clicks and likes. Everywhere you look, the internet feels a little less human—and increasingly, it is. Recent studies suggest that at least half of the content posted online is AI-generated.

How, then, do we reclaim the internet as a space for genuine human connection and creativity? That’s partly the question the Mozilla Foundation, a 23-year-old accessibility-focused nonprofit behind the Firefox browser, is trying to answer with this year’s MozFest in Barcelona.

Held this week, the Mozilla Festival is an annual global gathering of developers, policymakers, and tech builders focused on building a more human future for technology. This year, the festival asked how to make the internet more human—or, in its words, how to “rewild” it. 

Fast Company spoke to Nabiha Syed, executive director of the Mozilla Foundation, about this year’s festival, its themes, and why she believes nothing about the internet’s future is inevitable.

This interview has been edited for length and clarity.

What does “rewilding” mean in the context of MozFest?

We can acknowledge that we live in a moment where the internet sometimes feels like a monoculture. There’s like six dudes in hoodies who are making decisions about what everybody should be consuming: the platforms we should be on, the things that shape the market. And what we know from nature is that monoculture works until it doesn’t—until there’s a collapse in an ecosystem. 

So much of the open-source ideology that is core to Mozilla is about creating the structures for things to be wild and unpredictable and to go off in different directions and then later cross-pollinate. Last year’s theme for MozFest was “unlearning”: “Hey, we need to pause. What got us here isn’t necessarily gonna be what gets us to the next place.” Wilding is about saying, “What would it look like if we just let things bloom? What if we had more people have more say over the way that we think the internet’s supposed to unfold and be?” 

How would you describe the wild internet? And what is illustrative of a monoculture internet to you?

I think about when I first learned how to code it was because I really wanted my MySpace to be amazing. I would view the source and make my page look like whatever I wanted it to look like. When I couldn’t quite get it that way, I’d go to GeoCities and make my own site in my own way. And there was freedom, interoperability, and contestability. 

Even in the blogosphere, the way that we spoke to one another felt like we were actually having conversations and not worrying about bot interjections. There was a feeling of possibility and not just predetermined ways of interacting, like, “You can comment in this way” or “You can engage in this way.” 

You got to create and build with the architecture as you wanted it. Every generation is always creative on any communications medium, but within the confines of a platform like Instagram. They can be innovative in how they use Finsta, or the way they shitpost, or the way they talk. But it’s still within the architecture that someone else has designed. 

What the early internet felt like to me, and to so many people who have nostalgia for that time, is that you could play with the rules of the game, and they were committed to interoperability, transparency, and being able to peer under the hood. 

What are the consequences of that lack of interoperability, or the confined nature of our current platforms?

To me, the big thing is that it creates what we call imagination deficit. If you think that innovation comes from what the platforms provide to you—and your job is to play within the sandbox as it is created for you—it starts to inhibit your understanding of what is possible. Everything’s gonna be an endless scroll. Everything starts to look the same. Everything starts to look like it’s Instagram but for X. It’s Uber but for Y. Things start to become cookie-cutter.

Over time that changes our understanding of the possibility, which is ultimately very bad for, you know, big concepts like our cognitive liberty, but also our ability to play online. 

I think what was so special about the early internet is the feeling that you could just make new things and people could stumble upon them. It wasn’t mediated by algorithmic pseudo-wisdom or nudging you towards commercial ends. We are creating for one another—not to sell a course or become an influencer, or that always-lurking commercial logic that’s on these platforms. Removing that logic, or at least creating other spaces where it doesn’t matter as much, is what’s so interesting about early internet times versus platform-dominated times.

Right. Well, I guess then the question is, where are those spaces online undominated by commercial incentives? 

There are two kinds of spaces. You can see it in “dark social” environments. You see it in the group chat, you see it in the Signal chat, you see it in these closed spaces where people feel like there can be a clear organizing principle of why you’re there. You’re not trying to show off for everybody. We’re here to actually communicate and create for one another. 

The second are [part of the] Dark Forest Operating System, some of the Yancey Strickler [cofounder of Kickstarter] stuff of re-creating slightly bigger groups. What’s interesting to me about DFOS is that that mentality feels a lot like kind of bulletin board-y vibes. It feels like we’re re-creating spaces that were very Usenet-y. But they’re there, and if you know, you know. You’re invited into them. And I think that’s useful.

I don’t think everything needs to be radically, radically, radically open in order to have meaning. That’s not how human beings organize themselves. But what I worry we’re losing is that if you only have nodes of dark social, or dark social types of principles, and no common, or not as many common places to all be together, then we endlessly silo. And then we lose sort of the moments and spaces of cross-pollination.

Do you see any open spaces that still have that ethos?

Rudy Fraser from Blacksky is one of our fellows. We’re actively celebrating him as a fellow and supporting his work as a fellow. I think there are experiments there, and we want to encourage them.

One of the reasons we have MozFest is that you want enough of those spaces to have moments of intersection and cross-pollination. I think we’re trying to figure that out. We’ve been to a place of radical openness of Twitter, where everything is out there. Then everyone is swinging to, like, the dark social, where it’s, like, the eight people in your chat.

And now, if there are community-run federated platforms, then the work is like, “Okay, it’s not super small, it’s medium-size. How do they intersect in a mosaic together? Where does that happen?” And that’s really a lot of the purpose around MozFest: to have people’s own communities intersect with other communities in a dynamic way.

Let’s talk about that a little bit. You mentioned Rudy from Blacksky as a Mozilla fellow. I think we should give readers a little bit of tone-setting about what it means to be a fellow at Mozilla, for people who might only know Mozilla as the cute fox browser. What programs establish these spaces and foster these connections?

Thank you for the invitation to share that: Mozilla Foundation’s core thesis at this moment is that imagination is not the domain of Silicon Valley. It belongs to the world, and that reflects the core initial promise of the internet. 

Our work is to build from that imagination. We know that people are imagining different technology futures that are actually pro-human agency, pro-community—but they don’t always get the resourcing and the support that they need. Our intervention’s into that: our fellowships.

Over the last few years, we’ve supported over 200 different fellows who share our one fundamental belief, which is that no part of technology is inevitable, and we get to decide. And that can be everything from spinning up OpenNews and talking about a new way of doing journalism, to having fellows like Camille François, who just started Robust Open Online Safety Tools (ROOST), to having fellows like Rudy to having fellows like Matisse, who’s working on open-source quantum.

There are a lot of different areas of technology that all benefit from the idea. We’re looking at cybersecurity. The internet is not what the internet was 25 years ago. It shows up in all these different ways.

What kinds of projects does Mozilla’s incubator fund?

Earlier this year, we incubated Mozilla Data Collective, where we invested $10 million into a double-sided data marketplace that has ethical, fair licensing as the core of it. For a lot of creatives, they’re not against AI wholesale. They’re just against having their content taken and not getting compensated for it, and used to enrich a few. So creating a data marketplace where people can actually decide the terms under which they license their data is a way of having infrastructure for a better balance between creativity, intellectual property, and the reality of tech.

We have an incubator cohort on democracy and AI, where we were like, Is it democracy and AI? Democracy versus AI? We don’t know. Let’s figure it out. And so we use our studio to step in to say: There’s a lot of hype on all sides of the AI conversation. What does it look like to put half a million dollars behind folks—each different project—that are actually trying to figure out how this works and what it looks like to build in a good way? 

Half a million dollars is intentional. It is the same amount of money that Y Combinator gives out. But we think it’s important to have a different incentive of not short-term financial return, but building for mission and building for what we call product-community fit, not just product-market fit. So that’s the second big thing that we have, which is our studio.

We just opened yesterday our open call for what we’re calling human-scale AI. Bigger is not better when it comes to AI. You do not need to rush for superintelligence to get a lot of the benefits. You can use small, local, private, tailored models, and we are incentivizing people to do exactly that.

And then the last is: Look, you can champion amazing ideas, amazing people, amazing prototypes, and early-stage concepts, but you want them to find each other, and that’s where MozFest is really this flagship offering. Once a year, it has been happening for over 15 years, and it is a place for people all around the world to come together and say, like, “Hey, what are you building?”

Could you go into that incubator a little more? How are you rethinking incentives around funding, and what’s the intention behind that?

This was born out of a very simple observation: For a lot of ventures, the core incentive is getting a financial return—often in the short term. That incentive creates a whole lot of different behaviors that we think don’t encapsulate the full universe of everything that needs to exist in the world. 

We’re asking what it would look like to fund people who have projects where they’re like, “Hey, I don’t think I want this to just be venture-backed. I’m not trying to make crazy money off of this. I just think this should just exist.” 

Where are the spaces to do that? Philanthropy is one space to do it. You can get grants. We have philanthropic risk capital. We can make grants and also say, “Hi, welcome to the entire ecosystem of Mozilla”—of security researchers, of engineers, of developers, of policymakers, of operators who, over two decades, have this wealth of experience and this incredible network in over 120 countries—and open that up to people who are building. 

In our incubator, we offer $500,000 of support. It’s $300,000 at first and then $200,000 follow-on. We offer wraparound support, coaching, business model coaching, connections to other folks. It’s really bespoke support of, like, “What do you need?” In the last year, we helped someone figure out how to get a government contract, figuring out whether they should merge with another organization or not. Like, all of that messy reality of building something that’s beyond money, we bring to the table too, and the repeated encouragement that building for your community is going to be more important than building for a fast buck. 

I think that’s especially true in the AI age, where the cost of spinning up software is plummeting. The thing that will give durability and sustainability is having a community that’s bought in on what you’re doing, that will help you figure out the right business model. We do a lot of support around: What are the right business structures? What’s the right mission lock? How do you write your bylaws? (Obviously, as a nerdy lawyer by training, I’m like—a lot of this happens in governance—and all of those are the facets of what we bring to the table.)

It leads me to think that there are probably certain qualities that Mozilla’s risk capital looks for in founders or people looking to start projects that are out of the norm for other investment financiers. 

We have a spirit that we try to select for, which we call defiant optimism. You’ve got to believe the world is gonna be good ultimately, and can be good ultimately, but also stay ready to fight to get there, right? Hopeful optimism is “Yay, it’ll be fine.” That’s not our vibe. 

Our vibe is you gotta be scrappy, you gotta be willing to fight, you have to have a little bit of that defiance—and you also believe that things are gonna be okay.

Carney targets June deadline to finalize Canada’s historic EU associate membership

2 weeks 4 days ago

Canadian Prime Minister Mark Carney said Friday he expects to conclude negotiations with the European Union in June on an unprecedented partnership that would make Canada the bloc’s first “associate member,” even though the designation was left out of a joint declaration launching a new economic and security alliance.

Carney said he would attend a meeting of European leaders in June to conclude negotiations on the proposed relationship, which would stop short of full EU membership.

“In parallel, we are working towards the deepest possible partnership, short of accession—with Canada becoming the European Union’s first associate member,” Carney said.

The joint declaration issued at the end of a two-day summit in Montreal made no mention of associate membership. Instead, the leaders agreed to define an “ever-closer association” by their next summit in 2027.

European Council President António Costa said the alliance would be the closest relationship the EU has ever established with a country outside the bloc, expanding cooperation on trade, research, and security as U.S. President Donald Trump’s trade war strains relations with traditional allies.

European Commission President Ursula von der Leyen said shared values mattered more than geography in determining which countries should work together.

“It is not so important where the countries are on the map,” von der Leyen said. “What is important are the values that a country defends and where countries stand in international relations.”

Carney said the proposed alliance would strengthen the sovereignty of Canada and Europe and help them weather economic shocks.

“And it means, at its base, sovereignty, greater resilience, an ability to withstand shocks together,” he said.

Von der Leyen has been pushing for Canada to become the EU’s first “associate member,” despite resistance from some European governments to the designation. Carney has also used the term, although he has emphasized that the partnership’s benefits matter more than its name.

“We want Canada as close a partner as possible to the European Union,” von der Leyen said. “We know that Canada cannot be a full-fledged member of the European Union, but to have a relationship as close as possible, this is our goal.”

Carney said the proposed alliance would make life more affordable, give consumers more choices, and create opportunities for businesses, young people, and researchers.

Von der Leyen said Canada and the EU were also creating a common space with other trusted partners to confront challenges involving critical minerals, resilient supply chains, artificial intelligence, and space.

The leaders are expected to announce a digital trade agreement Friday, along with initiatives to expand scientific research partnerships and university student exchanges. Von der Leyen has also said the two sides would finalize an agreement giving Canadian companies and researchers access to Europe’s network of supercomputers to develop artificial intelligence.

Trump has suggested Canada’s proposed EU arrangement could be a “hostile act” and threatened tariffs on Europe if he considered it unfriendly. The United States accounted for 71.7% of Canada’s merchandise exports in 2025, underscoring the challenge Carney faces as he seeks to double exports to other countries over the next decade.

—By Rob Gillies, Associated Press

The company behind ‘PUBG’ wants to bring video game AI into the real world

2 weeks 4 days ago

The robotics industry wants humanoids to become part of everyday life, taking on work in homes and offices. Billions of dollars are flowing toward machines capable of increasingly complex physical tasks. The humanoid robot maker Figure raised more than $1 billion last year at a $39 billion valuation, while the robotics software startup Skild AI secured $1.4 billion in January to develop artificial intelligence that can control different kinds of robots.

But working alongside people requires understanding incomplete instructions and adapting when someone changes their mind midway through a task. A robot that needs constant explanations could leave its owner supervising the very work they hoped to hand over.

Krafton believes its video game experience could help solve that problem. The South Korean developer behind PUBG: Battlegrounds has spent nearly two decades creating virtual worlds where software responds to unpredictable player decisions. Its Palo Alto subsidiary, Ludo Robotics, is now applying that expertise to AI that helps robots understand human requests and decide how to act.

CH Kim, CEO of both companies, sees Krafton’s experience developing AI teammates as a starting point. Developed with the chipmaker Nvidia, PUBG Ally is what the companies describe as “the world’s first voice-enabled coplayable character (CPC),” an AI teammate that goes beyond traditional nonplayable characters (NPCs) by cooperating with players, adapting to gameplay and remembering previous interactions.

Ludo Robotics hopes to bring those capabilities into the physical world.

“We’ve been developing AI in the form of nonplayer characters, or NPCs, for our video games even before the recent wave of interest in the [AI] sector. With the latest large language models and AI technologies, we’ve been able to give our in-game NPCs greater intelligence,” Kim tells Fast Company. “We reached a point where we started thinking about what would happen if we brought these AI characters out of virtual worlds and into the real world. That’s essentially physical AI, and that’s how we came to work on Ludi.”

Ludo Robotics is set to present Ludi 0.2, its latest experimental robotic intelligence system, at the Bay Area Robotics Symposium at the University of California, Berkeley. The new version aims to improve conversational memory and gesture recognition, though Kim acknowledges that Ludi remains a research demonstration, not yet reliable enough for everyday use.

What PUBG Can Teach a Humanoid Robot

During a limited beta in PUBG: Battlegrounds’ arcade mode, players could bring along Ella, a PUBG Ally companion that responded to voice commands. In a recent blog, Krafton’s AI chief described players joking with her and treating some mistakes as entertainment. Yet her usefulness as a teammate was less convincing. The gaming news outlet Inven Global, citing Krafton’s PUBG Ally report, put combat performance at 2.58 out of 5. Half of respondents viewed the character as a tool, while just 18.5% considered her a teammate.

CH Kim [Photo: Krafton]

“The foundational technology we’ve applied to PUBG Ally is very similar to the technology we’re applying to Ludi,” Kim explained. “We’re building on those capabilities as we move forward, and our longer-term goal is to expand this work into developing our own proprietary foundation model.”

Krafton believes its gaming expertise offers a competitive advantage in humanoid robotics. Game engines simulate how objects move and interact, providing virtual environments that can generate training data. Kim says Ludo is already using synthetic data from these worlds and plans to incorporate more real-world information as its research advances.

Ludi 0.1, introduced in August, uses an architecture combining several existing AI models. A vision-language model from Alibaba’s open-source Qwen family analyzes camera images and transcribed speech to decide what the robot should do next. Nvidia’s GR00T robotics foundation model, fine-tuned by Ludo, controls physical tasks such as picking up objects. The system runs on Unitree’s G1, a roughly $16,000 humanoid robot, with Ludo’s software coordinating the components.

[Image: Krafton]

The company trained its reasoning model on 14,432 computer-generated interaction sequences. The robot learned physical manipulation separately from recordings of people remotely controlling it in a mock household. Although Ludo’s technical report identifies PUBG Ally as a design precedent, it does not list player conversations among its training data.

Kim sees initial opportunities in homes, offices, and factories, where robots must work alongside people rather than simply perform repetitive tasks. “Everyone seems to be working on building a foundation model, but I think that’s an enormous challenge,” he says. “If you’re actually going to develop a foundation model, it needs to be highly generalized. It has to be capable of handling all kinds of different situations, and I think achieving that level of generalization is going to take time.”

For now, Ludo intends to develop the intelligence rather than manufacture robots, working with hardware makers to integrate its software into their machines.

Krafton’s Investment Strategy for Ludo Robotics

Krafton reported second quarter revenue of 1.29 trillion won, about $890 million, up 95% from a year earlier. Its broader AI investments include $70 million for a computing cluster and a reported $33 million investment in HyperAccel, a South Korean startup developing AI chips. Ludo’s team includes Rob Nowak, chief scientist, and Yea-Seul Kim, human-robot interaction lead, both of whom bring academic experience from the University of Wisconsin-Madison.

Kim says Ludo is focusing on research this year before identifying a suitable environment for deployment in the real world in 2027. “We need to find an environment where those research results can be applied and where we can potentially create a data flywheel or a training flywheel,” he says. “That would allow us to continue developing our capabilities as we move toward building a foundation model.”

Krafton may eventually seek external funding. “It’s difficult to say what kind of commercialization can realistically happen with humanoid robots within a single year,” Kim says. “Krafton is investing heavily in AI overall, but robotics is a very long-term project. We may need to raise additional funds from outside the company to properly pursue this area.”

Ludo’s Push for Unified Robotic Intelligence

Kangwook Lee, Ludo’s chief technology officer and Krafton’s chief AI officer, says developing AI characters for video games was easier. “In a game environment, the appropriate programs and algorithms could make characters move and use their bodies naturally, allowing us to focus on improving their social intelligence rather than solving the underlying challenges of physical capability,” he explains.

Real-world robotics is more complicated because physical movement and social interaction must work together. Lee points to how people naturally gesture while speaking. “It’s almost impossible to think about physical movement and social interaction as two completely separate things,” he says.

Both Ludi 0.1 and 0.2 rely on multiple foundation models connected through an agent harness. One interprets the situation and decides what to do, while separate systems execute physical actions. That arrangement becomes less efficient when circumstances change midway through a task.

“The problem is that the other models have no visibility into the reasoning behind a decision. Instead, each receives a final instruction to execute without understanding the broader context, which becomes a significant limitation when circumstances change,” Lee explains. “The system must then return to the reasoning model to reassess the situation and determine the next course of action. This constant back-and-forth disrupts the connection between reasoning and physical execution, creating significant inefficiencies.”

Although Ludi 0.2 improves perception and social understanding, it retains that architecture. Ludo plans a more fundamental redesign with Ludi 1.0, scheduled for 2027. The goal is a unified foundation model that retains context and adapts to changing circumstances without repeatedly transferring control between systems.

“If something changes in the environment, the robot should be able to build on its previous reasoning and adjust its actions much faster,” Lee says. “Instead of passing the problem back to another model to reconstruct the context and determine what to do next, the system could use that shared understanding to respond directly to the new situation.”

That integration also presents computing challenges. Ludi 0.1 processed AI tasks locally on a workstation beside the robot rather than relying on the cloud. Ludo selected a smaller, 4-billion-parameter model to reduce delays. Kim sees hardware limitations as a major obstacle. “A foundation model needs to have generalized capabilities, but one of the biggest challenges is developing those capabilities within the specifications available on the robot itself,” he says.

Ludi’s AI Benchmarks Reveal Performance Gaps

Ludo measures progress through its Interaction Core28 benchmark, which tests AI across 28 simulated household scenarios, including delivering objects, interpreting ambiguous requests and responding when someone speaks while the robot is moving. Before additional training, the smaller Qwen model powering Ludi 0.1 completed 10 scenarios, improving to 17 after fine-tuning. An OpenAI frontier model, using a carefully optimized prompt, completed 26, revealing a substantial performance gap.

These results come entirely from simulations, making comparisons with physical robots difficult. Figure, for instance, reported a 56% task-completion rate across 420 trials in a home environment, though its evaluation did not include interrupted chores or situations involving people occupying the same space. Meanwhile, the consumer robotics company 1X markets its $20,000 NEO household humanoid with remote human assistance when needed.

For Ludi 0.2, Lee says the team retrained the visual reasoning model responsible for interpreting camera images, identifying people and determining how the robot should behave. “We needed to evaluate how robots use their bodies when interacting with people, rather than focusing solely on whether they can perform a particular action,” he explained. “It’s especially important to understand whether the robot behaves appropriately when people are nearby and how its physical movements affect those interactions.”

Krafton describes additional evaluations involving personal preferences, relationships between people in a room, and gestures made while carrying objects. However, the company has not published Ludi 0.2’s Interaction Core28 results or disclosed enough about its new evaluations for independent researchers to reproduce them.

Lee considers recognizing unclear instructions and asking for clarification, rather than making assumptions, an important measure of intelligence. “Our benchmark includes several scenarios in which people deliberately give the robot ambiguous instructions in a simulated environment. In those situations, the expected response is to ask clarifying questions rather than immediately attempting to carry out the task,” he says. “Once the person provides the missing information, the robot should have enough context to proceed.”

In a video game, players can laugh off an AI teammate’s mistakes and continue the match. In the physical world, misunderstandings could force people to repeat instructions or intervene in tasks they expected a robot to handle. That is the real test of Krafton’s gaming expertise: whether its experience making virtual characters responsive to players can translate into reliable physical assistance.

Hackers hijacked a Claude Code extension, exposing yet another new AI security risk

2 weeks 4 days ago

The world can feel particularly unsafe these days, with AI agents running rampant within private networks and billions of dollars disappearing from crypto wallets in less time than it takes to watch a sitcom episode.

But while the world was watching Binance’s massive crypto heist, another potentially consequential security breach was unfolding largely under the radar.

For roughly 11 hours, a malicious version of Anthropic’s Claude Code extension was available on Cursor’s VS Code extension marketplace. The extension—which allows developers to use Anthropic’s AI coding assistant within Cursor to write, edit, and debug software—had been tampered with to collect information about users’ computers and attempt to install a persistent backdoor. That backdoor used Secure Shell (SSH), a tool for remotely accessing computers, and could potentially have allowed hackers to access victims’ machines even after the extension was removed.

Cursor disclosed the breach in a blog post, saying it removed the compromised extension as soon as it discovered the problem. The company also emphasized that the mechanism intended to give hackers remote access was unlikely to have worked in many cases because of flaws in its code. Anthropic, meanwhile, said in a statement that the incident “only impacts Cursor users” and was unrelated to Anthropic’s own systems.

According to Cursor’s security notice, the malicious extension was designed to leave behind scripts that would attempt to reconnect to the hackers’ server every minute, even after the extension had been removed. Cursor says it has since worked with the server’s hosting provider to take the server offline and to temporarily shut down and rebuild its extension marketplace. The company found no evidence that other extensions were compromised. Still, it is urging anyone who installed the affected version on October 29 to check their computers for signs of infection, remove any unauthorized access, and consider changing potentially compromised credentials.

The hackers’ identities remain unknown, but security experts say the incident illustrates an increasingly troubling risk as AI coding tools become embedded in developers’ workflows.

“The scenario of Claude sending out poisoned agents everywhere makes me even more worried than I already was,” says Eerke Boiten, a professor of cybersecurity at De Montfort University in England.

Gwen van der Velden, a professor of education at the University of Warwick who’s working on a program to help rebuild Ukraine, says the choice of the Claude Code extension is particularly “interesting” given the potential for such tools to be exploited in hybrid warfare.

“We know from the Russian side, it is of great interest to them,” she says. “If you were to go for hybrid warfare, where you use this kind of technology, then actually getting 5,000 bits of malware reproducing itself through AI agents is a powerful thing.”

The broader concern isn’t simply that a malicious extension could compromise individual computers. It’s that the growing ecosystem of AI coding assistants and their associated tools is creating new pathways into sensitive systems.

“This is a classic example of how everyone has been focussing on LLMs attacking other systems, yet forgetting about the fact that LLMs and their associated tools are increasing the attack surface across some very sensitive IT estates,” says Alan Woodward, a professor of cybersecurity at the University of Surrey.

And even a relatively limited breach could carry outsized consequences when the targets are software developers with access to sensitive code and infrastructure.

“While 5,000 users might sound relatively small, these are developers who have access to highly sensitive systems and source code, so the potential impact will be far greater than the numbers suggest,” says Jake Moore, a threat adviser with the global cybersecurity firm ESET. “Software extensions have eased their way into many people’s workflows, but the concern is how easily trusted software extensions can become a vehicle for malicious code.”

Ahead of midterms, Republican attorneys general back South Dakota law that could criminalize abortion-related speech

2 weeks 4 days ago

Several states have abortion rights on the ballot again this midterm season—or candidates with strong stances on the issue. 

Among them are six state attorneys general who are seeking reelection and who recently voiced their support for an embattled South Dakota abortion law that critics argue violates free speech.

On October 19, Attorneys General Brenna Bird of Iowa, Tim Griffin of Arkansas, Mike Hilgers of Nebraska, Kris Kobach of Kansas, James Uthmeier of Florida, and Drew Wrigley of North Dakota joined 12 other attorneys general in filing an amicus curiae brief, a document in which someone with relevant expertise or an interest in a given lawsuit’s outcome makes a legal argument. 

All 18 of the attorneys general are Republican.

Law would restrict abortion-related speech

The brief supports a law passed earlier this year that prohibits “the dispensing, distribution, sale, or advertisement of certain articles or things for purposes of an unlawful abortion” in South Dakota and sets criminal and civil penalties.

South Dakota has had a total abortion ban in effect since 2022, with an exception only when “necessary to preserve the life of the pregnant female.”

The plaintiffs in Mayday Health v. Rhoden, which was filed in May, allege that the law is unconstitutional and violates the First Amendment. A preliminary injunction was issued, and later upheld, preventing the state from enforcing the law while the matter plays out in court.

Legal fight between state and nonprofit

The lawsuit follows a previous legal dispute between South Dakota and nonprofit Mayday Health, in which the latter was accused of violating state laws by placing signage at gas stations that said “PREGNANT? DON’T WANT TO BE?” and directing individuals to its website.

In a legal filing, the state argued that the ads contain deceptive information, and “immediate and irreparable injury, loss, or damage will result to South Dakotan consumers who are misled by the deceptive advertisements.”

Mayday Health’s website says it shares information about reproductive healthcare, including “pills, birth control, and gender-affirming care in any state,” but does not provide abortion pills or medical advice.

The abortion medication mifepristone has long been deemed safe and effective by the Food and Drug Administration (FDA) and major medical organizations. 

Politics writer Jessica Valenti characterized the law as “a chilling attack on the First Amendment and our ability to talk freely about abortion” on her Abortion, Every Day Substack.

The American Civil Liberties Union (ACLU) also opposed the law, noting that “there’s no way for the state to enforce this bill” without using invasive surveillance.

Brief argues for states’ rights

The amicus brief argues that states have sovereignty to regulate within their borders, which “is currently under attack by out-of-state actors like Mayday Health,” and that such actors are advertising illegal services. 

Some states have banned the mailing of abortion pills, but the U.S. Supreme Court ruled that the practice can continue for the time being.

Abortion access is on the ballot

In addition to the attorney general elections, voters in Idaho, Nevada, and Virginia will decide on abortion rights ballot measures next week.

Meanwhile, voters in Missouri—whose attorney general, Catherine Hanaway, signed the amicus brief—will vote on Amendment 3, which would repeal the right to abortion with limited exceptions and prohibit gender-affirming care for minors.

Sixty percent of Americans continue to say that abortion should be legal in all or most cases, according to a recent Pew Research poll; another Pew study found that Arkansas was the only state in which more than half of respondents were against abortion rights.

Patrick Stewart was thinking about AI ethics years before ChatGPT launched

2 weeks 5 days ago

Yesterday, the world lost one of its great thespians. Sir Patrick Stewart, known on stage for his riveting portrayals of Shakespeare’s characters and, more broadly, for his film portrayals of the X-Men’s Professor X, passed away after a short battle with Alzheimer’s disease and Lewy body dementia.

But of course, the role Stewart is best known for is playing Captain Jean-Luc Picard on the television series Star Trek: The Next Generation, which ran for seven seasons from 1987 to 1994.

Unlike other sci-fi of the time, particularly the more popular Star Wars, Star Trek: The Next Generation often explored the philosophical and social aspects of technology, with many of its most memorable episodes addressing the moral implications of scientific progress.

One particular episode, “The Measure of a Man,” from 1989, saw Stewart’s Picard make an impassioned plea and legal argument that his friend, the artificially intelligent android Data, played by Brent Spiner, should not be considered Starfleet property but a sentient being with the associated rights all individual sentient beings have.

This, of course, was decades before artificial intelligence went from being science fiction to science fact in our 21st-century world. But it’s not the only time Stewart contemplated the technology and its moral implications.

In 2014, Stewart gave an interview to Smithsonian magazine in which he, perhaps the most popular science fiction captain ever, was asked his opinion on the then-burgeoning field of artificial intelligence, and whether any future AI beings should have the same rights as humans.

At the time, Stewart was no longer just an accomplished actor. He was also in his tenth year as the Chancellor of the University of Huddersfield in the United Kingdom, a position he held from 2004 to 2015 (before continuing his relationship with the university as its Emeritus Chancellor).

During the time of Stewart’s chancellorship, the University of Huddersfield was known for its engineering and applied research programs and today runs an Institute for Applied Artificial Intelligence.

In his answer, Stewart reflected on “The Measure of a Man” episode:

“I remember at one script conference, Whoopi Goldberg, during the period when she was with us on the show, saying what we ought to make clear in this­—that we’re actually talking about slavery,” Stewart recalled. “And it was a great point to make. If we create independent life but keep it under our control, what is that? It can be said to be a form of slavery.”

But he continued: “I hope that the moral questions will be addressed as enthusiastically as technical questions will be addressed when it comes to artificial intelligence.”

A true Renaissance thespian

For anyone who is a fan of Stewart’s work, it’s little surprise the accomplished thespian, well-versed in the humanities, was always keen to consider the moral questions surrounding technology, particularly, in this instance, artificial intelligence.

Of course, despite some AI engineers reportedly believing they are creating a god, the AI that exists today still seems a far way off from being truly sentient, and thus deserving of rights. And it is nothing like the benevolent android Data. Instead, today’s moral questions about artificial intelligence revolve around its impact on jobs, social cohesion, mental health, privacy, and its use in warfare.

Yet some of those questions were also on Stewart’s mind more than a dozen years ago, and more than half a decade before chatbots like ChatGPT launched.

“I’m fascinated by this technology. I’m happy to say that my university is in the forefront of this kind of research,” Stewart told Smithsonian in 2014, before adding, “And I’m excited by that but . . . in the same way that we’ve had to start asking questions about personal liberty, and what is the nature of privacy as social networking and aspects of reality television and exposures become more and more successful and money-making, we have now to look at what is the nature of privacy and what can one expect as a right.”

If only we had someone like Captain Picard to help us navigate the artificial-intelligence uncertainty ahead.

Polymarket trading paused over a security issue involving third-party U.K. provider

2 weeks 5 days ago

Polymarket temporarily halted trading on its international platform Thursday because of a security incident that compromised a third-party provider, the company said.

Trading was paused for about an hour and resumed Thursday afternoon. The company has promised to fully reimburse any affected traders.

The prediction market platform said it halted trading “out of an abundance of caution” after identifying and fixing the issue affecting its international platform.

“We’ll make this right and are fully reimbursing all affected traders,” the company posted on X once trading had resumed.

Polymarket said it referred the incident to a law enforcement agency in the United Kingdom.

Reports have linked the source of the incident to DebugBear, a U.K.-based company offering website performance monitoring tools.

DebugBear’s own website was unavailable Thursday afternoon and the company posted on X that it was “impacted by an ongoing security incident.” It also advised clients who were using one of its products to remove the embed code from their websites. The company did not immediately respond to a request for comment.

It was not immediately clear to what extent traders using the popular prediction market platform were impacted, or the breadth of the impact. The company said anyone who was impacted “will be made whole.”

—By Kaitlyn Huamani, AP Technology Writer

Starbucks store closures update: Coffee giant reveals an optimistic plot twist after recent mass shutterings

2 weeks 5 days ago

Starbucks Coffee Company on Thursday provided an update on store closures, as well as fresh plans for future store growth, as it reported earnings for its Q4 and full fiscal year 2026. Here’s what to know:

Starbucks earnings snapshot

The Seattle-based coffee company reported adjusted earnings of 80 cents per share, a 54% increase over the prior year, beating analyst expectations. 

Consolidated net revenues decreased about 4% to $9.1 billion. Starbucks says the decline was largely driven by the conversion of its China retail operations to a licensed joint venture.

For the full fiscal year 2026, Starbucks reported net revenue of $37.9 billion, an increase of 2%.

Shares of Starbucks Corporation (Nasdaq: SBUX) were down 2% in early trading on Friday along with much of the market. However, the stock is up more than 23% year-to-date as investors appear to have warmed up to CEO Brian Niccol’s turnaround plans.

How many stores has Starbucks closed?

In Q4, Starbucks closed around 250 North American stores, 212 of which were in the United States, as part of its Back to Starbucks strategy.

The company has been closing underperforming stores to focus its attention on more profitable locations. Niccol, who formerly led a turnaround at Chipotle Mexican Grill, announced the Back to Starbucks plan on his second day on the job.

Starbucks ended the quarter with 16,781 U.S. coffeehouses, about 40% of its global portfolio of 41,607 locations.

On the earnings call, Starbucks executives said they expect to add around 90 to 100 net new North American stores in fiscal 2027 as Starbucks resets its “development pipeline.” They expect growth to ramp up from there.

“As you look out into fiscal year ‘28, we want to get back into building new units north of 300 or so in the U.S.,” Niccol said on the call. “We believe that is what gets us going towards that opportunity of 5,000 or more in North America.”

From turnaround plan to growth plan

Two years into the strategy, Niccol said it has entered a new phase. 

“Back to Starbucks started as a turnaround plan, and now Back to Starbucks is our growth plan,” Niccol said. “Our plan is working. More customers are visiting us because the Starbucks experience is better. Our coffeehouses are coffeehouses again.”

The next great founder may be hiding in healthcare

2 weeks 5 days ago

Academic medical centers and universities produce groundbreaking research every day. An autism researcher who builds an AI game to clinically improve behavioral problems. A basic scientist who discovers how the environment of tumors impacts treatment. An electrical engineer who nano-sized a method for magnets to image the body. An MBA who teamed up with two MDs to create a novel approach to solving complex pediatric cardiac cases. And all are at risk of remaining stuck in the ivory tower.

Venture capital is great at finding founders who tell a compelling story with confidence, but the best founders may not innately exhibit those traits. This is particularly true in healthcare, where physicians, researchers, PhDs, and faculty members often have the deep technical understanding and firsthand experience needed to build the next great company.

These scientists and clinicians are knee-deep in the problems they’re trying to solve. They’ve seen where existing approaches fall short, often for years, and have developed ideas for how to do things differently. But getting those ideas into the market requires a different set of skills. They need to learn how to test the market hypothesis, not just the scientific hypothesis.

Today, there are researchers and scientists developing promising healthcare innovations who aren’t thinking about building a company. And if they are, many show up to pitch venture capitalists with presentations better suited for an academic keynote or scientific meeting. Slides dense in technical detail, with little market analysis or validation, and without much of a story. It’s not surprising that many VCs pass. But how many promising ideas are we missing because the people behind them haven’t learned how to pitch?

Our experience

We’ve seen this firsthand, over and over again. And it’s made us rethink what it actually means to invest early.

Funds have competed for years to get into the best deals as early as possible. But in healthcare, getting in early may mean investing before there’s even a company. Before there’s a pitch deck, a business model, or a founder who has fully embraced the idea of becoming an entrepreneur.

This is the idea behind inception investing. Rather than starting with a company and evaluating whether it’s worth backing, the starting point is a person who is particularly well suited to solving a problem. That might be a clinician or researcher with decades of experience and a scientific approach that hasn’t yet been commercialized. Or it might be someone whose personal experience with the healthcare system has led them to become deeply focused on a problem that others have overlooked.

In either case, there may be no company yet. The market might not be defined. The founder might not know how to tell a compelling story. But those aren’t necessarily reasons to walk away.

At Treehub, we’ve found that the qualities we’re looking for in these early conversations are often different from what a traditional investor might prioritize. We’re less concerned with whether someone can deliver a polished pitch and more interested in whether they understand a problem well enough to build something that hasn’t existed before.

More than just science

Of course, scientific expertise alone doesn’t make a successful founder. There still needs to be a market for the solution, a viable business model, and a path to getting it into patients’ hands. And not every researcher wants to build a company. But those are questions worth exploring with someone who has spent years developing a potentially important solution, rather than reasons to dismiss them at the outset.

A scientist can learn how to evaluate a market. A physician can find a commercial partner. A researcher can learn how to tell a story that resonates with investors. The experience that led them to the idea in the first place is much harder to replicate.

If we want more of the research coming out of universities and academic medical centers to reach patients, we need to get better at identifying the people behind it and helping them figure out what comes next.

The next great healthcare company may not exist yet. And the person who will build it may still be thinking of themselves as a scientist, physician, or researcher rather than a founder. That’s exactly why we should be looking for them now.

Amazon posts 3rd quarter results that fall short of Wall Street’s expectations

2 weeks 5 days ago

Amazon offered a cautious sales outlook for the current fourth quarter that encompasses the critical holiday shopping period after delivering fiscal third-quarter results that came in below Wall Street analysts’ expectations.

The Seattle-based tech giant, however, posted strong sales growth in its prominent cloud computing unit. Sales in its so-called Amazon Web Services rose 39.3% to $46 billion during the latest quarter ended Sept. 30, higher than the 37% clip in the previous quarter. It marked the fastest growth rate in nearly five years, the company said.

Still, shares were down 7% in after-market trading when the results were released as investors focused on its core operations and worried about Amazon’s burning through cash as it invests heavily in artificial intelligence.

Investors have been closely watching Amazon’s quarterly earnings to see if the company’s big spending on artificial intelligence is starting to pay off.

Amazon’s CEO Andy Jassy told investors on the call Thursday that he still expects to spend $220 billion in capital this year, largely to build AI data centers. He believes that sales of its AWS unit will increase 40% each of the next two years, and the unit will ultimately become a $1 trillion-plus business in the next five to 10 years.

“The opportunity remains extraordinary,” Jassy told investors.

Jassy said Amazon is installing an AI chip about every 14 seconds and “almost as fast as we installed capacity.”

“Customers are gobbling it up,” he said.

But Amazon reported a free cash flow loss of $42.4 billion for the trailing 12 months as it makes big investments in AI.

In July, Jassy disclosed to investors that Amazon was increasing this year’s capital spending on technology by an additional 10% to $220 billion. That number includes outlays on robots, semiconductors and satellites. That’s up from the $200 billion investment plan announced in February and well above the $128 billion in capital spending for all of last year.

Worries are weighing on the entire AI industry following a big push to build data centers and fill them with AI chips.

The latest earnings report from Alphabet late Wednesday raised alarm bells when it reported weaker growth for the summer than analysts expected. The company behind Google and YouTube also said it burned through more cash during the latest quarter, after its purchases of property and equipment topped $51 billion. Its stock plummeted Thursday.

Amazon’s earnings report also comes as big tech giants promise to keep fewer secrets or invest more in communities willing to allow data centers as they face backlash from politicians and residents over their expansion.

Amazon said earlier this month that it will invest more than $1 billion over the next five years into communities where it operates data centers. The money will support education, job training, water and energy preservation projects and other local priorities, Amazon said.

Meanwhile, Amazon has been speeding up its order delivery times through a combination of robotics, AI technology and more efficient warehousing. In fact, speedier delivery helped Amazon dethrone Walmart in February from its status as the nation’s largest company by revenue.

Amazon executives said increased speed is fueling its grocery business, including perishables.

Amazon reported net income of $20.69 billion, or $1.89 per share, in the three months. That compares with $21.19 billion, or $1.95 per share, in the year-ago period.

Net sales rose 12% to $201.94 billion, compared with $180.17 billion from the same year-ago period. Amazon said sales were hurt by the shift in the calendar for the company’s big sales event Prime Day, which fell in the second quarter this year for most large countries, including the U.S.

In 2025, Prime Day occurred entirely during the fiscal third quarter.

Excluding this shift, third-quarter year-over-year sales growth would have been approximately 400 basis points, or 4%, higher, Amazon said.

Analysts were expecting $1.97 per share on sales of $202.2 billion, according to FactSet.

For the fourth quarter, Amazon said it expects net sales to fall between $240 billion and $247 billion, a 12% to 16% increase compared with the year-ago period. Analysts were expecting $244.16 billion.

This story has been updated to correct that Amazon reported a free cash flow loss of $42.4 billion, not that cash flow fell to $42.4 billion.

—Anne D’Innocenzio, AP Retail Writer

As AI slop floods feeds, luxury brands are going all in on handmade ads

2 weeks 5 days ago

In the age of AI, handmade work is becoming radical (and chic), and consumers are taking notice.

Prada’s new holiday campaign, Drawing Prada, celebrates craft and process, earning praise on social media for its use of handmade animation.

The campaign transports the actors Chen Haoyu, Luther Ford, Kelvin Harrison Jr., Maya Hawke, and Sadie Sink into a fairy tale–like animated forest, complete with a cozy cabin and charming woodland creatures.

In one scene, Harrison Jr. walks through a snowy path only to be hit by a snowball thrown by an animated cat-like figure who quickly turns around and whistles, trying to have nothing to do with it.

In another, Sink kneels beside a fireplace and peers into a mouse hole, finding a trio of mice atop a Prada bag and a shoe, all while munching on cheese.

But it is not just the whimsy that is capturing the attention of consumers. The animations are hand drawn by animator John Pomeroy, whose work includes films like Thumbelina and Disney’s Pocahontas and Fantasia 2000. Prada went out of its way to highlight the handmade process.

In the campaign’s images, the actors are superimposed on top of the artist’s sketches and process, as a sort of reminder to viewers that the work is made by humans. It’s a refreshing counter to the AI slop metastasizing across media.

The campaign draws on what luxury fashion has always sold: handcraft and quality. Now the ads are catching up, made as couture as the clothes.

“In an AI world, I’m crying with Prada reviving 2D—go back to making animations like that immediately,” one user said on X. Another added, “The only good thing that AI is going to bring is that big brands, to stand out, will choose the traditional.”

A growing trend

Unsurprisingly, brands tapping into craftsmanship for their marketing efforts has become a trend—particularly for luxury brands.

Earlier this year, Hermès hired an artist to hand-illustrate its website, filling it with oceanic creatures like seahorses, fish, eels, and starfish. The texture of the paper grain peeked through the design, emphasizing its handmade nature, all crafted with ink and paper by French artist Linda Merad.

Sometimes, the return to traditional crafts can take a less analog approach. Ralph Lauren, for instance, created a Wimbledon-inspired campaign with Tom Hiddleston, using CGI to bring its iconic teddy bear mascot to life. The brand’s Instagram caption specified it was made using “live-action footage and CGI technology,” going out of its way to communicate that the video was not AI slop.

The trend is also making its way outside of the fashion space. Take luxury hair care brand Oribe, whose holiday collection is a collaboration with French mosaic artist Mathilde Herrero. The campaign’s videos feature the artist crafting the mosaics that wrap around the boxes for the holiday sets, bringing the product’s signature colors like purple and gold into the design.

These new craftsmanship-centered campaigns are also redefining how brands can understand what meaningful investment in creative looks like, especially as they fight for attention.

“The industry has trained itself to read ROI as a click-through rate,” says Dayna Carney, founder of creative agency Daynas House. “The Prada campaign is already moving on its own, through shares and saves. It’s being talked about in the group chat, which is the only distribution that still holds any weight.”

Carney also says that the effort of world building can translate beyond a single campaign, with possibilities to create follow-up stories with similar motifs.

“The creatures could end up on a scarf or a bag charm. That is the part that lasts,” Carney added. “Most holiday work is forgotten by December 31. This will be remembered and talked about.”

[Illustration: Linda Merad for Hermès] Human touch is the new luxury

Craftsmanship is having its moment just as AI slop floods feeds. But while it can read as rebellion, it is also a reminder that taste, however you define it, is inherently human.

“In order to develop taste, you have to have a rich inner life,” says Liana Satenstein, a former senior fashion writer at Vogue and author of the fashion newsletter Neverworns. “You have to read, you have to go out, you have to have a world of experiences to develop that. As much as you try to do that with AI, you can’t do that. it’s almost like putting lipstick on a pig.”

It makes sense that the fashion world, which banks on tastemaking, is going all in on handmade techniques in its campaigns.

“Luxury fashion has always run on trust and taste, usually in that order,” Carney says. “Embracing handcraft isn’t a deliberate pushback on AI. It’s just that AI is technology without either.”

Lose your job to AI? Here’s what to do if a company wants you back

2 weeks 5 days ago

Companies around the world are increasingly citing artificial intelligence when slashing jobs, fueling anxiety that a lot more workers may lose their livelihoods to machines.

Microsoft, Barclays Bank and supermarket chain Kroger are among the latest entries on a growing list of companies that have announced AI-related workforce reductions this year, including the elimination of jobs deemed redundant or shed to free up money for investing in the technology.

But the job cuts of the AI era, which are affecting tens of thousands of people, occasionally have an epilogue. Some companies, after discovering that AI models, applications and agents don’t perform as well as expected, end up rehiring laid-off workers to fill eliminated positions.

It’s unclear how common such reversals are, but they are happening often enough to acquire a nickname: the “AI boomerang.”

“Two years ago, we said, ‘Would this happen?’ and now we’re saying, ‘Yes, it is happening in real time,'” said Lucy Beaumont, a global senior vice president at SHL, a U.K.-based technology company that develops employment-related tests and tools.

Hiring managers in fields including finance, marketing, law and health reported earlier this year that their companies had eliminated positions primarily because of AI or automation, only to hire again for the same or similar roles, according to recruitment and staffing firm Robert Half.

The most frequently cited reasons for restoring dropped jobs included the technology needing more human oversight than anticipated and its inability to replicate institutional knowledge, the firm said.

“When you need critical thinking, collaboration, creativity, institutional knowledge, those are things that AI just can’t replicate,” Bill Driscoll, a senior district president at Robert Half, said. “And I think that’s where companies are starting to realize, ‘Wow, we really do need people to do these jobs.'”

Klarna, the Swedish provider of buy now, pay later services, began recruiting human customer service representatives again last year. CEO Sebastian Siemiatkowski acknowledged in an interview with Bloomberg News that the company’s reliance on an AI chatbot to cut costs had gone too far.

In June, Ford said it had hired or promoted 350 veteran engineers after AI tools the automaker deployed to do quality control work did not perform at the expected level.

“AI utilization has been an important part of our quality gains and that — along with deep technical expertise of our teams — is a powerful combination,” a Ford spokesperson told The Associated Press.

However, companies don’t always broadcast their U-turns on shifting resources or responsibilities from employees to AI systems, Beaumont said.

“Maybe their shareholders love that they did this right-sizing motion … and then they have to turn around and quietly, with their tail between their legs, admit that they weren’t ready for that or they shouldn’t have done that,” she said.

Management experts offered employment advice for workers as companies make greater use of AI tools and consider — or reconsider — their staffing needs while assessing the technology’s capabilities.

Rehiring suggests the invisible work of humans is not easily replaced

When companies use AI to perform tasks previously done by people, they often miss the invisible work employees do beyond their formal job descriptions, said Betsy Summers, a principal analyst on the Future of Work team at market research and advisory firm Forrester.

That might include sensing emotions and moods, holding impromptu meetings and putting in extra effort when feeling inspired or otherwise motivated, Summers said.

“We don’t know all the work that’s required to do things well within the organization,” she said. “The citizenship behaviors of team cohesion and team building and making sure that you have trust within your team, none of that is accounted for. So when you rip out humans who are doing that work and then expect agents to be able to then deliver exactly what humans have delivered, they are frequently mistaken.”

Another factor that complicates assigning more work to AI-powered tools is that company procedures for tasks such as data processing and bookkeeping are not always well documented, experts said.

“Organizations are not spending enough time doing the up-front work to really truly understand workflows and where humans add value in the process,” said Brooks Holtom, a professor of management at Georgetown University’s McDonough School of Business.

In roles such as customer service, AI agents and chatbots frequently need to refer questions and complaints to a worker who can recognize emotions, assess the severity or complexity of a problem and respond creatively within established policies, Holtom said.

“It’s that combination of creativity but staying in the guardrail that’s really hard for a machine to figure out,” he said.

Figure out what gives you a competitive edge as a human

As companies redirect resources toward AI, current employees and those asked to come back might want to emphasize skills that complement the technology, such as emotional intelligence and sound judgment, Holtom said.

“Really try to be able to articulate, what’s your value proposition as a human?” Holtom said. “Don’t be afraid to lean into the technology as well, and understand how to differentiate yourself from what machines do well. Embrace that, understand that, and you can become uniquely qualified to help with implementation.”

Empathy is a human quality that technology has not fully mastered, Beaumont said.

“Think about an angry customer. They want to be listened to, they want to be heard, they really want a human on the other end of the line,” Beaumont said. “Can an AI replace that right now in a way that feels meaningful to individuals? Perhaps not.”

Know your leverage and use caution if you’re asked back

Many people are understandably reluctant to return to a company that laid them off as part of a reorganization around artificial intelligence, according to Beaumont.

“To come back to a business where you’ve been let go, replaced by AI, to be asked back to that business, your trust is gone, your psychological safety there, your relationship with the employer, and that’s very, very difficult to build back,” she said.

Workers offered their old jobs back are asking for more money and to know how AI might impact future staffing, Beaumont said. Requesting additional training and a clear explanation of your opportunities for advancement could be part of the discussion, she said.

Communicate ways you would contribute value as a returning employee, outlining a new job description “that leverages the technology but also your expertise,” Holtom said. You could end up in a higher-ranking position that pays more than the one you left, he said.

Before accepting an offer, seek a commitment such as a one-year renewable contract that includes a buyout clause if the job is terminated in less than a year, Holtom suggested.

Don’t discount the distinctly human values of motivation and curiosity, which make people efficient and effective, Summers said.

“Within our societies, work serves a broader purpose than just creating value for shareholders or customers or constituents,” she added. “Work serves a human purpose as well. It’s the manifestation of our search for meaning and purpose. We want to contribute to our communities as human beings.”

Share your stories and questions about workplace wellness at cbussewitz@ap.org. Follow AP’s Be Well coverage, focusing on wellness, fitness, diet and mental health at https://apnews.com/hub/be-well

—Cathy Bussewitz, Business Writer

7 last-minute Halloween costume ideas inspired by pop culture

2 weeks 5 days ago

Halloween lands on a Saturday this year, which means there’s lots of pressure to find the perfect costume.

If you have procrastinated and put off this great responsibility through no fault of your own, there is still hope. Pop culture is a great source of inspiration, and most of these costumes are low effort.

Here are seven ideas to make you the cleverest person in the room without a ton of prep work.

Be Elizabeth Holmes and Nathan Fielder

Elizabeth Holmes, the disgraced founder of the now-defunct health technology company Theranos, is back in the culture zeitgeist thanks to a new documentary by Nathan Fielder.

Since everyone is talking about You Can See Everything, it’s the perfect costume inspiration.

Dressing up like Holmes is easy since she is known for revering Steve Jobs and sporting a minimalist tech-bro look. All you need is a black turtleneck, black dress pants, a messy ponytail, and red lipstick.

To rock Nathan’s essence on Halloween, simply layer a blue crewneck shirt over a white t-shirt. Be sure to accessorize with black over-the-ear headphones.

Be Kelly Clarkson and the magic shop bird

Kelly Clarkson, the first-ever American Idol, recently had a beef with the Brooklyn-based store Art of Play. She was denied the opportunity to buy a mechanical bird and did not appreciate the way she was treated.

She took to X to publicly lament the situation and the internet took notice.

Since video footage of the exchange was released, Clarkson fans can recreate her look by wearing a yellow t-shirt or sweatshirt, baseball cap, and loosely fitting jeans.

You can even make this a couples costume by having your partner dress as a yellow bird, complete with a $15,000 price tag. A onesie from Amazon or a quick trip to Spirit Halloween should do the trick for this second costume.

Be someone from ‘Off Campus’

Off-Campus, the hit college romance show based on the book series of the same name by Elle Kennedy, premiered on Amazon Prime Video in May of this year. There is sure to be lots of Briar University representation this Halloween.

The simplest and most lowkey way to go is find a Briar U hockey jersey or t-shirt and pair it with leggings or jeans.

The next level up is to recreate Hannah and Garrett’s look from the dynamic duo’s party.

For Hannah, the bunny, all you need is a white dress, bunny ears, and white knee-high boots. For Garrett, the magician, all you need is a white button-down shirt, black dress pants, and a cape.

Expert level is dressing up as Ali and Dean, who met at the party while dressed as J.Lo and Maverick from Top Gun.

Be ‘Freaky’ Nikki

It was a great year for scary movie fans. Curry Barker’s Obsession grossed more than $263 million at the domestic box office alone, and only cost $750,000 to make, according to a New Yorker interview.

To represent this horror film on Halloween, dress up as Nikki from the party scene or ending. The party scene Nikki requires a red-and-black animal print halter top, black shorts, and black boots.

A fake or prop broken beer bottle and fake blood can also be used to enhance the look.

The final scene costume requires a purple floral dress, fake blood, and a brick. NBCUniversal is even selling a set that comes with your own one wish willow.

Just be careful what you ask for should you use the wish. 

Be an Etsy Witch

A witch is a classic Halloween costume. Put a witty spin on it by printing out an Etsy logo and wearing it as an accessory.

And yes, it is true some spiritual practitioners try to sell their spells on the popular online marketplace even though it’s technically not allowed.

Be a ‘Widow’s Bay’ character

Widow’s Bay, the critically acclaimed Apple TV series, is what happens if a Stephen King novel and a Parks and Recreation episode had a baby.

It’s a funny horror series about an isolated New England island town that’s definitely not cursed.

Patricia Moyer, played by Kate O’Flynn, is the mayor’s quirky assistant. She also happened to survive a serial killer attack in high school, which no one really believes.

When the Boogeyman returns, she has to fight for her life in her comfy clothes. All you need to honor her struggle is a reddish orange t-shirt with stars and moons on it, biker shorts, and uggs.

Bonus points if you can make a book cover with the title Your Turn: Out with the Old, and In with the New by Lucy Fors and find a taser.

Be a Backstreet Boy

Oh my god, they’re back again.

Since 2025, the Backstreet Boys have been performing at Sphere in Las Vegas, celebrating the 25th anniversary of their 1999 album “Millennium.”

The cover picture of that album has all five boy-band members wearing all white. A super simple costume idea is to rock all white and call yourself a Backstreet Boy. If you really want to get into character, ask your friends if you are sexual all night long.

Whatever you plan on dressing up as, we hope you have a blast.

Binance hack: Bitcoin and other tokens fall as the crypto world is rocked by a historic attack

2 weeks 5 days ago

The cryptocurrency world is rattled on Friday after hackers stole $6.3 billion from Binance.

The world’s biggest cryptocurrency exchange shared the news on X, explaining that the incident appeared to stem from an engineer’s laptop. The compromised device “allowed an attacker to gain access to a service container and submit unauthorized withdrawal transactions.”

North Korea link?

Blockchain intelligence firm TRM Labs has tied TraderTraitor, “North Korea’s elite hacking unit,” to the attack. TRM Labs calls it the “largest crypto theft on record,” surpassing last year’s $1.5 billion Bybit heist. 

In response to the attack, Binance has suspended withdrawals and deposits temporarily. It’s also working with its blockchain partners to freeze any impacted assets. 

“Our teams are working around the clock to resolve this issue safely and thoroughly,” Binance said in its post on X. “We are working closely with our law enforcement partners to confirm the identity of these criminals and bring them to justice.”

Binance claims the company will cover all losses using its Secure Asset Fund for Users (SAFU). However, an expert expressed doubts to CNBC about a full reimbursement. The fund holds about $1 billion in Bitcoin. 

In 2025, Binance reported $34 trillion in trading volume and over 300 million users. It also had $162.8 billion in assets at the time. 

Cryptocurrencies react to the hack

Cryptocurrency tokens and crypto-adjacent companies saw their values and share prices fall in response to the attack. 

Bitcoin (BTC), the biggest cryptocurrency on the market, was down nearly 4% on Friday morning to $77,594—the lowest it has been in more than a month. 

Shares of Circle Internet Group (NYSE: CRCL), Robinhood Markets (Nasdaq: HOOD), and Strategy Inc. (Nasdaq: MSTR) all fell over 6% in premarket trading. 

Meanwhile, shares of Strive (Nasdaq: ASST), a Bitcoin treasury company, and Bitmine (NYSE: BMNR), an Ethereum treasury company, were down more than 4%. 

Tech stocks down: Intel, Reddit, and Amazon are getting hammered. Has the AI reckoning arrived?

2 weeks 5 days ago

There’s one thing scaring investors today, and it’s not Halloween.

As of this writing, some of the biggest tech stocks are getting hammered in premarket trading, just one day after Google owner Alphabet Inc. saw its own stock price suffer the worst one-day decline in its history. Here’s what you need to know.

Nasdaq lower as tech stocks sink

It’s shaping up to be nothing short of a bloodbath of a week for some stocks.

Yesterday, Alphabet Inc. (Nasdaq: GOOG), parent company of search giant and AI hyperscaler Google, saw its shares drop 12.29% in a single session after reporting its most recent quarterly results.

It was the largest single-day post-earnings share price drop in Google’s history, with more than half a trillion in value wiped out in a single session.

The main driver: a significant increase in Google’s negative cash flow amid its AI buildout. Moreover, revenue for its all-important Search division came in below expectations, fueling fears of disruption by AI chatbots.

Today, Google’s bleeding has stopped—mostly. Currently, GOOG shares are down about eight-tenths of a percent.

But now, other Big Tech stocks are getting hammered.

Intel down 25%

The hardest hit is chipmaker Intel Corporation (Nasdaq: INTC), whose shares are currently down a staggering 25% in premarket trading. Intel’s share price drop comes after it reported its Q3 2026 quarterly results yesterday.

Investors did not love the results. Specifically, as noted by CNBC, Intel issued a weaker-than-expected outlook for its current Q4, driven primarily by a slowdown in PC sales.

The average cost of a personal computer is surging, thanks to skyrocketing memory prices, and it turns out—shocker—that when PC prices go up, consumers want to buy fewer of them.

Intel expects its Q4 revenue to come in between $15 and $16 billion, well below the $17 billion analysts had expected.

Reddit down 20%

Then there is Reddit Inc. (NYSE: RDDT). The social media site reported its Q3 2026 quarterly results yesterday, too, after which its stock price plummeted.

As of the time of this writing, RDDT shares are down nearly 20% on the news that its search referral traffic—mainly from Google—is “down meaningfully” in the current Q4.

As Google embraces AI search results, fewer Google users are either seeing Reddit referral links or clicking on them. This has led to less traffic to Reddit.

In response, Reddit says it will now institute a policy where a person must be logged into Reddit to see its content.

Amazon down 6.2%

Finally, there is Amazon.com, Inc. (Nasdaq: AMZN). The e-commerce giant, too, reported its Q3 results yesterday, and today, its stock price is down significantly—about 6.3% as of the time of this writing.

Amazon’s revenue came in at just under $202 billion, and its EPS was $1.89. Those figures, as CNBC notes, were below what analysts expected.

But the company also posted negative free cash flow of $42.4 billion for the trailing 12 months, largely thanks to its continued capital expenditures on its AI buildouts. (Google’s negative free cash flow announcement a day earlier is what caused the search giant’s stock to plummet this week).

The sinking of Intel, Reddit, and Amazon stocks this morning has helped spur a broader premarket sell-off in tech stocks, sending the tech-heavy Nasdaq futures down 2.4% and S&P 500 futures down 1%. 

An AI reckoning?

There’s one common thread driving the negative news Amazon, Reddit, and Intel posted yesterday: AI. 

For Amazon, its AI buildout is leaving the company with significant negative free cash flow, even as it continues its massive capital expenditures on its hyperscaling ambitions.

For Reddit, Google’s embrace of AI overviews and AI answers has caused a significant decline in search referrals to the social media site.

For Intel, the rising cost of PCs, driven by RAM-hungry data centers that are driving RAM prices sky-high, is causing consumers to abandon or delay plans to buy new computers.

Does this mean that an investor-led AI reckoning is upon us? 

It’s too early to tell. But this week’s stock price falls for Google, Amazon, Intel, and Reddit show that investors are becoming increasingly concerned about the impact that artificial intelligence is having on companies’ bottom lines and businesses. 

Specifically, for hyperscalers like Google and Amazon, investors are starting to want to see a return on investment.

Google and Amazon might have the know-how on how to build great AI products, but investors are increasingly worried no one has cracked the way to make AI profitable.

The hundreds of billions that tech giants are spending on AI buildouts—mainly in the form of data centers—will need to be paid back eventually. And if those buildouts are causing negative free cash flow, that leaves investors worried about how these massive tech companies plan to pay their bills when they come due.

I tried plug-in solar and loved it—but there’s a catch

2 weeks 5 days ago

Getting solar installed on your roof typically involves weeks of planning and red tape and hiring an installation crew to do the work. I tried a very different option: a plug-in solar panel. Like the name implies, installing it essentially just means plugging it into an outdoor wall yourself. Then power starts flowing back into your house.

A single panel, of course, generates a tiny fraction of what you’d get from a full rooftop system. But it’s still a simple way to begin to offset your electric bill, and it’s something that renters can use, not just homeowners. This year, states have started rolling out a wave of new laws designed to clear the way for plug-in solar. I wanted to see how easy it was to use.

[Photo: Bright Saver] How plug-in solar works

I tested a kit from Bright Saver, a nonprofit that advocates for plug-in solar. The 180-watt solar panel is somewhat large, at 4 by 3 feet, but lightweight and easy to move. You can attach it to a wall or fence, or buy a small separate stand so it can sit at an angle on the ground. In the back, the panel plugs into an inverter, a small box that converts the current from the solar panels into electricity that you can use in your house. The inverter plugs into the wall.

Unlike portable solar panels designed for camping, which typically work with a battery, this system is designed to sync with the grid, pushing current into your outlet at the exact frequency and voltage of the grid signal. If grid power goes out, the inverter shuts down in milliseconds, so there’s no risk of the solar panel sending electricity back into the grid when utility workers are fixing an outage on a line.

My first challenge was where to plug it in, since my 180-year-old house didn’t have outdoor outlets. When I started looking around for where wiring could be added—a process that involved opening a trap door and discovering a second, hidden basement—I realized that reaching the sunniest wall would be too difficult. The only easily accessible spot for my electrician to add an outlet was on an eastern wall; ideally, to generate the most energy, panels should face south or southwest. Still, it was a reasonably sunny place to test.

The kit was fast to set up. It took only a few minutes to plug everything together, and to begin testing it, I propped it up temporarily rather than trying to attach it to the wall. After a couple of minutes, the inverter started flashing green. Power was beginning to flow. The app for the kit was clunky, and repeatedly failed to connect to the system. But it finally recognized the inverter, then showed on the screen exactly how much power the solar panel was sending back into my house.

I live in upstate New York, and late October isn’t the sunniest time of year. When I first connected the system, early on a hazy morning with no direct light on the panel, it generated only 8 watts—roughly enough to power my WiFi router. Later in the morning, as it got sunnier, the output jumped up to 81 watts, more than enough to power my fridge or my MacBook Air.

The solar panel might only reach its peak of 180 watts under the right conditions in the summer. But even on a fall day, it was able to provide a meaningful chunk of the power I was using. Bright Saver offers the kit with either one 180-watt solar panel ($285 for members; $499 for nonmembers) or two, for a total of 360 watts. The panels are designed to last for 20 years, and depending on your electricity rates, can pay for themselves in savings within the first couple of years.

With a single small panel, you’re unlikely to produce extra power beyond what your house uses. If you use multiple panels and don’t have a battery, any extra power that you don’t use goes back into the grid. While your utility won’t pay you for it, it could eventually help you in an indirect way: If solar panels like these were used by large numbers of people, it could help shrink the load on the grid at times of peak demand. That reduces the need to turn on expensive “peaker” power plants that raise bills for everyone.

[Photo: Bright Saver] Where you can use plug-in solar now

There’s a catch, in my case: In New York, as in many states, there isn’t yet a law in place to clear a path to using plug-in solar. It’s not illegal to use. But it exists in a gray area with utilities. Some utilities have said that plug-in solar is allowed only if customers go through the process of “interconnection” to the grid; others don’t have any process in place, or have said that they won’t approve the systems and reserve the right to shut off service if they think a system is a safety hazard. My own utility, NYSEG, declined to answer questions about whether it allows plug-in solar now.

The new laws that are rolling out in several states now are designed to make the tech easy to use by eliminating the need for utility approval or permits. The first passed in Utah in 2025. This year, laws have taken effect in Colorado, Connecticut, Maine, Maryland, and Vermont; early in 2027, they’ll take effect in California, New Hampshire, New Jersey, and Virginia. New York’s legislature passed a plug-in solar law in May 2026, though the governor, Kathy Hochul, has yet to sign it; the governor’s office told me only that the governor would consider the bill. (Advocates told me that it’s not clear yet whether she will sign.) A long list of other states introduced bills this year, and will reintroduce them in 2027.

The laws are different in each state, though most say that it’s legal to plug the systems into a standard outlet, without an electrician. A few require a special new certification created by UL Solutions, the certification company. This is where things get complicated. The standard is still evolving, but currently requires plug-in solar panels to use a unique plug and outlet. In the states with this requirement, an electrician will have to verify that the outlet you want to use is on a dedicated circuit, and swap in a new specially-designed outlet. (Bright Saver is taking preorders for this version of the kit, which will soon be available.) The idea is to prevent someone from using a solar panel on a shared outlet; the concern is “breaker masking,” which can prevent a circuit breaker from tripping when wiring overheats. But Bright Saver told me that the risk isn’t a concern with small, 180-watt panels. And it adds enough friction to the process that it may dissuade some people from getting the panels in the first place.

That requirement doesn’t exist in Germany, where the idea of plug-in solar first took off. Germany has a similar climate to where I live—it’s not especially sunny. Still, the country gets nearly 20% of its electricity from solar power. And plug-in solar panels, while a tiny part of that supply, are helpful enough for power bills that around 1.5 million plug-in solar systems are in place there, often attached to apartment balconies. With luck, the same thing may soon be possible in more parts of the U.S.

5 reasons AI evaluations need designers’ unique perspectives

2 weeks 5 days ago

AI evaluations look like the least designerly corner of product development. They are the tests that decide whether an AI feature is ready to ship, and they run on scripts and dashboards full of percentages. No wonder companies hand them to engineers and product managers.

The more I read about evals, the more familiar they look. Behind each one, there is a decision about what “good” means for the person using the product. They all seem to answer questions about user experience. What should the AI do when it isn’t sure? Which mistakes can people live with?

Designers make calls like these every day. However in AI evals, we tend to have a say after someone else has written the criteria, if we are considered at all. As catastrophic as it may sound, this can lead to shipping a product that is technically functional but fails its users.

Here are five reasons why designers should always be in the loop, right next to engineers and product managers.

1. TASTE AND JUDGMENT CATCH WHAT THE NUMBERS MISS

Providing users with the right answers might be the hardest quality score to quantify in evals. Even user ratings can mislead: a Stanford study found that people trusted flattering AI answers more than ones that pushed back.

To decide what the right answer looks like, evaluation teams need a professional with trained taste and judgment. As designers, we develop such attributes by working with users every day. We look at a product through a human lens, so we judge an answer by whether it helps the person who asked.

Teams can train AI to grade other AI outputs, but someone with taste has to teach it what to do, like jargon a new user won’t understand or a cheerful reply to a complaint. Details like these can slip past an engineer, for instance, but a professional in UX/UI design knows to look for them.

2. DECIDE WHAT HAPPENS WHEN THE AI DOESN’T KNOW THE ANSWER

AI often sounds too confident and seems to have all the right answers for the person prompting it, even when it’s wrong. In some cases, chatbots mix good and bad medical advice, and people struggle to tell the difference.

Hallucinations and false information aren’t a good user experience. They make the product feel untrustworthy, pushing users to drop sessions or switch to a competitor.

Whenever the AI doesn’t know the answer, I’d have it say so and point the user in the right direction. That way, the user can rest assured they will still get help, even if the AI can’t give it at that particular moment.

Designers have worked on error messages and dead ends for years. An AI not knowing the answer is just another dead end. We should set this criterion in evals: The AI must clarify what’s missing in its answer and, if possible, where to find it.

3. REAL USERS ARE MESSIER THAN ANY TEST

One skill people working in design have is knowing how to recruit the right set of people to run tests. I’m talking especially about UX researchers here.

The products we built, way before AI was even a thing, required heavy research. We understand how people talk, how they mumble and jump from one topic to another, how unique each answer can be, and how much can depend on mood, time of day, and other external factors.

Some teams running AI evals build test sets with synthetic users. I don’t see that working, and I’m not alone: 88% of researchers in one recent survey raised concerns about the quality and accuracy of AI-generated personas.

4. QUESTION THE ASSUMPTIONS BEHIND THE EVAL

Evals are built on assumptions about what users want, and they’re rarely written down. Most of the time, they come from business goals or from whatever is easiest to measure.

For example, a team building a customer service chatbot to cut support costs may assume people would rather not talk to a human. In that case, any handoff to a live agent counts as a failed test, and the bot that never hands anyone over gets the best score, even when people are getting nowhere.

If you’ve ever been stuck in a loop with a support bot, you know sometimes you just want to talk to a real person.

As designers, we write these assumptions down and check with real users before designing a feature. I’d recommend doing the same with evals.

5. SOME EVAL PROBLEMS ARE DESIGN PROBLEMS

When an eval fails, the first reaction is usually to tweak the prompt or switch models. Sometimes that’s the right call, but other times the problem is the product itself.

Hamel Husain, a machine learning engineer with more than 20 years of experience who now teaches AI evals, wrote this summer: If an AI’s answers are hard for a team to check, they’re often hard for users too.

Imagine a travel assistant that books the wrong date because a user typed “next Friday,” which can mean two different days. A better prompt won’t fix that, but asking the user to confirm the exact date will. Now the answer is easier to check. That’s a design decision.

We catch problems like this in usability tests. So before a team spends another week refining prompts, I’d ask a designer to review the failed cases.

FINAL THOUGHTS

Engineers and product managers will keep running AI evals, as they should. But most of the decisions behind them are about the people using the product. Aren’t designers among the professionals who would know users the best?

My advice to design leaders is to help write the eval criteria before they’re final. After all, deciding which mistakes users can live with is our job, too.

Arin Bhowmick is chief design officer at SAP.

Forget Zillow. Vermonters are fighting the housing crisis with catalogs

2 weeks 5 days ago

One state is preapproving a whole neighborhood’s worth of home designs in advance, and it’s a step that a growing number of local governments across the U.S. are taking to build more homes—and fast.

In Vermont, state officials are preparing a 10-home catalog called 802 Homes that’s named after the state’s sole area code and is expected to be released early next year. Once it’s out, the catalog’s blueprints will be free to use for home builders and developers, and since those designs have already been approved, they can expect to zip through the zoning and approval process.

The U.S. is short 10 million homes, the White House estimated in April, especially starter homes. To fill that gap, local governments have taken to preapproving building plans to get the ball rolling and reduce costs. Vermont is one of the first states to launch its own statewide preapproved housing program, and across the country, more officials are considering it.

“The state needs attainable housing that matches the home types Vermonters want and need, including homes that allow older Vermonters to age in place or downsize, and young families to get their foot in the door,” Jeff Dube, community planning and project manager at the Vermont Department of Housing & Community Development, tells Fast Company. “We also need housing that is located in places that support walkable communities and our historic downtowns and village centers, which tend to be areas that are served by infrastructure and can bring down development costs.”

[Image: Vermont 802 Homes]

About 40 jurisdictions in the U.S. have a preapproved building plan program, according to Pew, the social research center. In South Bend, Indiana, more than 90 units have been built or approved as of June after the city approved seven plans in 2022, while in Kalamazoo, Michigan, crews built nearly 50 units by 2025, a year after it launched eight plans of its own. Hawaii County, Hawaii (on the state’s Big Island), which has been doing it longer, has built 225 units since 2012 out of a library of 56 preapproved plans. 

Officials in cities from Helena, Montana, to Washington, D.C., have proposed preapproved housing plan programs of their own this year, while in California, state law mandates that cities offer preapproved plans for accessory dwelling units, a route other municipalities in the country have also taken. Pew says preapproved housing plans available for free or a nominal fee shortens the preconstruction approval time and thus reduces delays and costs.

The houses in Vermont’s 802 Homes were designed to be affordable and versatile. Importantly, they need to meet Vermont’s pressing housing needs for smaller households, or what’s called the “missing middle” of housing, while still looking distinctly Vermont. 

The home plans were designed for Vermont’s Agency of Housing and Community Development in partnership with the Boston architectural firm Utile, and they include single-family and starter homes, a townhouse, duplex, multifamily units, and a 981-square-foot single-unit railroad flat.

Architectural features like steep rooflines, clapboard siding, and an abundance of pedestrian-oriented street connections, porches, and porticos were designed into the homes based on responses from a statewide visual preference study. The study aimed to determine whether or not residents thought they looked at home in Vermont neighborhoods, and whether they would be happy to live in them or have them built next door.

[Image: Vermont 802 Homes]

The plans in the 802 Homes catalog are designed to be built through either off-site prefab construction or traditional on-site construction, and they’re being tested first in three pilot communities: Essex Junction, Hartford, and Manchester.

Vermont officials next have to finalize construction documents for both on- and off-site construction methods and develop the public-facing catalog and website, and they’re coordinating with developers on potential demonstration projects.

The idea of building homes you can pick out of a catalog isn’t new. Sears, Roebuck and Co. sold and shipped materials to build homes from its catalogs in the early 20th century. Under Operation Breakthrough, a public-private home-building initiative from the Department of Housing and Urban Development under its-then secretary, George Romney, the agency collected proposals for mass modular homes; thousands of units were built before the program was defunded. 

Vermont’s plans take the idea of preapproved home designs and adapt them for local aesthetics and community needs in a way that still scales statewide. If this and other programs can deliver quality, affordable housing at a faster pace, the housing crisis could one day meet its match.

Lukas Bentel created Mschf’s Big Red Boot. Now, he’s striking out on his own

2 weeks 5 days ago

In 2024, Lukas Bentel helped pull off a heist at the Metropolitan Museum of Art. Sort of.

Bentel is a cofounder of the iconic Brooklyn-based art collective Mschf and, until recently, served as its chief creative officer. After nearly eight years with the brand, he’s parting ways with the collective to start his own design practice. He says it is still in an “exploration phase” but will include a major focus on “making things physically with my hands.”

[Photo: Mschf]

As Bentel moves on to his next venture, he brings along a unique formula for creating products that manage to combine high art and satire with mass appeal. He was the creative mastermind behind some of Mschf’s most commercially successful launches, including its era-defining Big Red Boot, culture-war-starting Satan Shoes, and museum-worthy ATM Leaderboard.

Looking back on his time at Mschf, though, Bentel is even more proud of some of the projects that built cult followings: the time his team stole a sink from the Met a piece at a time, “Ship of Theseus-style,” he says, or the time he bought an authentic $30,000 Damian Hirst painting and cut it up into 88 resellable pieces.

In many ways, Bentel’s work defined Mschf’s knack for toeing the line between poking fun at and profiting from modern consumerism. Here, Bentel looks back on his art philosophy, recounts the book of guidelines that turned Mschf from a small art collective into a commercial powerhouse, and explains why it’s time for him to “get back into the dirt.”

This conversation has been condensed and edited.

[Photo: Mschf]

What were some of the learnings that shaped your career pre-Mschf? 

[In grad school at RISD], I stumbled into three or four friends who would make a lot of work together. We started an artist group called Hello Velocity, which made things that were, conceptually, very similar to Mschf. 

There was generally a problem in art school where you would spend so long making something, you’d put something up on the wall in a gallery space, and the only people it would actually engage with would be your peers, maybe your friends, maybe the teacher. Beyond that, almost no one else. That was so frustrating.

I’d read this one book by these two people called Dunne and Raby, and they wrote about speculative design. They argued that objects were great for making critical work, because everybody knows how to interface with an object. Everybody knows how to hold a cup and drink out of it. So if you fuck up a cup, the people that are engaging with it will know exactly what’s wrong. Whereas I think a lot of fine art is, one, put in a space that is completely protected from a physical touch and interaction perspective. And, two, is often only trying to cater to an audience that, to be honest, I’m not particularly interested in all the time.

Hello Velocity eventually became its own marketing and branding agency. What were some of the most memorable projects from that era?

I think probably the best project that we did at Hello Velocity was in 2012 or 2013, and it took the form of a biotech company called Byte Labs. This was at this tech-optimistic period in time where I remember all my peers being like, “I need to get the job at Google.” We were using all that language around that space to make something that, in my mind, was a critical art object—just put in a different environment. What we were doing was soliciting celebrity biopsies so that we could grow in-vitro meat from those celebrities and basically sell human celebrity salami back to their fans.

In my mind, it was a critique of celebrity consumption at its base level. It could have taken the form of a weird neat sculpture in a gallery space, but instead it took the form of this fairly real-looking startup, and then actually got a ton of traction because people didn’t know how to interface with it. 

After running Hello Velocity for five years, how did you end up forming Mschf?

Kevin Wiesner [Hello Velocity cofounder] and I met a guy named Gabe Whaley who had a marketing studio which was called Miscellaneous Mischief. As soon as we met him, we realized that he knew how to operate a business. He had a great team of young, hungry developers and people that could really make things. We realized that we fit each other’s puzzle pieces quite well, in terms of him knowing how to operate and run a business and us trying to make as crazy work as possible. 

Honestly, the ups and downs of the business of Hello Velocity were a lot—I never wanted to work in an advertising space, necessarily. So, we decided we were going to fire all our clients and start something totally from scratch. Kevin and I sat down and came up with a series of rules, sort of like a book of guidelines for how we were going to run this practice. 

In my mind, this was the ultimate Warholian artist practice exercise. It was, “Let’s use all the tools and trappings of a business to make as great of a creative artwork as we possibly can.”

[Photo: Mschf]

You mentioned Mschf’s original book of guidelines. What are some others that stand out?

This is a rule I always bring up because I find it interesting—it was very good for the art practice, but bad for business. We told ourselves we’d never do the same medium twice. By that, I mean if we made a painting, we’d only make one painting ever. If we made a shoe, we’d only make one shoe ever. If we made a book, we’d only make one book ever. Basically, to force ourselves into a headspace of never being too comfortable within a medium.

For the first two years of Mschf, we operated exclusively like this. It was that punk mentality of, “I’ve learned how to play the guitar too well. Shoot, let me give this to somebody else and have them play it.” I’ve worked now with a number of people in the footwear industry, and they almost never think outside of the normal trappings of a shoe, because that’s all they’re thinking about. It was a test of being able to jump into a new medium, try something, launch it, deploy it, see what happens, fail, move on to the next thing.

Are there any projects from that early era that you’re particularly proud of?

We kind of quickly stumbled into some early success. The third thing that we did was this thing called Jesus Shoes, which was a Nike Air Max 97 where we injected holy water that we shipped over from the River Jordan that was blessed by a priest so that you could walk on water like Jesus. 

This was in the height of the very consumer speculation footwear, hype beast time, and we were trying to make fun of that. We were like, okay, well, everyone’s smashing up all these collaborations and spending a lot of time and energy—why don’t we put the two of the most famous brands together, the Roman Catholic Church and Nike? 

To be honest, I didn’t think anybody would consider buying it for real. We priced it at $1,425 because the Bible verse Matthew 14:25 is the verse where Jesus walked on water. We made 12 of them because we were like, “Oh, let’s make 12 for the 12 disciples and each of us can take a pair home,” basically. We launched them, and the people in this footwear universe ate them up like crazy. We sold out in about a minute. 

What are some Mschf releases that you think don’t get enough hype?

Maybe the only piece we did that was actually illegal was a concept I’d come up with to steal a sink from the Metropolitan Museum of Art. The way we did it was Ship of Theseus-style. The idea was, we would go to this one bathroom in the Met and take the faucet, take the handles, take the plumbing, take the drain, piece by piece. But every time we were there and we took a piece, we would replace it. So technically the sink never left the Met.

It never went to court, and I think the Met refused to comment on it, which is probably a good thing from their perspective PR-wise. You don’t want the Streisand effect to happen with the Met sink. In my head, I was always like, ‘It’d be so funny to have a lawyer in court arguing the Ship of Theseus philosophical problem.”

[Photo: Perrotin]

Is it the Met sink? Who can say!

Yes, exactly. It’s sort of the perfect crime. Even though, at the same time, I very much acknowledge that maybe taking a thing outside of the Met is a museum heist of sorts. That’s a piece I loved that skirted under the radar.

So why was now the right time to leave Mschf and go your own way?

It was a bit of a long time coming. It’s been seven years since we started Mschf. From my perspective, I see it in a very linear arc from what Kevin and I were doing in college. We might’ve been the only people there that were as concerned with the art historical canon in comparison to everybody else. For the first five years, we put out a ton of individual work. We kept that schedule going. We technically put something out every 10 days for the first three years.

But at some point in time, I think we decided to focus on other aspects of the business and slowed down our production cadence on individual pieces. Mschf has a lot of great work, and I know a few of the projects that are currently inbound. I just personally felt like I needed to get back to the feeling of making things and being scrappy. I mean, for years I’ve talked about making things with my brother, and we’ve always been in different parts of our professional lives. Either I was running a studio with other people or he was running his practice very successfully. I’m very, very much looking forward to that.

Ultimately, I’m very, very proud of all the work that we all did, and I think we’ve left a legacy there that they’re going to be making crazy things for a long time. But at this point in time, for me, I need to get back into the dirt. I’m very much looking forward to getting back to making things physically with my hands. I feel like I really need to scratch that itch again.

Checked
9 minutes 38 seconds ago
Fast Company
Fast Company inspires a new breed of innovative and creative thought leaders who are actively inventing the future of business.
Subscribe to Fast Company feed