RSI and AGI talk spike, but safety hiring stays tiny

Claims of a capability inflection and rising safety risk meet our live count of 4,206 open AI roles, which shows enterprise build-out surging while safety hiring remains small.

A week of big claims, bigger tension

In 48 hours, leading voices painted a picture of AI lurching forward on capability and risk, while enterprise demand surged. Zvi Mowshowitz said, "The first Millennium Prize, Navier-Stokes, has fallen to AI." He doubled down with, "This new AI has, eight days after it started training, solved Navier-Stokes." Ethan Mollick argued frontier players have crossed a line: "This week brought some of the clearest statements we've heard from both Anthropic & OpenAI that some form of recursive self-improvement has been achieved, though it still sounds early" and warned "RSI would cause a rapid gain in AI ability & the first firms to RSI may get an unsurmountable lead." Azeem Azhar reported that enterprise hands are up and spending is rising: "This year, nearly every single hand went up; I estimate some 95%." On risk, Noah Smith opened with, "This was the week that AI safety hit the big time," while Gary Marcus urged caution and pacing and asked, "Could rogue agent swarms take over the entire internet in the next six months?" before judging, "We think that’s (a) a vague claim, and (b) pretty implausible as best we can understand it."

We test these claims against our tracker of employer job feeds. Our count shows 4,206 open AI roles across 178 employers. In the last 30 days employers opened 1,014 modelling and engineering roles and 508 data roles, far outpacing closures. The story in the jobs is a rapid build-out of applied capability, not a pause.

Enterprise adoption claims match the hiring tape

Azhar wrote, "There are decades when nothing happens. This week, I am allowing myself that cliché." He then offered a concrete pulse check: "I spoke to 250 IT executives in Las Vegas last week, and I asked my usual question: “How many of you have serious, meaningful results from your AI initiatives?”" He reports, "Every one of them plans to spend more next year than they have this year." and "it’s no surprise that our latest revenue numbers show AI revenue grew faster in August than in July, and faster in July than in June."

Our data supports the direction of travel. Consulting and enterprise adoption engines dominate openings. Accenture leads with 591 open roles and 797 opened in 30 days. PwC lists 125 open and 97 opened in 30 days. Financial services demand is visible with Capital One at 194 open and 138 opened in 30 days. These figures are consistent with the pattern Azhar describes of widespread deployment and budgets pointing up. The role mix backs that up: 1,638 modelling and engineering roles and 804 data roles open, across 141 and 123 employers respectively, indicate large-scale implementation work rather than tentative pilots.

Frontier leaps and RSI narratives meet a dispersed jobs market

Mowshowitz praises Astra’s step up: "Astra is an excellent model." and "This is a big deal." He even writes, "This is the first time a debate over whether a model ‘was AGI’ felt non-silly." Mollick raises the stakes, positing early recursive self-improvement. If that dynamic yields an "unsurmountable lead," we would expect hiring to concentrate heavily in a few frontier labs.

Our tracker shows robust frontier hiring but not concentration consistent with a runaway lead in the jobs market. OpenAI has 166 open roles and 46 opened in 30 days. Anthropic has 125 open and 45 opened in 30 days. These are significant, but they are far outnumbered by the integrators and adopters that are standing up AI inside client environments and large enterprises. Accenture’s 591 open roles and PwC’s 125 open suggest the broader economy is staffing to absorb and apply capabilities, not just watch frontier players pull away. On this evidence, Mollick’s warning about an unsurmountable lead is not visible in hiring yet.

Safety is loud in commentary, modest in headcount

Smith’s frame is categorical: "This was the week that AI safety hit the big time." Marcus, reacting to Dario Amodei’s call for pacing, writes, "We should all salute their tentative agreement to slow down the acceleration of this technology, given that not one of the AI companies seems to have good control over it." He adds, "But yeah, there are some reasons to be cynical, too." On threats, Marcus scrutinizes claims that agent swarms could take the internet, concluding, "“Taking over the entire internet” is, once again, immense in scope yet so vague it hurts."

Our data shows only 40 open roles in evaluation and safety across 19 employers, with 11 opened and 5 closed in the last 30 days. Research roles total 230 open across 50 employers. By contrast, modelling and engineering plus data account for 2,442 open roles. If safety has hit the big time in public debate, it has not yet translated into big-time safety headcount in open hiring. Nor do we see signs of a hiring pause. Openings at OpenAI and Anthropic remain in the triple digits, and broader industry hiring is brisk. The jobs tape does not show a slowdown consistent with a near-term industry pact to pace.

Agents are useful but opaque; hiring tilts to infra, not product

Simon Willison ran an everyday test of an ambitious agent. "It worked for 27 minutes and produced exactly what I'd asked for, as both an embedded visualization and downloadable GPX file and GeoJSON files." He then hits a practical snag: "Frustratingly, the actual code it ran and exact details of what it did weren't visible to me in the ChatGPT UI." and "I see this lack of transparency is an anti-feature." He also summarized the capability in short form: "This is pretty neat: ChatGPT Work and GPT-6 Astra (I used "Max") can take an address and produce a 5K/10K circular running route starting from that address, using OSM data"

Our tracker suggests businesses are staffing the layers that might address these gaps. Infrastructure roles total 413 open across 93 employers, while product and design sits at 93 open across 46 employers. That balance implies organizations are investing in the plumbing to run, monitor, and harden agents before a large wave of productized, end-user features. Willison’s transparency friction aligns with a market still building the observability and governance stack for agentic systems.

Willison also quotes Paul Ford to recalibrate expectations for developer work: "A.I. can write very good software, but it also makes it easy to do someone else’s job badly, which is part of why all those projects fail." Our data is compatible with that view. The 1,638 open modelling and engineering roles and 804 data roles suggest teams are hiring humans to steer, evaluate, and integrate systems rather than replacing those roles at scale.

Profit pressure and margins meet steady lab hiring

Mollick pushes back on a common narrative: "There seems to be a persistent belief that frontier AI companies are unprofitable serving models but it appears that Anthropic has 80%+ gross margins on inference." Our data cannot verify margins. It does show Anthropic with 125 open roles and 45 opened in the last 30 days. That is consistent with continued investment in growth, not retrenchment. OpenAI’s 166 open roles show the same pattern. If inference margins are strong while training is expensive, we would expect frontier labs to keep hiring in revenue-facing and platform roles. We do see a heavy tilt toward modelling and engineering hiring overall, though our tracker does not break out those roles inside specific firms.

Mollick’s broader caution is relevant to policy and workforce planning: "We don’t need better models for AI to have wide impacts on jobs & society and we need to be preparing to encourage good outcomes & mitigate bad." Our live count of 4,206 open AI roles across 178 employers is a concrete sign of those impacts now. Our headline jobs-created indicator is calibrated to the World Economic Forum’s Future of Jobs Report 2025, which projects 11 million AI roles created and 9 million displaced by 2030. That long-run framing underscores why today’s hiring mix matters.

What the market is telling us

Capability claims are rising. Mowshowitz’s "This is a big deal" and Mollick’s early RSI note set that tone. Our numbers cannot validate capability claims, but they show no cooling in hiring.
Enterprise adoption is real. Azhar’s room of raised hands lines up with leadership positions at Accenture, PwC, and Capital One, and with the dominance of modelling, engineering, and data roles in openings.
Safety is loud in discourse, light in hiring. Only 40 open evaluation and safety roles suggests organizations have not yet shifted significant headcount toward safety, even as Smith and Marcus elevate the debate.
Agents work, but ops is the bottleneck. Willison’s success and transparency pain points line up with a market focused on infrastructure, not yet on large product rollouts.

Across the commentary, one prediction does not show up in the jobs yet. Mollick worries first movers may achieve an "unsurmountable lead". In hiring terms, the lead is still with integrators and adopters. Frontier labs are growing, but the bulk of new seats to fill are in the rest of the economy, getting AI into production.

What we read

Every quote above is taken verbatim from one of these posts.