I don’t need another model scoreboard. I need to know what I can type, speak, or click tomorrow morning without a special enterprise ticket. On the Codex side of DevDay 2026 (29 September 2026), that everyday surface is the refreshed Codex CLI plus code review in the ChatGPT desktop app.
This is companion coverage to the Codex Cloud piece. Same event, different question: what shipped for people who live in a terminal and a PR list.
Refreshed Codex CLI — what OpenAI spotlighted
From the official recap, the CLI refresh includes:
- Voice to start and steer tasks (community coverage calls it two-way voice — I’d confirm in the live CLI docs before promising “conversation quality”)
- A new
/agentsview to delegate and track multiple tasks - Better prompt editing, session resumption, worktrees, and a cleaner terminal UI for long sessions
Access: available on all plans, per the recap.
Docs: learn.chatgpt.com/docs/codex/cli
One caution I’d put in the notebook: secondary changelogs in the wider internet suggested some voice / agents-adjacent CLI pieces showed up in builds before Sept 29. Safe wording for us: DevDay bundled and spotlighted a refreshed CLI. I’d avoid “first-ever voice in Codex CLI” until we cite a dated changelog.
What I’d try in the CLI this week
- Update to the current Codex CLI from the official docs path — don’t invent install flags here.
- Run one short task with voice steering if your environment supports mic input cleanly.
- Open
/agents, park two tiny tasks (lint fix + README typo), and see whether tracking stays honest when one fails. - Resume a session after closing the terminal once. If resumption is flaky, write that down before trusting it on a merge-day branch.
Desktop code review — GitHub PRs and GitLab MRs
The other everyday ship: a new review experience in the ChatGPT desktop app. OpenAI says you get summaries, diffs, and a way to ask Codex about issues before you drop feedback on GitHub pull requests or GitLab merge requests. Automatic reviews can run a first pass in the cloud while you’re away.
Access: all plans, per the recap. Docs: learn.chatgpt.com/docs/code-review?surface=app
Quotas for auto-review aren’t on the recap. Same rule as Cloud: eligibility ≠ unlimited.
I’d use it as a first pass, not as the only reviewer. For a family-and-work week, the win is catching the obvious miss while I’m offline — then I still read the diff before I approve.
How this pairs with Codex Cloud
CLI and desktop review are the hands. Cloud is the room the agent can stay in. If you’re only adopting one habit this week, I’d pick automatic cloud review on one active repo plus one voice-steered CLI task — not every feature at once.
Watch the keynote cut
OpenAI didn’t ship a separate “CLI only” YouTube short that I trust as the primary embed for this angle, so I’d put the full DevDay 2026 keynote near the top and jump to the Codex developer tools segment when you watch.
Takeaway
If you’re on any paid ChatGPT plan that includes Codex surfaces, the DevDay question isn’t “did AGI arrive.” It’s whether voice + /agents + desktop review remove friction you already feel. I’d test that on a real PR by Friday, then decide what deserves a team default.
