GitHub Copilot cloud agent review

GitHub Copilot's agents come in two forms. Agent mode runs in VS Code, Visual Studio, JetBrains, Eclipse and Xcode, editing files and running commands on your machine ^1,7. Copilot cloud agent, called Copilot coding agent until April 2026, takes an issue or prompt, works in the background in a GitHub Actions environment, and opens a pull request or leaves a branch for you to review first ^8,9. Pro+ and Max users can also hand tasks to Claude and OpenAI Codex agents from the same Agents tab ^1,20.

Packaging changed sharply in 2026. GitHub paused new self-serve Copilot Business sign-ups in April ¹¹, moved every plan to usage-based GitHub AI Credits on June 1 and added a $100 Max plan ¹⁰, then reopened sign-ups with stricter account vetting and upfront seat charges for card and PayPal customers from October 1 ^12,29. The billing switch drew loud complaints from developers ²⁴.

You assign an issue, mention @copilot on a pull request, or start a task from the Agents tab, GitHub Mobile, Linear, Slack or Microsoft Teams ^5,21. The agent runs in an ephemeral environment powered by GitHub Actions, which you configure with a copilot-setup-steps.yml file, larger runners or self-hosted runners ⁸. It checks its own code with CodeQL, secret scanning and an advisory-database check on new dependencies, and gets a second opinion from Copilot code review ⁶. Each session is capped at 59 minutes, one repository and one pull request ⁵.

Guardrails are strict. Only users with write access can trigger it, it pushes only to its own branch, Actions workflows wait for human approval, and the person who asked for the change cannot approve it ⁶. In the IDE, VS Code offers per-session permission levels, approval for sensitive files and agent sandboxing ¹⁵, plus checkpoints to revert a request ¹⁶. Enterprise admins can block or require approval for shell commands, file edits and network domains centrally ¹³, and local sandboxing became generally available in October 2026 ¹⁴.

Every plan now bills in GitHub AI Credits: one credit is $0.01, and each interaction costs the model's per-token rate ^3,4. Pro costs $10 a month with 1,500 credits, Pro+ $39 with 7,000 and Max $100 with 20,000; part of each allowance is a flex amount GitHub may change ^1,2. Business costs $19 per seat with 1,900 credits and Enterprise $39 with 3,900, pooled across the organization ^2,3. Extra usage is on by default for organizations, so admins should set budgets or switch it off ³. Inline suggestions stay unlimited on paid plans ³, and Copilot code review also uses GitHub Actions minutes ¹⁰. A daily agent user on Pro+ pays $39 for $70 of credits, and anything beyond that bills at $0.01 per credit ^1,3.

Copilot Business and Enterprise are covered by GitHub's SOC 2 Type 2 report and ISO 27001 certificate, and the Copilot trust center also lists ISO 42001, SOC 1, SOC 3, CSA STAR Level 2 and TISAX ¹⁷. GitHub does not train on Business or Enterprise data, but may train on Free, Pro, Pro+ and Max interaction data unless the user opts out ¹⁸. GitHub holds zero data retention agreements with OpenAI and Anthropic, and xAI models run under a zero-retention policy; Claude Fable models need a separate exemption ¹⁸. Enterprises on GitHub Enterprise Cloud with data residency can keep Copilot inference in the US or EU ¹⁹.

GitHub Copilot holds 4.4 stars from 399 G2 reviews ²⁷ and 4.4 from 478 ratings on Gartner Peer Insights ²⁸; both cover Copilot as a whole rather than the cloud agent alone. GitHub publishes Copilot results from Duolingo and Grupo Boticário, quotes Uber and Redfin, and shows FedEx, Shopify and General Motors logos on its Copilot page ^22,23. Independent coverage in 2026 focused on the billing backlash ²⁴, a breach of GitHub's internal repositories ²⁵ and repeated outages ²⁶.