OpenAI is building AI agents for everything. Will everyone use them?
Get the Tech newsletter
Daily tech — startups, AI labs, chips, the launches that shape the next decade. Free.
- OpenAI released ChatGPT Work last month at its lowest $20/month tier, built on a modified version of Codex and aimed at accountants, investors, doctors, and other white-collar workers beyond software engineering.
- An OpenAI-backed study found 98% of OpenAI employees were using Codex in June, but just 17% of organizational subscribers and under 1% of individual subscribers were — the gap that the Work product is designed to close.
- The joint Codex/Work app reaches 20 million users, compared with the 1 billion users OpenAI says prompt ChatGPT online.
- ChatGPT Work links agents to email, Slack, Notion, Figma, and other SaaS tools; lead engineer Andrew Ambrosino told TechCrunch he grants the app access to his own inbox and Slack despite acknowledging the risk that private DMs could leak into outputs.
- Vertical-specific competitors Harvey (legal) and Clay (sales) are taking a model-agnostic approach, plugging in whichever model works best — a strategy that, per a16z's Christian Catalini, could capture value if the major labs can't acquire the complementary assets needed to scale agents themselves.
- Hands-on testing surfaced real friction: cloud-drive permission setup defaulted to full access only, key settings live exclusively on the web app, and the product can create Google Calendar events but not new calendars.
- Sam Altman is among the early users, with OpenAI saying he has used the tool to plan vacations, while employees use it for weekly metrics reports and VCs are using agents to assemble investment memos.
Why it matters: Agents that run longer burn more tokens, so per-user revenue scales with how broadly non-developers adopt them — making the <1% individual-subscriber usage figure a direct commercial problem for OpenAI's model-agnostic competitors like Harvey and Clay to exploit if it persists.
Ask SkimNews
