GitHub Copilot's September Reset: Prepaid Seats, One Unified Agent Experience, Balanced Reviews by Default

GitHub published three separate Copilot changes on August 28, packing a policy shift, a product consolidation, and a review default change into one week of deadlines. None of them change list prices. All of them change how Copilot Business and Enterprise teams budget, where Copilot runs, and what a default code review costs.

The billing shift: seats are prepaid now#

New Copilot Business and Enterprise customers paying by credit card or PayPal can sign up again starting September 1. GitHub paused those signups earlier this year; the stated reason for the vetting and billing changes is "to improve availability and reliability of Copilot services," and the mechanism is prepayment. Every new seat assignment requires payment before the user gets access, and starting October 1 every assigned seat, including existing ones, is charged upfront at the start of the billing cycle.

Three details matter for anyone managing a seat roster:

• No prorated refunds. Revoking a seat removes it from the next cycle; it does not refund the current one. Mid-cycle seats are still prorated from assignment date.
• Overage gates access. If a team exceeds included usage, additional payment may be required for users to keep using Copilot. This is the strongest enforcement lever GitHub has shipped so far - usage-based billing from June allowed overage to accrue against budgets, while this change ties continued access to payment.
• Prices are not changing. The AI Credits model, spend controls, and additional credit purchases all stay as they are. See our usage-based billing guide for how that metering works.

The practical read: this is an anti-abuse and cash-flow move, not a price hike. For finance teams it means October invoices carry the full seat count upfront, so the budget impact lands a month earlier for anyone who added seats mid-cycle.

One Copilot, built on agent sessions#

The larger product signal is consolidation. No earlier than September 28, Copilot cloud agent, Chat on github.com, and Chat in GitHub Mobile are replaced by a single unified experience, enabled by default, with one policy. The cloud agent leverages Sandbox for a faster cloud experience, and github.com chat fully migrates to the agent sessions model that the cloud agent already uses.

Two consequences are easy to miss:

• Chat data is retained for the life of the account instead of 28 days. That aligns github.com and Mobile with the cloud agent's retention. For teams with data-retention obligations, this is a policy change worth reading before opting in - the same class of consideration zero-data-retention model limits already introduced for premium models.
• Opting out means losing access. If you or your teams opt out of the unified experience, Copilot disappears from github.com and GitHub Mobile after launch. There is no "old chat, new agent" split to retreat to; agent sessions become the only surface.

Admins should review the single policy before September 28 and set it explicitly. GitHub's settings path is Copilot settings on github.com -> Copilot cloud agent (coming soon).

Balanced becomes the review default#

Since effort levels went GA on August 7, Lite has been the default review mode. From September 28, repositories and organizations whose setting is "Default" will get Balanced: a higher-reasoning model that spends longer on repository context, and burns more AI credits and Actions minutes per review, as our effort-levels breakdown documented at GA.

Keeping Lite is explicit work: teams that want it must change their repo or org setting away from Default before September 28. For organizations that auto-request reviews on every pull request, the silent flip multiplies review spend on routine changes. GitHub recommends Balanced for complex logic and security-sensitive code - the risk is that "Default" now applies that depth everywhere unless someone decides otherwise.

What this tells us#

GitHub is finishing Copilot's transition from a chat and completion product into an agent platform with three consistent moves: one runtime (agent sessions everywhere, sandbox-backed), one billing posture (prepaid, usage-metered, enforcement-backed), and one review default (deeper). That follows the governance layering of the summer: enterprise team model policy targeting, MCP allowlists, and the impact dashboard's ROI section. The direction is consistent - as agents get more autonomous, GitHub gives admins more precise levers, then biases the defaults toward the capable option.

Three action items before the deadlines:

1. Review the unified policy and your model, MCP, and data-retention stance before September 28.
2. Confirm card details and seat counts before the October 1 prepaid cycle.
3. Set explicit review effort defaults now if you want Lite to stay Lite.

The Copilot agent traces production-scale analysis shows what these sessions look like under real load - the unified experience just makes that the only shape Copilot has.

Continue Reading#

• Copilot code review effort levels, explained - what Lite and Balanced cost, and how to set defaults
• GitHub Copilot usage-based billing, explained - how AI Credits metering works underneath the new prepaid seats
• Enterprise team model policy targeting - the admin levers that pair with the unified policy
• Copilot agent traces at production scale - what agent sessions do under real load
• The Copilot impact dashboard ROI section - how GitHub is measuring the spend it is now collecting upfront

Sources#

• Upcoming changes to GitHub Copilot policies and billing - GitHub Changelog, August 28, 2026
• GitHub Copilot Quick Start - GitHub Docs
• About GitHub Copilot code review - GitHub Docs