OpenAI Codex Surpasses 4 Million Weekly Active Users, Knowledge Workers Growing 3x Faster Than Developers
N.R. Finch
OpenAI's Codex has crossed 4 million weekly active users, up more than 5× since its desktop launch in February. Knowledge workers now make up a fifth of the base and are growing over 3× faster than developers — This means → the tool is shifting from a coding assistant to a white-collar operating system.
4 million weekly actives — what does that speed tell us?
OpenAI told Axios exclusively that Codex weekly active users have hit 4 million, up more than 5× since the desktop app launched in February.
This means → Codex has left the niche-developer category and entered the user scale of mainstream workplace software.
Before OpenAI moved, Anthropic's Claude Code and Cowork had already built scale among non-programmers. OpenAI launched the Codex desktop app the following month — a fast-follower play.
Why are white-collar users growing faster?
Knowledge workers — office professionals — now account for roughly one-fifth of all Codex users, growing over 3× faster than developers.
The three fastest-growing task types: data analysis (week-on-week +110%), research (+37%), and knowledge outputs such as reports, memos, contracts, PDFs, and spreadsheets (+36%).
In plain terms = white-collar workers discovered Codex does more than write code — it builds spreadsheets, drafts reports, and pulls research. Once the use cases open up, growth outpaces the developer base.
What kind of product is Codex trying to become?
OpenAI noted in its report that previous waves of workplace software created massive volumes of files and messages, but most of these "workplace artifacts" remain siloed in their own apps.
Codex aims to aggregate key context across platforms: email, calendars, documents, spreadsheets, design apps, and messaging tools like Slack and Teams — with one-click daily auto-briefings.
This reflects a repositioning of Codex from "coding assistant" to knowledge-work operating system — not just writing your code, but connecting all your work software.
Running tasks in parallel: more output, but more strain?
Over 60% of users now run multiple Codex tasks simultaneously at some point during their day. In mid-April that figure was below 50%.
This means → users treat AI like multiple production lines running at once — output goes up, but so does the cognitive load of supervising them all.
OpenAI co-founder and current Anthropic member Andrej Karpathy said on a podcast that he has been in a state of "AI-induced derangement" since December, pushing the tools to their limits. Rootly co-founder and CTO Quentin Rousseau compared the experience to "the difference between running a marathon and binge-watching a show" — one leaves you exhausted, the other keeps you up all night.
Can AI replace experts? A Stanford professor's real-world test gives the answer
Stanford GSB professor Andrew Hall had Claude Code update a paper published five years ago. The tool handled data collection, analysis, chart generation, and drafting — "with very little prompting."
But after a graduate student reviewed the output: data collection had gaps, and some coding was inaccurate.
In plain terms = AI can handle roughly 80% of the work, but the remaining 20% is exactly what matters most. Hall's own words: "You very much need a PhD-level expert closely supervising." This reflects the real position of current AI tools: powerful assistant, not independent expert.
Content is for reference only, not financial advice.