Quit OpenAI Because Its Culture Is Broken
By Gammatek ISPL, Industrial Systems & Compliance Analyst at Gammatek ISPL
Last updated: September 2026 | 14 min read
Author block: Gammatek ISPL writes about organizational systems, compliance, and governance at scale for Gammatek ISPL, drawing on direct experience helping fast-growing companies build the operational infrastructure their headcount has outpaced. This piece is based entirely on publicly available, sourced reporting — no claims here are speculative.
Why This Matters
Every few months, another headline claims OpenAI's "culture is broken." It's a satisfying story — the world's most important AI company, quietly falling apart from the inside. But the most detailed first-hand account of what's actually happening there tells a more specific, more useful story than "broken culture": a company that scaled from roughly 1,000 to 3,000 employees in a single year, and is still catching up on the internal systems that kind of growth requires. If you run, manage, or are about to scale any organization — AI company or not — this is the version of the story worth paying attention to, because the failure mode isn't unique to OpenAI. It's what happens to almost any company that grows faster than its systems can support.
What Actually Happened
In July 2025, Calvin French-Owen, a senior engineer who had worked on OpenAI's Codex coding agent, left the company after about a year and published a detailed personal account of his time there. He was explicit that he wasn't leaving over internal conflict — he said roughly 70% of his reason for leaving was wanting to return to building his own company, not dissatisfaction with OpenAI itself.
What he described, though, was a company straining under its own growth rate. He joined as roughly the 1,000th employee; a year later, headcount had tripled to around 3,000, and he was already in the top 30% by tenure. In his own words, paraphrased here rather than quoted directly: everything about running a company that size breaks when you scale that fast — communication, reporting structures, how products ship, how people are organized and hired.
He also described an unusual operational detail that stood out to outside observers: OpenAI runs almost entirely on Slack internally, with email use so rare he estimated receiving around ten emails total during his entire year there. He noted this created a genuinely chaotic, distracting work environment, though one he considered "workable" for people who actively managed their channels and notifications.
Separately, and more critically, other departures painted a sharper picture. Jan Leike, who co-led OpenAI's safety-focused "superalignment" team, left citing a view that safety culture and processes had "taken a back seat to shiny products." Gretchen Krueger, an OpenAI policy researcher, raised similar concerns about internal divisions between teams focused on ethics, safety, and governance versus teams focused on shipping.
Worth noting directly: French-Owen's account actually pushes back on the idea that OpenAI neglects safety broadly — he described this as the "biggest misconception" about the company, and said there are, in his words, many people internally still focused on safety work. The two narratives — "scaling chaos" and "safety deprioritized" — aren't identical, and a fair account needs to hold both without collapsing them into one simple story.
The Pattern This Actually Represents
Strip away the specific company name, and what's being described is a well-known organizational failure mode: growth outpacing systems. It happens constantly outside the AI industry too — in manufacturing, logistics, healthcare, any sector where headcount, locations, or output scale faster than the underlying infrastructure meant to manage people, processes, and accountability.
What typically breaks first, in roughly this order, based on patterns seen across fast-scaling organizations generally:
Communication infrastructure — informal tools (Slack-only, no email discipline) work fine at 50 people and become genuinely chaotic at 3,000, exactly as French-Owen described.
Reporting and management structure — org charts drawn for a 500-person company don't hold at 3,000; reporting lines blur, and decisions that used to happen in one conversation require navigating five.
Hiring and performance processes — recruiting and performance management built for slow, selective growth buckle under the pace of hiring hundreds of people in months rather than years.
Governance and oversight functions — the parts of a company responsible for compliance, risk, and safety are often the most under-resourced relative to headcount growth, precisely because they don't directly ship product — which is exactly the dynamic Leike and Krueger both pointed to.
Why the "Governance Strains Last, Worst" Pattern Matters Specifically for AI Companies
This is the part generic coverage of the OpenAI story misses: in most industries, a company that under-invests in governance infrastructure during hypergrowth creates internal friction and inefficiency. In an AI company shipping systems used by hundreds of millions of people, the same under-investment creates a different category of risk — product decisions about AI safety, alignment, and deployment timing get made inside the same strained, informal, Slack-driven structure as every other decision, without necessarily getting the dedicated process a decision of that consequence arguably needs.
This isn't a claim that OpenAI specifically mishandled any particular safety decision — that's a more specific claim than the available reporting supports, and it's not one this article makes. It's a structural observation: the organizational pattern French-Owen described (chaotic, bottoms-up, Slack-first, built for speed) is, by design, not the same pattern you'd build if governance and risk oversight were the top priority driving the org chart. Both things can be true — a company can genuinely care about safety (as French-Owen insists OpenAI does) while also having built organizational infrastructure optimized primarily for shipping speed rather than oversight rigor.
An Implementation Consideration for Any Fast-Growing Company
The lesson here isn't really about OpenAI specifically — it's about what any organization scaling quickly should build before the strain shows up in departures and press coverage, not after:
Enterprise HR software built for scale, not improvisation, before headcount triples — informal people-management processes that work at 200 employees create real risk (legal, retention, culture) at 2,000.
Enterprise recruiting software with structured pipelines becomes essential once hiring moves from dozens to hundreds of people per quarter — ad hoc hiring processes that worked early tend to produce inconsistent quality and burnout on recruiting teams exactly when the company can least afford it.
Enterprise performance management software to keep reporting structures and accountability clear as management layers multiply — without it, the kind of reporting-structure confusion French-Owen described becomes close to inevitable.
Enterprise scheduling software for coordinating teams working on genuinely different rhythms (research, applied engineering, go-to-market, as French-Owen distinguished) rather than assuming informal coordination will scale with headcount.
None of this is abstract advice — it's the direct, structural fix for the exact pattern described in the most detailed account available of what happened inside OpenAI during this period.
What This Means If You're Scaling Right Now
If your own organization is growing fast — adding headcount, adding locations, adding product lines — the OpenAI story is worth reading less as gossip about a specific company and more as an early warning system. The signs described (everything running through one informal channel, reporting structures blurring, teams unknowingly duplicating work) show up in plant operations and manufacturing companies going through fast growth just as often as they show up in AI labs — the industry changes, the pattern doesn't.
The organizations that handle this transition well tend to share one trait: they build the structural systems — HR, recruiting, performance management, compliance and governance tracking — proactively, before the strain becomes visible externally, rather than reactively, after departures and press coverage force the issue.
How This Connects to Your Own Governance Stack
Whether you're running an AI company or a chemical manufacturing plant, the underlying problem is the same: growth that outpaces the systems meant to track accountability, compliance, and oversight. That's precisely the gap Gammatek's compliance and governance platform is built to close — giving fast-growing operations the structured audit trail and oversight infrastructure that prevents exactly the kind of reporting-structure chaos described above, before it becomes a retention problem, a safety problem, or a headline.
[See how Gammatek helps growing operations build structured compliance and governance systems → https://www.gammateksolutions.com/post/it-s-all-fun-and-games-until-you-give-ai-your-credit-card
Comments