August 24, 2026: Copilot moves deeper into Slack and Teams

Today is August 24, 2026, and GitHub’s freshest signal is operational rather than theatrical.

The biggest thing to know is the postmortem for the August 17 outage. GitHub says the incident lasted 7 hours and 47 minutes and affected GitHub.com services including Issues, Pull Requests, APIs, Actions, and Copilot. The company says the outage was made worse when delayed Copilot token responses triggered a retry bug in VS Code, which amplified traffic instead of letting the system calm down. In plain English, retries are supposed to help when something is slow or flaky, but if too many clients retry too aggressively, the extra traffic can make recovery harder. GitHub says that is now one of the areas it is fixing with retry limits, retry budgets, and variable timeouts.

That matters because it shows GitHub’s current risk is not just raw capacity. It is also the interaction between AI features, client behavior, and backend recovery. GitHub says it has already added more than 3 million CPU cores, 120 petabytes of high-speed storage, and significant network capacity, and that Azure now serves roughly 58% of platform load and half of Git operations. That is a concrete scale-up response, not just a promise. But it also means the company is still working through how to harden the platform at a very large footprint.

The second notable move is product expansion for Copilot in the places teams already work. On August 21, GitHub put the new Copilot experience in Slack into public preview for Copilot Business and Enterprise organizations. Users can mention @GitHub to start an agent session that can answer questions, triage bugs, update or create issues, investigate failures, and open pull requests. GitHub says usage counts against existing Copilot entitlements and cloud agent budgets, so this is not just a feature demo. It is also a governed consumption surface.

GitHub made a similar move in Microsoft Teams. Shared agentic work with GitHub Copilot in Teams is also in public preview, and GitHub says Copilot cloud agent sessions consume AI credits while cloud sandbox usage is billed separately. That split is important for enterprise buyers. It means the company is making the spending model explicit enough that admins can reason about budget, policy, and where a session actually uses resources.

A practical reading is that GitHub is turning Copilot from a code assistant into a coordination layer. Instead of living only in the editor, it can now start from chat, inspect work, and then move into issues or pull requests. That broadens the surface area where GitHub can become the default workflow system for software teams.

There were also a couple of smaller but still useful workflow updates. On August 20, GitHub said saved views can now be pinned to the repository issues sidebar in Issues and Projects, and that the blocked_by, blocking, and relates_to REST endpoints now behave more narrowly based on token scope. The first part is a small productivity gain. The second is the more interesting platform detail, because it is a least-privilege improvement. In other words, the API now exposes less than before unless the token really needs broader access, which is the kind of change enterprise administrators usually want.

On the reliability side, GitHub Status also closed a few recent incidents. The Copilot model incident involving Fable 5 was resolved on August 13 after an upstream provider issue, and GitHub advised users to choose another model or Auto while it was degraded. The August 18 runner-group and runner-related permissions incident was also resolved, with GitHub saying the root cause was an expired authentication certificate in backend requests. One correction worth noting: GitHub later said the August 18 Actions incident was posted in error and that Actions workflows were not impacted. So the open question there is not whether workflows failed, but how GitHub distinguishes between a UI permissions issue and an actual execution problem in its incident labeling.

So the scorecard for this week is straightforward. GitHub is widening Copilot’s reach into Slack and Teams, tightening some Issues and Projects and API controls, and still dealing with the reliability lessons from a major outage that involved retry behavior as well as capacity. The company is clearly treating these as connected systems, not separate products.

What to watch next: whether GitHub publishes a fuller root cause update on the Copilot Cloud Agent visibility delay and task-start issue, whether the August 26 Copilot default-availability policy shows up in production or UI behavior, and whether the September 1 model deprecations leave the expected set of Copilot models available.