Copilot Automations can now learn from outcomes and propose their own prompt changes
GitHub's Copilot Automations can now use signals like merged PRs and accepted comments to propose prompt changes, which you approve before they apply.
GitHub's Copilot Automations can now learn from their own outcomes and propose changes to their prompts, according to a GitHub changelog post dated October 28, 2026. The signals are ones your repo already produces. A triage automation can learn from whether people change the labels it applied. A review automation can learn from whether developers accept its comments. An optimization automation can learn from whether maintainers merge its PRs. You review every proposed change before it's applied. For builders, this automates a prompt-tuning loop you'd otherwise run by hand.
When you ask Copilot to set up a self-improving automation, GitHub says it can create a group of automations. For a custom code reviewer, that might be a review automation that runs when a PR opens, plus a weekly improvement automation. The weekly job analyzes how developers responded and proposes changes to the reviewer's prompt, tools or triggers. GitHub notifies you, and you accept or reject each change. The feature is available on Copilot Pro, Pro+, Max, Business and Enterprise.
The launch came alongside other Automations updates at GitHub Universe, The New Stack reports. One of those updates adds chained automations and AI credit limits per run, day, week or month. That matters because GitHub hasn't said what a weekly improvement run costs. Shared automations bill credits to the repository owner and keep a change history. In VS Code, cloud automations need version 1.144 and the chat.automations.cloud.enabled setting.
Several things are still unclear. GitHub hasn't documented how the improvement job handles noisy signals, such as comments accepted just to clear them. It also hasn't said how much signal a small repo needs, and no independent hands-on tests have appeared yet. If you only need review, CodeRabbit's Essentials plan lists "learning from your feedback," per a review by its competitor cubic, so weigh that source accordingly.
Sources
- Copilot Automations now learn from outcomes and self-improve - GitHub Changelog · The GitHub Blog
- GitHub debuts Copilot Fusion, a new model selection & orchestration system that beats Claude Code for half the API cost · The New Stack
- Build shared workflows for your team with Copilot Automations - GitHub Changelog · The GitHub Blog
- Create and manage Copilot Automations in Visual Studio Code - GitHub Changelog · The GitHub Blog
- New tools, triggers, and cost controls for Copilot Automations - GitHub Changelog · The GitHub Blog
- CodeRabbit review: pricing, setup and how it compares | cubic · Cubic