GitHub Copilot Is Becoming One Agent Policy Surface
GitHub's late-August Copilot updates unify cloud agent, web chat, and mobile chat policies while expanding Copilot code review to bot-authored and very large pull requests.
GitHub published two Copilot changelog entries this week that look administrative at first glance. They are actually a useful signal about where agentic developer tools are going: away from isolated product surfaces and toward one managed system with shared policies, shared billing controls, and review behavior that assumes agents are part of the normal development loop.
On August 28, 2026, GitHub announced upcoming changes to Copilot policies and billing for Business and Enterprise customers.1 The most important product change is that Copilot cloud agent, Copilot Chat on github.com, and Copilot Chat in GitHub Mobile are planned to converge into a single Copilot experience and policy no earlier than September 28, 2026.1
That matters because these surfaces used to feel like related but separate entry points: a chat box on GitHub, a mobile assistant, and a cloud agent that could work on repository tasks. GitHub is now saying administrators should think about them as one policy-managed experience. After the relaunch, the unified Copilot experience will be enabled by default, and Copilot on github.com will migrate fully to the agent sessions experience that had previously been used only by Copilot cloud agent.1
There is a data-retention implication too. GitHub says chat data on github.com will be retained for the life of the account instead of 28 days, aligning it with the existing Copilot cloud agent behavior.1 That is not a small detail. Agent sessions are no longer treated like ephemeral autocomplete context. They are becoming durable work records that administrators need to understand, govern, and explain to teams.
The same changelog also sets dates for billing changes. Starting September 1, 2026, GitHub plans to reenable sign-ups for new Copilot Business and Enterprise customers paying by credit card or PayPal, with stronger account vetting and updated billing behavior.1 New seat assignments will require payment before users gain access, and existing credit card or PayPal customers should expect upfront seat charges at the start of the next billing cycle beginning October 1, 2026.1 GitHub says Copilot Business and Enterprise prices are not changing, and that spend controls, usage tracking, and the option to purchase additional AI credits remain available.1
The second update landed on August 27, 2026 and focuses on Copilot code review.2 Copilot code review can now review pull requests that it did not fully cover before: automatically requested reviews on bot-authored pull requests, including Copilot cloud agent pull requests, and very large pull requests.2
The bot-authored pull request support is especially important. If coding agents are opening pull requests, the review system has to handle work created by non-human accounts. GitHub says organizations can use the policy that allows members without a Copilot license to use Copilot code review, with usage billed directly to the organization.2 For pull requests opened by Copilot cloud agent, GitHub says Copilot code review can now provide a full agentic review instead of falling back to a limited experience.2
The size limit change is also practical. GitHub says Copilot code review previously had a limit of 300 files or 20,000 lines of code, and that this limitation no longer applies.2 That does not mean giant pull requests are suddenly a good engineering habit. It does mean the review tooling is being adjusted for the kind of mechanical, broad, and sometimes messy changes that agents are likely to produce: migrations, dependency updates, codemods, generated files, and cross-repository maintenance work.
GitHub also added resolution reasons for Copilot review comments: "Addressed", "Won't fix", or "Incorrect".2 That gives GitHub structured feedback about whether a Copilot finding was useful. In other words, review is becoming a feedback loop, not just a comment stream.
Put together, these changes show a sober direction for agentic coding. The interesting work is not only making agents write more code. It is making agent sessions governable, billable, reviewable, and understandable inside existing software delivery systems.
For engineering teams, the takeaway is simple: agent adoption is becoming an administration and process question, not just a developer preference. Before September 28, Business and Enterprise admins should review Copilot policies, retention expectations, code review defaults, and budget controls. The agent is moving into the workflow. The workflow needs to be ready for the agent.