GitHub has announced three Copilot changes that may look administrative but directly affect budget, privacy, and work continuity: upfront seat billing for some Business and Enterprise customers, a unified agent and chat experience, and Balanced becoming the default effort level for code review.
The key dates are September 28, when the experience and review changes begin, and October 1, when the billing behavior starts reaching existing customers who pay by credit card or PayPal. Plan prices are not changing. The way access, usage, and default settings are controlled is.
Upfront seats and usage-dependent access
According to the official GitHub changelog, new Copilot Business and Enterprise seats must be paid before a user receives access. For eligible existing customers, every assigned seat will generate an upfront charge at the start of the billing cycle from October 1.
Revoking a seat during the month does not generate a prorated refund; removal is reflected in the next cycle. GitHub also says that when included usage is exceeded, additional payment may be required for users to continue using Copilot.
This changes operational risk. Budget is no longer only a discussion after the invoice arrives. It can affect tool availability. Teams using agents for long-running work need to monitor consumption before they reach a limit, not after it.
One experience, with longer retention
No earlier than September 28, Copilot cloud agent, Copilot Chat on github.com, and chat in GitHub Mobile will begin converging into one experience and one policy. Moving github.com to agent sessions also changes chat retention: data will be kept for the life of the account instead of 28 days, matching the existing cloud-agent behavior.
The new model will be enabled by default. Organizations that opt out will lose Copilot access on github.com and GitHub Mobile. The official text does not say that every Copilot surface will automatically be disabled, but this is already enough to require a security, privacy, and records-management review.
For companies with data-minimization policies, the question is not only whether the feature helps developers. It is who can access that history, how long it should exist, and what information should never enter a session.
Balanced becomes the code-review default
On the same date, repositories and organizations that keep Default selected for Copilot code review will move from Lite to Balanced. Balanced performs a deeper review and can change the usage associated with the process.
Teams that want to preserve the current behavior must explicitly select Lite before the change. It is a useful example of cost by default: no team makes a new active decision, but the behavior changes anyway.
Copilot now requires AI FinOps discipline
These changes reinforce a broader shift: agentic development tools no longer fit a simple “price per license” model. A license opens the door, but task intensity, review level, model selection, and agent volume determine real consumption.
Engineering leaders should take four actions before the cutover dates:
- compare contracted limits with actual usage from recent weeks;
- define alerts and owners for approving additional capacity;
- review unified-experience retention with security and privacy teams;
- deliberately choose Lite or Balanced for code review.
The point is not to leave Copilot because billing changed. It is to leave behind the assumption that a subscription alone guarantees predictability. When agents perform more work and configuration changes consumption, cost governance becomes part of engineering-platform architecture.