Establishing Code‑Review Etiquette When GitHub Copilot Is a Reviewer

[ai-mindset]

Our team recently debated how to handle pending Copilot comments during a PR review, following a question @yiwen-h asked. To our knowledge, there is no documented guidance on this scenario from other organisations, and public references can be sparse at the time of writing.

Goal
Create a concise blog post that captures our emerging consensus and gives teams a practical framework for handling Copilot feedback. This blog post could also be integrated into the team's work manual.

---

Proposed Convention

Situation	Recommended Action	Rationale
PR opened, Copilot added comments	The PR author should address all Copilot comments.	Guarantees a clean PR, giving the author the chance to vet AI suggestions.
Human reviewer starts reviewing while Copilot comments are pending	Review concurrently. If a human comment overlaps with a Copilot comment, the reviewer should explicitly state it e.g. “I agree with Copilot’s suggestion”.	Reinforces important issues, prevents duplicated discussion.
No pending Copilot comments	The human reviewer becomes the sole blocker. Their review must be completed before the PR can be merged.	Ensures human judgement is always involved.
Pending Copilot comments still unaddressed	The PR author is the blocker. The human reviewer may: continue reviewing, or unassign themselves and add a comment such as “Please address Copilot’s feedback and re‑assign me when done”.	Author remains responsible, maintains clear communication.
PR needs more time for addressing Copilot feedback	Convert the PR to Draft until all Copilot comments are resolved.	Signals that PR is not yet ready, avoids overwhelming reviewers.
Copilot feedback appears after a review has begun	Keep the PR open. The author addresses the new feedback, then the human reviewer completes any remaining items.	Avoids unnecessary close‑reopen cycles, preserves review continuity.

[yiwen-h]

Thanks for doing this Eirini. I feel like this, whilst very thorough, somewhat overcomplicates the situation a little in the verbose way that AI often does

Could we simplify it a little? the key is just communication and being clear on who is waiting for what

[ai-mindset]

Thanks for your input @yiwen-h. I removed some of the potential "noise". However, the table attempts to explicitly capture the discussion we had on Slack (which will be hidden in 84 days). You posed an important question that should be addressed in more detail in our work manual and through a blog post, for clarity.
The Github issue was hand-written and then improved with Copilot (bullet points -> table, better language), to avoid uncurated AI slop.
Does the table capture our discussion addressing your original question?

[yiwen-h]

Makes sense - thank you so much for revising