AI Companies Are Hiring More

Today on the AI Daily Brief: the latest numbers on AI and jobs. Before that, in the headlines: is OpenAI about to give 5% of the company to the US government? The AI Daily Brief is a daily podcast and video about the most important news and discussions in AI. Alright friends, quick announcements before we dive in. First of all, thank you to today's sponsors, KPMG, Rackspace, Blitzy, and HyperAgent. If you want to get an ad-free version of the show, go to patreon.com/aiDailyBrief, or you can subscribe on Apple Podcasts. And to learn more about sponsoring the show, head on over to aidailybrief.ai/sponsors or shoot a note at sponsors@aidailybrief.ai. Well, all those conversations about AI companies getting more tied up with the government are getting a lot more real. OpenAI has proposed handing over a 5% stake in the company to the US government. Now, obviously there has been chatter about this for some time now, and honestly, if you go way, way back in Sam Altman interviews from 3 or 4 years ago, it was clear at the very beginning of the project that he thought this sort of pursuit of AI was something that inherently should have government involvement from the very beginning. I can't find the exact interview, but I remember thinking that he almost seemed surprised that the US government wasn't interested in getting involved back then. Now, coming back to the deal now, this potential 5% stake would be contributed to a sovereign wealth fund, structured in a similar way to the Alaskan Permanent Fund, which collects oil and mining revenue for the benefit of citizens. At current valuations, the stake would be worth around $42 billion, and it's unclear whether the administration would be purchasing the stake or whether it would be a gift to the American people. The other wrinkle is that OpenAI has proposed that all leading AI developers should contribute 5% of their equity. Beyond Anthropic, this might include Google, Meta, and others, and it's not clear that any of these other firms would agree with OpenAI's proposal or if they're participating in discussions. The FT pointed to what many people assume is a pretty clear quid pro quo, writing, giving the government an ownership stake could help secure good relations with the administration and would mark an attempt to address political blowback by sharing the wealth generated by AI with the public. Now, in terms of handicapping where these conversations are, sources said the discussions were still in the early stages, referring to them as, quote, conceptual. They also added that such a deal might require an act of Congress to implement regarding this strange new era of tributary capitalism. Also on Wednesday, memory producer Micron agreed to invest $250 million in Trump accounts, the government-funded investment accounts for children that were introduced last year. On Truth Social, Trump declared Micron a, quote, truly great American company, one of the hottest anywhere in the world, adding that this is, quote, the biggest corporate investment of its kind and will help jumpstart the American dream for these fabulous children as we celebrate America's 250th anniversary. Trump accounts so far have been largely funded by individual philanthropy, including a $6.25 billion donation from Dell CEO Michael Dell and his wife Susan, although other companies have made smaller contributions to the program. Look, it's pretty clear at this point that the Overton window on these sort of deals is shifting, and I'm sure we'll have more context to talk about it in the future. So before I say something that makes a big chunk of you angry, let's move on to the next story. Meta is reportedly going down an Elon-esque path Planning to launch a cloud services business as a way to monetize their excess AI capacity. According to sources speaking with Bloomberg, Meta is developing plans to launch a cloud business that will sell access to their AI infrastructure. One potential plan is selling access to models hosted on Meta's data centers, similar to the AWS Bedrock platform. Another approach would be to sell access to raw compute along the same lines as neoclouds like CoreWeave. Sources said the plans are still in development and subject to change, but they're far enough along to have a name and a leadership team in place. The business line will be called Meta Compute and is led by Head of Infrastructure Santosh Janardhan working alongside Daniel Gross of Meta Superintelligence Labs and President Dina Powell McCormick. Now careful listeners will know that this is not the first we've heard of this plan. Back in May, Mark Zuckerberg told investors that the idea was definitely on the table and that they'd been fielding requests from other companies. At the time, Zuckerberg said no deals had been struck as Meta had strong internal demand for all that compute. However, he mentioned that external sales would be a natural outlet if Meta overbuilt capacity. Now this plan of course mirrors Elon Musk's pivot to cloud. In early May, SpaceX signed a $1.25 billion a month deal to provide compute to Anthropic. This was followed closely by deals with Google and later a more modest deal with Reflection AI. These deals dramatically improved the bottom line for SpaceX, with compute sales now estimated to be their primary revenue driver ahead of Starlink. And just as the market liked Elon's moves, they seem to like Zuckerberg's bob and weave as well. The report triggered an immediate response, sending Meta stock soaring by as much as 10% before closing the day up 8.8%. This was Meta's best single-day performance in 6 months. Neo Clouds CoreWeave and Nebius were pummeled, losing 14% and 17% respectively. Jefferies called the plan strategic, comparing it to the early days of AWS, where Amazon was able to monetize excess capacity from website hosting to help finance further investment. Mizuho wrote that they don't believe the business will be a meaningful short-term revenue driver for Meta, Instead, they cast it, quote, more as planning for all potential scenarios, including a Plan B. Still, analysts wrote that compute sales would be an overall positive, adding what they call a margin of safety to medium-term earnings. Now, in the finance corner of X, the debate was raging. Some believe this will lead Meta to cut back on CapEx, with the pivot representing a tacit admission that they've overbuilt. The bullish view, however, is that Meta will finally deliver a solid revenue narrative for their AI spend while also cutting investment costs, triggering a significant boost to earnings. Look, my guess is, as we've seen with SpaceX, this does not mean the end of Meta's efforts to train models or anything like that. It's just taking advantage of a resource they've built. Wright's rune from OpenAI: you either die a frontier lab or live long enough to see yourself sell compute. Now speaking of SpaceX, the Wall Street Journal reports that SpaceX showed investors a prototype of an AI device prior to IPO. The journal described it as a, quote, "handset-like device designed to reshape how humans interact with AI." They wrote that it's slimmer than an iPhone, would run on a proprietary operating system, and would integrate technology from xAI. Elon was quick to deny the reporting, calling it utterly false within minutes. And it's kind of difficult to know what to make of the story. It could be Elon responding to Sam Altman's desire to develop the dominant AI device. It could just be another part of the IPO hype train. Or it could be a new vertical play from Elon Inc. SpaceX bought up a slice of the wireless spectrum in 2025, and analysts have recently suggested that they should acquire T-Mobile as a way to break into the mobile market. So, are we going to see the combination of a mobile carrier, Starlink for mobile internet, and an AI-first handset? Over in Anthropic land, the company has rolled back spyware targeting Chinese labs after controversy erupted on Reddit. Earlier this week, a Reddit user called LegitMichael wrote, Anthropic embedded spyware in Claude code and attempted to hide it from you. They explained that since a software update in early April, Claude code has been checking whether users have a proxy enabled, and if so, the software is covertly transmitting information back to Anthropic through surreptitious changes to the system prompt. Michael claimed the software will tell Anthropic whether a user is in China or even if a user is associated with a Chinese lab. This information was extrapolated from timezone settings and other metadata. Now, we knew Anthropic was doing something to monitor the activities of Chinese labs reportedly using their models for distillation. Their recent letter to Congress and prior research reports had specific numbers that implied some level of usage monitoring. And despite the Claude automoderator on the Reddit page calling this a nothingburger, Claude co-developer Tariq announced that this approach had been rolled back, posting: This is an experiment we launched in March that was meant to prevent account abuse from unauthorized resellers and protect against distillation. The team has landed stronger mitigations since then, and we've actually been meaning to take this down for a while. We've merged the PR, and this should be fully rolled back in tomorrow's release. Now, I think the thing to keep an eye on here is not so much Anthropic trying to decrease Chinese distillation. It's another reminder of the visibility Anthropic has into the work being done on their systems. And yet, if anyone does have concerns about Anthropic, they're certainly being shoved to the side right now as everyone flocks excitedly back to Fable. After an early morning of mashing the refresh button, users regained access to Fable 5 on Wednesday, and it seems just as good as people remembered. Elvis Sun wrote, Fable 5 is so effing good. Here's everything I did in the last 2 hours. 1. Audit my business, found 3 high ROI tactical things to work on, mostly narrowing down on retention and acquisition levers. 2. Solved my hardest backend problem that Codex and Opus were too dumb to crack despite many attempts. 3. Solved my hardest engineering problem with media lists agent reliability and future design directions. 4. Designed a content engine I've been trying to build. Honestly, I'm starting to feel Fable is smarter than me. The future of this is getting pretty weird. Andrew McCallip tweeted, Fable is an absolute monster. The output quality is wild, crisp, comprehensive, fast, and the aesthetics are just gorgeous. It's just machine gunning PRs. I don't mean to gush, but I'm floored. I don't need a benchmark chart telling me, actually, SweepBench Pro is 2 points higher. If you can't viscerally feel a model's performance by now, no chart is going to help you. Still, one of the big questions for the re-release was how often Fable would kick benign coding tasks over to Opus, and on that front the reports are pretty mixed. Some are complaining that it is happening near constantly. Developer Nick Dobos wrote, Fable is mega nerfed. Two chats on different projects, each routed 4-10x more tokens through Opus than Fable. Fable did only 20% of the work. WTF, disappointing. Is anyone else seeing this? On the flip side, Bridgemind also noted that quite a lot of their work was being routed to Opus, commenting, "I just paid $321 for a coding session where Fable-5 refused to do the work." And yet others like analyst Max Weinbach wrote that after a few hours of working, he was yet to have Fable refuse a request. Now other people are focused on figuring out the best way to get full value out of Fable. AI content creator Theo, for example, found that it was at its best when running other agents, commenting, "It's the first model that feels like it actually gets how to use and orchestrate agents. It has way more taste than OpenAI models, so the code it writes is way less cringe. I won't ship APIs or SDKs without Fable taking a look first. Going back to the point that I was making yesterday, policy reacher Miles Brundage suggested that having Fable as an orchestrator makes the Sonnet 5 release make a lot more sense. Taking a step back, Professor Ethan Mollick wrote, been reading all sorts of posts about the best way to develop workflows for Fable and it reminds me of how little we actually know about the best way to organize work for long-running agents. Nobody has enough experience or has done enough testing to reach any real conclusions. In other words, figuratively and literally, we are still on day one of figuring out how this new class of models work. And in welcome news, late on Wednesday, Anthropic reset weekly usage limits so users can experiment a bit more. However, that's going to do it for the headlines. Next up, the main episode. One of the most important AI questions right now isn't who's using AI, it's who's using it well. KPMG and the University of Texas at Austin just analyzed 1.4 million real workplace AI interactions and found something surprising. The highest impact users aren't better prompt engineers. They treat AI like a reasoning partner. They frame problems, guide thinking, iterate, and push for better answers. And the good news? These behaviors are teachable at scale. If you're trying to move from AI access to real capability, KPMG's research on sophisticated AI collaboration is worth your time. Learn more at kpmg.com/us/sophisticated. That's kpmg.com/us/sophisticated. One of the more interesting shifts in enterprise AI right now is how quickly the conversation is moving towards infrastructure and operations. As AI moves into core workflows, regulated data environments, and agentic systems, enterprises need governed infrastructure and inference that can operate reliably day-to-day with clear operational accountability built in from the start. As those systems scale, the operating model increasingly becomes part of the AI strategy itself. Rackspace Technology is the operator of the full enterprise AI stack, from agents to infrastructure across private cloud, hybrid cloud, and edge environments. Rackspace builds and operates governed AI infrastructure, inference, and production AI systems for organizations where sovereignty, compliance, and uptime are non-negotiable. Their foredeployed engineers stay embedded beyond deployment to help operationalize and run AI in live environments. To learn more about where enterprise AI runs and outcomes scale, go to rackspace.com. Want to accelerate enterprise software development velocity by 5x? You need Blitzy, the only autonomous software development platform built for enterprise codebases. Your engineers define the project—a new feature, refactor, or greenfield build. Blitzy agents first ingest and map your entire codebase, then the platform generates a bespoke agent action plan for your team to review and approve. Once approved, Blitzy gets to work autonomously generating hundreds of thousands of lines of validated, end-to-end tested code. More than 80% of the work completed in a single run. Blitzy is not generating code, it's developing software at the speed of compute. Your engineers review, refine, and ship. This is how Fortune 500 companies are compressing multi-month projects into a single sprint, accelerating engineering velocity by 5x. Experience Blitzy firsthand at blitzy.com. That's B-L-I-T-Z-Y dot com. This episode of the AI Daily Brief is brought to you by HyperAgent, where you run fleets of agents your team can manage together. New users get $1,000 in inference. Forget local agents and chat workflows waiting on your laptop to be prompted. HyperAgent deploys always-on agents in the cloud, doing real work across the tools your team already uses. Marketing's agent turns competitor moves into landing pages. Sales' agent enriches leads, drafts emails, and updates the CRM. Ops' agent chases the paperwork and tracks the budget. Every agent has access to shared context and follows your rules about scope and approvals. It's time you add agents that feel like teammates. Hire yours at HyperAgent, built by the team at Airtable. Claim your $1,000 in inference at hyperagent.com/ai-daily-brief. Welcome back to the AI Daily Brief. Today we are checking in on the AI and jobs narrative. We actually haven't had all that much of this for a while, but there are a couple of new interesting studies that are providing some more information in the unending question of what AI is and will do when it comes to jobs and employment. The first is a new report from the Center for AI Safety suggesting that Fable is a step change in AI's ability to do real work. The report was an update to the center's Remote Labor Index. The benchmark measures a model's ability to complete economically valuable tasks drawn from common freelance professions. The tasks include 3D modeling, architecture, graphic design, video and audio editing, data analysis, and programming web apps. Now, what makes the index interesting is that whereas something like the Meter Long Horizon Task Test uses something like 50 or 80% as the standard metric, for the Remote Labor Index, it measures a model's ability to do these tasks at a quality that a paying client would accept. Every deliverable that is tested is judged by human evaluators against what they call a gold standard deliverable, which was produced by a paid professional. Meaning that this is a much harder test to score well on. GPT-5.5, for example, only scored a 6.3%, with Opus 4.8 just a little bit ahead of that at 8.3%. Fable represents a big jump up, coming in at 16.1%. When they first ran the benchmark late last year, GPT-5.2 was the top performer at 2.5%, meaning that in a very short period of time, the numbers have increased significantly. The center wrote, "The frontier has more than quadrupled in under 8 months, a concrete signal of how quickly economically capable AI agents are advancing." The pace of improvement is also clearly evident in the examples offered by the center. Some of the examples they gave were photorealistic renders of rings, producing an advertisement video for an imaginary company called Skyline Tree Services that needed to be a 60-second flat design 2D animated advertisement with a voiceover provided. Finally, there were floor plan renders. Now, to be clear, completing 16% of freelance tasks at a professional level is very different to being able to do 16% of all human jobs. The center itself was even critical about their own findings, writing, "Today's AIs still fall short of professional quality on most projects." However, they added, "This increase in automation rate has been rapid, occurring in less than a year. The index spans a wide range of economically valuable work, so this trend directly captures how quickly the automation of remote work is advancing." Interestingly, although most people agree that the exponential increase is impressive, there are wildly different interpretations of the implications. On one end of the spectrum, you have folks like V who write, "Remote workers and freelancers need to think twice. This benchmark shows just how high the success rate of AI models can be in handling digital tasks such as creating 3D models, ad videos, rendering house floor plans, etc. Even though Fable-5 is still considered low, it won't be long before this number climbs higher." It's becoming harder and harder to deny that digital work can be automated by AI. On the other end, you have folks like Scott who writes, "Even Fable-5 is still at a 16% automation rate. That means 84% of the time, human is needed. In my opinion, the real work has complexities orthogonal to what current AI systems can cover and require a fortress of priors that are a given for humans but not inherently installed for agents." Frankly, I think this report has a little something for everyone, no matter where you find yourself on the AI jobs question. It shows that the quality of work is advancing very quickly, but also that there are huge challenges in actually getting to a fully economically viable product. And the space in between those two represents a lot of opportunity for people, including the freelancers themselves, to redesign their economics, redesign their deliverables, and figure out how to adapt to a changing landscape of opportunity. Indeed, one important distinction that we talk about a lot here, but that is finally finding its way into the broader discourse, is the distinction between tasks and jobs. At a recent event run by the European Central Bank, OpenAI's chief economist Ronnie Chatterjee said that he doesn't think AI will replace human workers. He said, "Just because a task is exposed to AI doesn't mean it's going to substitute for that. We need to think a lot harder about what jobs are, how they will evolve, and that will help us give advice to people about labor market trends rather than being optimistic or pessimistic." Chatterjee discussed his own profession and the way that economists have been on the list for technological replacement for decades but are somehow still able to find work. He commented, "My dad was an economist also in 1985. His job was very exposed to the personal computer when he first put one in his office. But instead of using a punch card in a big room and a mainframe computer to run regressions, now he could run them on his computer. And it was a complement to his work over time that made him more productive." Chatterjee cited software developers as the profession most in the firing line for AI replacement, but claimed that we haven't seen a lot of evidence thus far. He argued, "Those jobs shrinking as AI capabilities increased, that really hasn't happened to the same extent people were predicting." Now, obviously this is coming from an OpenAI source, but as we've seen, that certainly doesn't guarantee an optimistic take. Now, there has of course recently been a shift in this discourse, particularly from OpenAI, not so much from Anthropic yet, where Sam Altman has basically come out and said that he was wrong about the way that he thought AI was going to interact with jobs, much to his delight, as he and OpenAI are now much more convinced that this is going to be an augmentation rather than a replacement situation. Now, adding some messiness to this, There is still something clearly happening in the sectors most exposed to AI work replacement in the US. According to the most recent labor data, tech and finance are seeing the worst hiring outcomes. Payroll data from the Bureau of Labor Statistics shows that these two sectors are now losing 28,000 jobs per month on average so far this year. Overall, the year has seen a boost in hiring with 113,000 jobs added per month on average, but if you excluded tech and finance, that figure would be much higher. John Challenger, the head of private payroll data analytics firm Challenger Gray Christmas, blames AI disruption. His firm has tracked over 100,000 job cut announcements this year and maintains that based on the amount of times AI is mentioned, quote, "It's certainly making an impact in a way that no technology has before." Now, of course, it remains quite difficult to separate actual AI replacement from the useful narrative of AI replacement. In the tech sector, many still believe that AI is being used as a cover story— for inevitable mass layoffs who have more of their origins in overhiring back in 2022. Pooja Sriram, a senior U.S. economist at Barclays, said, "Some of this could genuinely be productivity replacing workers, but the narrative that keeps coming up is really a cost-cutting exercise by a lot of firms given the amount of investments they have committed towards AI." Now, obviously, the politics of AI are getting more and more acute, and one interesting recent story in the New York Times was about what China is doing around AI and jobs. Now, one of my long-held theses is that if we started to see significant job displacement from AI, I think that one of the policies the United States would try before moving to something more dramatic like universal basic income would be to provide incentives for firms to not fire people. I don't know whether it would be carrot or stick type incentives, i.e., job cuts and subsidies on the positive end, or fines and fees on the negative end. But basically the idea would be for policy to create an incentive for organizations to keep people employed. Part of why I've been at least theoretically interested in some type of policy like that is that while it is more involvement in the private sector than I like for my government in general, it is also my longstanding belief that most companies are going to have to get through the efficiency phase of AI to get to the opportunity phase of AI. In other words, they're going to have to go through all the cost cutting and efficiency gains to realize that the real opportunity of AI is in new opportunities. To the extent that there were policies that incentivized firms to keep people employed that might provide an acceleration over to the opportunity side because people have to figure out what to do with all that excess labor that they don't theoretically need anymore to deliver on their current product suite, leading potentially to some new thinking and new ideas. Now according to this New York Times report, China appears to be exploring policies that are somewhat in this vein. The Times writes, "The government is leaning heavily on companies to avoid layoffs, and those who don't fall in line might find themselves in court." Indeed, the idea of AI as augmenting people rather than replacing them is increasingly a matter of legal precedent in China. In April, the Times writes, a court ruled that a tech company had illegally laid off a worker after replacing him with AI software. The Hangzhou Intermediate People's Court wrote, the development of AI technology should be applied to liberating labor, promoting employment, and improving people's livelihood. Labor laws allow employers to undertake technological changes and upgrade their operations, but it should also take into account the protection of workers' legitimate rights and interests. The New York Times concludes, just how this will work in practice and how far the government is actually prepared to go with companies that don't comply remains to be seen. But what these rulings underscore is how much China is thinking about the problem. Still, meanwhile, back here in the US, markets might be doing some of this work for us. You're starting to see more and more stories of companies that were quick to fire based on AI, or at least blame AI for their firings, reversing those decisions. Ford has recently made news for their rehiring of what they call graybeard engineers. Writes Bloomberg, Ford Motor Company took an unusually human approach to fixing its stubborn quality problems. It brought back what it calls graybeard engineers to help train younger staff and to reprogram the artificial intelligence tools that weren't getting the job done. Over the last 3 years, Ford said it has hired 350 veteran engineers, many of them former employees and others from suppliers, to help address seemingly intractable quality woes that have cost the automaker billions. The result? Ford is the top mainstream brand in the latest J.D. Power Initial Quality survey, said Charles Poon, Ford's vice president— Ford's VP of vehicle hardware engineering. Artificial intelligence is a fantastic tool, but it's only as good as the information you use to train it. Over prior years, we didn't pay as much attention as we should have to the experience of our most knowledgeable engineers that have been with us through many product cycles. And Poon actually went farther. He said, "Mistakenly, we thought that just by introducing AI and ingesting the design requirements that we had, that we could produce a high-quality product. But we recognized that for us to enhance some of our automation and machine learning and AI tools, we needed to ensure that they were trained by the most experienced individuals." Now, obviously this is just one company, but I do think it's a leading indicator of a narrative you're going to see more, which is companies recalibrating a little bit where AI is going to sit in their overall labor stack. Of course, one company is just an anecdote, but the last report I wanted to cover today is based on a lot more data. In a research collaboration with Revelio Labs, Ramp recently correlated firm-level AI spending against payroll data for 21,000 US businesses. The headline stat? They found that companies with high AI adoption were growing headcount at 10% on average across the past 2 years, while companies with low AI adoption were basically flat. What's more, the beginning of aggressive headcount growth syncs up to the beginning of the company's AI adoption, suggesting a strong causative factor. And as to concerns about entry-level employment, well, Ramp found that headcount growth was also stronger at the entry level, running at an average of 12% compared to 10% overall. Now, Ramp's lead economist, Eric Karazian, made sure to point out that data can hide a lot of nuance and tell a lot of stories. In sharing the results, he wrote, "You should be skeptical. Companies that adopt AI are already fast-growing." However, he pointed out that the study attempted to control for a lot of the obvious factors. For example, the study attempted to match like-for-like firms with a control group that hadn't adopted AI, hoping to ensure that RAMP wasn't just measuring the propensity of firms with high existing growth to spend the money on AI adoption. He also noted a prominent learning curve to firm-level AI adoption. Headcount growth didn't begin until 6 to 12 months into AI adoption plans. Now importantly, while this study was looking at firms with AI adoption, This is not primarily about the firms that are token maxing and spending millions of dollars on AI. Ramp's threshold for high levels of AI adoption was fairly modest, with companies spending an average of $30 per employee per month in the early phases. AI spend did ramp up alongside headcount growth, but Karazian said it wasn't a dramatic cost, well below $1,000 a head. As far as takeaways go, Karazian thought that this was very good news, especially for young people. He wrote, this is our first evidence that high AI adopting firms are hiring different kinds of employees. We believe they are selecting for a new set of skills, specifically people who know how to use AI and use it well. Entry-level workers, especially recent graduates and college students, are a natural place to look. Ultimately, it is way too early to know exactly how AI and jobs are going to play out. Still, any of you who are even a remotely regular listener will know that I am extremely optimistic on this front, and I am encouraged to see not only the narrative, but some of the numbers validate at least parts of that optimism. One final set of numbers that I'll leave you with come from Box CEO Aaron Levy who wrote: "At Box, we recently did a survey of 1,600+ mid- and large-sized companies, and the findings were similar to RAMP's. 58% of respondents expected headcount to rise over the next 3 years. That figure climbs to 79% among the most mature adopters of AI. The more advanced AI adopters expected to grow their headcount at a greater rate in the future than others." In reality, he concludes, this is what actually you should expect to happen. If a company can get more customers because they use AI in sales for a counter-market intelligence, they hire more salespeople, not fewer. If you can build way more software than before, you end up hiring more engineers because the projects get bigger and you take on more. As always, my long-term optimism about this does not mean that I think there isn't going to be important displacement that happens in the short term. I think that there are categories of jobs that likely just get entirely wiped off the face of the planet. And I think that there are going to be vulnerable populations who have built their entire careers around something that AI happens to do well, who are at a stage in their career where just switching to something new isn't going to be exactly viable. But boy, creating good policy to support very specific interventions around particular at-risk roles and populations is a way different task than dealing with an AI job apocalypse. Anyways, guys, that is going to do it for today's AI Daily Brief. Appreciate you listening or watching as always. And until next time, peace.