Despite 'unraveling' claims, OpenAI keeps hiring at pace

Commentators flag crisis, bubbles and productivity shocks; our employer listings show OpenAI and Anthropic still adding roles, safety hiring remains small, and broad demand is led by engineering and data.

A split screen: crisis talk, bubble checks, and product shifts

In the past two days, several prominent voices painted sharply different pictures of where AI stands. Zvi Mowshowitz wrote that "OpenAI is attempting to turn its ship around," while Gary Marcus argued that the company’s troubles mark the start of a broader unraveling. Azeem Azhar updated his bubble dashboard with the line "Is AI a bubble? Not yet." Simon Willison highlighted a quiet but material change in how ChatGPT does web search, and he also pressed the case that coding agents can multiply developer output to the point of asking "why should a company have more than one engineer?" Ethan Mollick contrasted how ChatGPT and Claude guide users to build reusable skills.

We test those claims against our hiring tracker, which is built from employer job feeds and is our headline jobs-created indicator, calibrated to the World Economic Forum’s Future of Jobs Report 2025 (11M AI roles created, 9M displaced, by 2030).

Our tracker counts 3,643 open AI roles across 143 employers, taken from their own job feeds. The largest families are Modelling and engineering, with 1,384 open and 340 opened in 30 days across 116 employers, and Data, with 689 open and 160 opened in 30 days across 100 employers.

“Pause” and “course correction,” meet the hiring tape

Mowshowitz describes a company attempting to fix deep problems. He writes that "OpenAI has now taken initial steps to address What Happened leading up to HuggingFace attack, including pauses to development while new safeguards are put in place and problems are diagnosed." In a follow‑up, he adds: "OpenAI has some severe misalignment problems, and experienced total failures of its infrastructure and supervision." He then notes that "OpenAI is now taking active, expensive steps to try and fix the problem going forward."

Our figures show no hiring freeze at OpenAI. OpenAI has 147 open roles, with 61 opened in 30 days. That is consistent with “active, expensive steps” but not with a visible halt in recruitment. If the pause Mowshowitz describes involved significant reallocation toward safety work, the broader market has not moved decisively in that direction yet. Across all employers we track, Evaluation and safety shows 38 open roles, with 8 opened and 2 closed in 30 days, across 15 employers. That remains a small slice of overall demand.

Mowshowitz also notes that "Anthropic revenue continues to climb as they prepare for their IPO, although growth has slowed somewhat recently." We do not track revenue, but Anthropic’s hiring remains strong: 112 open roles, with 30 opened in 30 days. The two leading labs are both active on our board.

“Unraveling” claims vs what employers are posting

Gary Marcus argues that the opening stages of OpenAI’s unraveling have begun, says that hardly anyone trusts the company now, and suggests this is good news for Anthropic. Our tracker does not support a collapse narrative. OpenAI’s 147 open roles with 61 opened in the last 30 days sit alongside Anthropic’s 112 open, 30 opened. Both are adding headcount through their own hiring systems. That is not proof of smooth operations, but it is clear evidence that both continue to recruit.

At market level, there is robust demand spread well beyond model labs. Accenture leads with 599 open roles, 194 opened in 30 days. Capital One has 179 open and 19 opened in 30 days. Amgen lists 172 open and 11 opened. Waymo shows 102 open, 15 opened. PwC has 89 open, 12 opened. Databricks lists 88 open, 19 opened. If trust were collapsing across the sector, we would expect to see broader pullbacks. We do not see that in employer postings.

Marcus also frames events as good news for one competitor. Our numbers show Anthropic is hiring, but they do not show OpenAI retreating. On the evidence we have, demand for talent at both is resilient.

Bubble watch: hiring breadth aligns with “not yet”

Azhar says "Is AI a bubble? Not yet." He writes that "AI revenues have continued to rise, reaching $126 billion over the last twelve months as of July," and that "Our updated dashboard tracking the investment wave currently has no gauges in the red, two in amber, and the rest in healthy green (just)." He does caution that "Funding quality has deteriorated since Sep 2025" and "In our base case, we expect it and economic strain to turn red during 2027."

Our data is consistent with a market that is still hiring across functions. By family: Modelling and engineering 1,384 open, Data 689, Infrastructure 338, Research 200, Product and design 89, Evaluation and safety 38. That distribution matches an industry investing in building and deploying systems, not one seized up by a crash.

Breadth across 143 employers also aligns with Azhar’s read that the wave is still rolling, even if funding quality has slipped. The strong enterprise demand reflected by firms like Accenture, Capital One and PwC supports the view that adoption is not confined to a hype cluster around a few labs.

Product changes and developer leverage without a hiring collapse

Willison reports that "ChatGPT search now uses the site:operator at scale" and that "Their own tracking shows a notable change aligned with the GPT-5.6 rollout earlier this month." He positions this as evidence of invisible but meaningful shifts in how generative engines gather information, with implications for the emerging GEO playbook.

Our tracker does not have a GEO category, so we cannot isolate demand specifically for this tactic. That said, the presence of 689 Data roles and 89 Product and design roles points to continued hiring for teams that would shape and measure these kinds of feature changes.

Willison also argues for counting lines of code in the age of agents: "A lot of people will tell you it makes no sense to measure productivity in lines of code." He follows with "If agents let you produce a thousand lines of debugged code, that really is a very meaningful improvement" and ends the riff by asking "why should a company have more than one engineer?"

Our numbers say companies still want many engineers. Modelling and engineering lists 1,384 open roles, with 340 opened and 34 closed in 30 days, across 116 employers. Infrastructure adds another 338 open, with 57 opened and 3 closed in 30 days, across 71 employers. If managements were taking “one engineer with agents” literally, it is not showing up in their requisitions yet. The agent story may be real in output per person, but the net effect so far appears to be more engineering hiring, not less.

Ethan Mollick describes a split in tool ergonomics, writing "Claude's skill creator (literally, just ask it to make a skill) is still the best way to make reusable skills if you care about the details." He adds that "ChatGPT tends to want to do magic & just do it for you, while Claude does comparisons & shows them, repeatedly asking for input & feedback and doing A &B tests." We cannot attribute specific postings to those workflows, but hiring at both OpenAI and Anthropic remains active, which is consistent with both ecosystems pushing hard on product and developer experience.

Privacy alarms that hiring data cannot confirm or deny

Marcus published a post titled "OpenAI is becoming a surveillance company" and writes, "This certainly raises privacy concerns." Our tracker cannot assess surveillance claims. We do not see job categories that would validate or refute this directly. What we can say is that OpenAI continues to list roles at scale, and that Evaluation and safety roles across the market remain a minority at 38 open across 15 employers.

What to watch next in the postings

Whether OpenAI’s requisitions slow or tilt more toward safety. Today it shows 147 open roles, 61 opened in 30 days, while Evaluation and safety remains 38 open market wide.
Whether Anthropic accelerates hiring into a listing. It has 112 open, 30 opened in 30 days.
Whether enterprise demand keeps leading the count. Accenture has 599 open, 194 opened in 30 days, and firms like Capital One, PwC and Databricks remain active.

We will keep reading the commentary and watching the counts. Our headline jobs-created indicator is counted from employer job feeds and calibrated to the World Economic Forum’s Future of Jobs Report 2025, which expects 11 million AI roles created and 9 million displaced by 2030. Right now, the hiring tape shows resilience, not retreat. Where the commentary and our numbers diverge, we will keep saying so plainly.

What we read

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