What it does
Two autonomous features on github.com. The coding agent takes an issue or task, works in its own cloud environment and opens a pull request. Copilot code review reviews pull requests, flags issues and suggests fixes.
Key facts
- Coding agent went GA on 25 September 2025; Copilot code review went GA in April 2025.
- You hand the agent work by assigning it an issue, using the agents panel on github.com, or the "Delegate to coding agent" button in VS Code.
- The agent runs in its own dev environment via GitHub Actions, pushes to a branch and opens a draft PR, then requests your review when done.
- You iterate by leaving PR comments; the agent picks them up and pushes further changes.
- Code review can run automatically per repo or org via branch rules ("Automatically request Copilot code review"), with an optional "review new pushes" setting.
- Automatic review is available on Pro, Pro+ and Max individually, and through Business/Enterprise org configuration.
- On Business/Enterprise, an admin must enable the coding agent on the Policies page before anyone can use it.
- You can now request a Copilot review from the terminal via GitHub CLI (gh pr create / gh pr edit), added March 2026.
When to use / skip
The coding agent earns its keep on well-scoped, low-ambiguity issues — dependency bumps, test coverage, small features with clear acceptance criteria. Give it a vague epic and you'll get a plausible-looking PR that misses the point. Code review is a good first-pass filter that catches the obvious before a human looks, but it doesn't replace a senior reviewer on anything that matters. Use both as a funnel, not a rubber stamp.
Configuration decisions
- Decide whether the coding agent is enabled at all, and for which repos/teams — it's off until an admin turns it on.
- Choose the agent's model per task in the coding-agent UI; it does not inherit your IDE chat model.
- Decide whether code review is manual (requested per PR) or automatic via branch rules, and whether it re-reviews new pushes.
- Set branch protection so agent PRs still require human approval and passing CI before merge.
- Work out GitHub Actions cost and runner capacity — the agent consumes Actions minutes.
Gotchas
- The agent opens a draft PR and runs in Actions; runaway or repeated runs burn Actions minutes and AI credits.
- Its model is chosen separately in the coding-agent UI, so don't assume it matches your editor selection.
- Content exclusions have not applied to cloud agents (as of May 2026) — the agent may read paths you excluded elsewhere.
- Automatic code review on every push gets noisy fast; scope it to the branches that warrant it.
- Review quality improved sharply after the March 2026 agentic overhaul — older impressions of it being "barely useful" are out of date.
- Treat agent PRs as untrusted input until reviewed — a confident diff is not a correct one, and it shouldn't auto-merge.
Consultant notes
- Pitch the coding agent as a junior teammate for the backlog's tedious end, not a replacement for engineers. The review burden shifts to humans, so staff for that.
- Code review is the easier sell and the safer first step — low risk, immediate value, no autonomous merges.
- Watch Actions and credit spend once the agent is live; unbounded delegation is where costs surprise people.
- The admin-gated enablement on Business/Enterprise is a governance feature, not a nuisance — use it to phase rollout by team.
Confirm branch protection blocks auto-merge of agent PRs before you let a client switch the coding agent on.