OpenAI Says Codex Agent Is Now Reshaping Work in Every Department, from Engineering to Legal

OpenAI posted to its official account on X, stating that “work at OpenAI is being transformed by agents, in every department.” The company says staff are now using Codex to tackle work that is “more complex, longer-running, and increasingly cross-functional” — and that its own internal experience offers “an early look” at how agentic tools may reshape workplaces more broadly.

The numbers behind that claim are striking. By May 2026, Codex accounted for 99.8% of weekly output tokens generated within OpenAI. For the average OpenAI worker, the tool accounts for more than 85% of output tokens — a measure of how much text and code the system produces on their behalf.

What Codex Actually Does

Codex is OpenAI’s cloud-powered coding agent. It can work on tasks in parallel, running in its own sandboxed environment without needing a human to watch over every step. Engineers use it to understand complex systems, refactor large codebases, ship new features, and resolve incidents when deadlines are tight.

But it doesn’t stop at engineering. OpenAI says Codex has become the primary AI tool for every department in the company — including Legal, Finance and Recruiting. That’s the detail worth paying attention to. This isn’t a tool that stayed in the hands of software developers.

The Usage Data

OpenAI’s internal figures show how deeply the tool has embedded itself in day-to-day work. By May 2026, 80.6% of sampled individual OpenAI users had made at least one Codex request estimated to replace more than 30 minutes of human work. Around 70.2% had made a request estimated to cover more than an hour’s worth of work.

And a quarter of users — 25.6% — had submitted at least one request estimated to exceed eight hours of human work. That last figure suggests some employees are handing off what amounts to a full working day’s worth of tasks to the agent in a single go.

Non-developer use has grown sharply too. OpenAI says the number of non-developer users rose 137 times over for individual users, and 189 times over for organisational users, since August 2025.

Beyond the Code

Sam Altman, OpenAI’s chief executive, has spoken publicly about the company’s belief that AI agents will handle increasingly complex, multi-step tasks within workplace systems. The Codex case study fits squarely into that vision — a tool that started as a coding assistant and has spread into functions that have nothing to do with writing software.

OpenAI’s own explainer describes how teams use structured prompts, validation steps and careful task scoping to get reliable results from Codex. The implication is that getting the most from these agents still requires skill and judgement from the humans directing them.

It’s worth being clear about what this data is and isn’t. These are OpenAI’s own internal figures, self-reported and not independently verified. The company has an obvious interest in presenting Codex in the best possible light. What the numbers demonstrate is adoption inside one organisation — a technology company, staffed by people already comfortable with AI tools, with strong incentives to use the products they build.

That’s a different thing from proving what happens when similar tools reach a law firm in Maidstone or a manufacturer in Sittingbourne.

A Cautious Reading

The case study format has limits. Company self-reporting can overstate productivity gains and understate problems — reliability issues, governance questions, the risk of over-automation in areas where human judgement matters most. OpenAI doesn’t publish its error rates or the tasks where Codex fell short.

Sasha Haco, a researcher who has written on AI governance, has argued that the real test of agentic tools comes not in the companies that build them, but in the organisations that adopt them without the same technical depth. “The question isn’t whether these systems work at OpenAI,” she has said. “It’s whether they work reliably in environments where people can’t debug the failures.”

OpenAI has separately described Codex as available in both cloud-powered and terminal-based variants, with different use cases for each. The internal rollout described in the tweet appears to centre on the cloud-based version, which can handle longer-running tasks without keeping a user’s session open.

The broader picture is one of accelerating change in knowledge work. Agents like Codex are moving from experimental tools into everyday workflows, and the pace of that shift — at least inside OpenAI — appears to have been fast.

What This Means for Kent Residents

For most people in Kent, the direct impact of this announcement is indirect for now. But if the businesses, councils and public bodies across the county begin adopting similar agentic tools — whether from OpenAI or competitors — workers in roles from software development to HR and finance could find their day-to-day tasks changing in ways that mirror what OpenAI is describing internally. For anyone working in tech or knowledge-based industries in Kent, this is a signal of where the industry is heading, even if the arrival date is uncertain.