Knowledge workers are adopting OpenAI's Codex at a striking pace, now representing roughly one-fifth of the platform's user base and growing more than three times as fast as developers. The shift signals a fundamental pivot in how AI tools are being deployed beyond software engineering.
Codex has crossed 4 million weekly active users, more than five times the number at launch in February. The surge reflects a broader market movement toward AI systems that can orchestrate work across scattered digital tools: email, calendars, documents, spreadsheets, design software and messaging apps like Slack and Teams.
For office workers, the appeal is practical. A single click can automate a morning briefing that pulls calendar events, unread emails and other flagged items into one dashboard. The fastest-growing use cases among non-developers reveal where the demand lies. Data analysis jumped 110% week over week. Research tasks climbed 37%. Knowledge artifacts, like reports, memos and contracts, rose 36%.
Usage patterns show deepening reliance. More than 60% of users now run multiple Codex tasks simultaneously during their workday, up from under half in mid-April. The trend reflects OpenAI's larger bet: that AI agents can become the central nervous system for knowledge work, connecting and summarizing the sprawling mess of files and communications that modern offices generate.
Andrew Hall, a Stanford Graduate School of Business professor, deployed Claude Code to update a research paper on vote by mail that he published five years earlier. The agent collected data, ran statistical analysis, generated tables and drafted a revised paper with minimal prompting. The result was impressive on the surface. But when Hall tasked a PhD student with auditing the work, gaps emerged. The agent missed data it should have captured and made coding mistakes that required expert-level oversight to correct.
The gap between capability and reliability highlights a persistent tension. Quentin Rousseau, CTO of incident management platform Rootly, describes the psychological cost of managing multiple fast-moving AI workstreams. Where a demanding workday produces the fatigue of effort, directing AI agents creates a different kind of exhaustion. "It's kind of like the difference between running a marathon and watching a really gripping TV series," he said. "One tires you out and the other keeps you up all night."
OpenAI co-founder Andrej Karpathy, now at rival Anthropic, recently told the "No Priors" podcast he had been in a "state of AI psychosis" since December while experimenting with what agentic tools could accomplish. The comment reflected both the intrigue and toll that supervising AI at scale can exact.
The competitive landscape shifted when Anthropic released Claude Code in October 2025, attracting non-developers at scale for the first time. Holiday downtime let casual users experiment. By January, Claude Code had gone viral. Anthropic then released Cowork, an office-focused variant of Claude itself. OpenAI followed with its desktop app in February, joining a market that has quickly matured from niche developer tools to office infrastructure.
OpenAI's strategy appears to be repositioning Codex less as a specialized coding tool and more as an operating system for knowledge work itself, weaving together the fragmented artifacts that define modern professional life.
Author James Rodriguez: "The growth is undeniable, but Hall's experience with Claude Code proves that AI agents still need humans watching closely, not managing them from a distance."
Comments