OpenAI bets ChatGPT Work can bring AI agents to non-coders

OpenAI is trying to extend the AI-agent model that transformed software engineering to everyone else. ChatGPT Work, released last month on the company’s lowest subscription tier at $20 a month, is a modified version of OpenAI’s Codex coding tool aimed at letting white-collar workers — accountants, investors, doctors — hook AI agents into their daily digital workflows, connecting to email, Slack, calendars, and apps like Notion and Figma. “If I’m asking it to write a document, is there a possibility that it’s going to pull from a private DM on that subject and not know that it’s not supposed to share some info? Yes,” said Andrew Ambrosino, lead engineer for OpenAI’s desktop app, who gives the tool access to his own inbox, Slack, and phone. “I’ll do it for the job. I will take the personal hit here and there if I have to. And I haven’t had to.”

Reaching non-engineers matters commercially: longer agent runs burn more tokens, and OpenAI needs professions beyond coding to justify the scale of its compute investment. An OpenAI-backed study found that in June, 98% of OpenAI’s own employees were using Codex, but just 17% of organizational subscribers and less than 1% of individual subscribers were using the agentic coding tool — a gap the company is trying to close with Work. OpenAI wouldn’t say how many people use Work versus Codex specifically, but the combined app has 20 million users, compared with more than a billion people OpenAI says are prompting ChatGPT overall.

Making the tool usable for non-technical workers required rethinking its interface: giving an agent read-only access to a cloud drive, for instance, proved confusing, and some settings are only configurable on the web app rather than mobile. Cost is also a live issue — one journalist testing the app used more than 80 million tokens in four days, at a cost of about $65 according to the model’s own analysis, more than three times the $20 subscription price for that period. OpenAI’s Thibault Sottiaux pointed to a recent 80% price cut on the company’s Luna model as evidence the gap will narrow: “If you wake up six months from now, you should be able to do all of the same [tasks] with less spend.”

Why it matters

Whether ChatGPT Work succeeds hinges on a harder problem than coding tools solved: unlike software, which either works or doesn’t, most white-collar output — a presentation, a strategy memo, a sales pitch — has no clean way to measure if an agent got it right, making trust and adoption a slower climb than in engineering.