OpenAI Codex release notes: app 26.602 shows usage

This OpenAI Codex update is smaller than Sites or CLI 0.137.0, but it is useful. Codex app 26.602 makes agent work easier to review after the fact: activity insights and share cards now sit in the Profile section, while Computer Use startup and appshot error reporting get cleaner.

For small teams, that matters. Once Codex becomes part of weekly work, you need to see how it is being used, not just feel that it is being used.

OpenAI Codex release notes: what 26.602 changes

Codex app is OpenAI's desktop environment for running coding agents against local projects, threads, and review flows. A coding agent is an AI assistant that can read code, propose changes, and sometimes use tools under human control.

In the 2026-06-04 changelog, OpenAI lists three changes worth tracking:

• Activity insights and share cards were added to the Profile section. Users can review Codex usage highlights and save a profile card. Sharing is available on consumer ChatGPT plans.
• Computer Use startup readiness and appshot error reporting improved. Computer Use means Codex can work in desktop apps, while an appshot is Codex's way to capture context from an app window.
• Browser and review surfaces received UI fixes, including fullscreen browser composer controls, hex color swatches, terminal scrollbar alignment, and animated diff stat alignment. Onboarding now has more role choices too.

Source: OpenAI Codex changelog, 2026-06-04, Codex app updates 26.602. Appshots are described in the OpenAI Codex changelog, 2026-05-21.

Why this matters for Nordic teams

This is not a big model release. It is closer to a small operations view for Codex usage. That kind of view helps when a team moves from "we try Codex sometimes" to "Codex should help us every week".

Three questions become easier to ask:

• Are we using Codex for the right work, or mostly as an expensive chat tab?
• Do we see repeated patterns that should become skills, plugins, or clearer AGENTS.md rules?
• Do some repos, environments, or change types need more human review?

This is where Tool Forge fits. Not as another big platform, but as a way to make Codex work measurable: scoped API keys, secrets in a secret manager, least privilege, approval gates before deploy/auth/payment work, and logs that show what the agent actually did.

Short example: use the new Codex feature

Human step: open the Profile section in the Codex app and review your activity insights or usage highlights. If you share anything, note that OpenAI only lists sharing for consumer ChatGPT plans in this release.

Source: OpenAI Codex changelog, 2026-06-04: activity insights and share cards are in the Profile section, with sharing on consumer ChatGPT plans.

Then paste the summary into Codex or the AI agent you use for work planning:

Here are my Codex activity insights from the Profile section: [paste short summary]. Compare them with our three most important repo or automation goals for this week. Suggest two changes to how we use Codex, one thing to measure next week, and where human review is needed. Do not change files.

Good output should:

• connect usage to concrete repo or automation goals
• separate better practice from simply using Codex more
• flag where approval gates or review are needed
• suggest one simple metric for next week

What to test today

Pick one repo where Codex already helps. Look at the usage highlights, compare them with this week's goal, and write one small rule: "Codex may do X, but should ask before Y." That is often enough to make the next agent run safer without slowing the work down.

If you already connect Codex to more tools, make the check more concrete: which keys live in environment variables, which secrets live in a secret manager, which permissions are scoped, and which agent runs end up in a log a human actually reads?