OpenAI folded three separate products into one desktop app on 9 July 2026: ChatGPT, the Codex coding agent, and the Atlas browser. Inside the unified app, three modes sit side by side — a standard Chat mode, a new Work mode for office-style agentic tasks, and Codex, which OpenAI says is unchanged from the standalone coding agent developers already use. All three are available on every subscription tier, including free. The standalone Atlas browser is being retired outright: it shuts down completely on 9 August 2026, with its browser-based agent capabilities absorbed into the new app’s built-in browser.
The practical effect is that OpenAI’s coding agent no longer requires a separate install, a separate subscription decision, or even a paid plan to try. That’s a meaningfully lower barrier than Codex carried a month ago, at a point when the “which AI coding agent” conversation — Claude Code, Cursor, Copilot, Windsurf, Codex — is already crowded and shifting fast (see this week’s other signal on GitHub Copilot’s own model swap).
So what
If your team standardised its AI coding tooling six months ago and hasn’t revisited the decision, both of this week’s developments are reasons to: Codex is now a zero-friction trial on any ChatGPT plan, and Copilot is about to run on a different model entirely. Neither of those facts should drive a tooling switch on its own — but a development partner that’s actively tracking which tools are worth evaluating, rather than defaulting to whatever they picked a year ago, is the difference between AI-assisted development that compounds and AI-assisted development that quietly stalls. That evaluation work is part of how we run projects — see AI-assisted development or get in touch to talk it through.