OpenAI Repurposes Codex to Bring Agents to Office Workers

OpenAI's newest product doesn't look like a chatbot at all. ChatGPT Work, released last month at $20 a month, the company's lowest subscription tier, is built on a modified version of Codex, the coding assistant originally designed for software engineers.

The twist is who it's for. Rather than helping developers write code, ChatGPT Work repurposes Codex's agentic scaffolding for accountants, investors, doctors, and other professionals who have never touched a terminal.

The idea is to let non-technical workers direct AI agents through the digital workflows that define their jobs: pulling data, cross-referencing documents, executing multi-step tasks inside the software they already use. That's a meaningful jump from a chat window that answers questions to a system that acts on a user's behalf across an entire workday.

Thibault Sottiaux, who leads OpenAI's core product organization, oversees Work as part of that broader push. Andrew Ambrosino, the lead engineer on OpenAI's desktop app, is building out the technical surface that lets these agents operate outside the browser.

A crowded field of specialists

OpenAI isn't entering an empty market. Startups like Harvey, built for law firms, and Clay, aimed at sales teams, have already staked out vertical niches with model-agnostic products that can swap in whichever LLM performs best for a given task.

That positioning cuts against OpenAI's approach. Where Harvey and Clay sell deep, industry-specific tooling built around customer workflows, ChatGPT Work bets on a single horizontal product that flexes across accounting, medicine, and finance alike, all running on OpenAI's own models.

ChatGPT Work aims to give white-collar workers AI agents that can hook LLMs up to digital workflows used by accountants, investors, doctors, and other professionals.

The wager is that broad reach and OpenAI's existing user base outweigh the specialization advantage of vertical competitors. Whether generalist agents can match purpose-built tools on accuracy and workflow depth inside any one profession is the open question the product now has to answer in the market rather than in a pitch deck.

Pricing the product at OpenAI's entry-level tier also signals an attempt at fast, broad adoption rather than a premium enterprise rollout, a contrast to how Harvey and similar vertical players typically price for large institutional clients.