GitHub published a developer guide on November 6 describing how teams can use Copilot across four stages of agentic engineering: plan, build, verify and observe. The post, by Nick Liffen and Philip Holleran, ties into GitHub Universe 2026.
The premise is that AI has accelerated writing code while planning, review and operations have lagged. GitHub says it saw more than 1.2 billion commits in a single October week, up over 400% year on year. It also cites a Faros AI study finding that pull requests sat in review 441% longer as teams moved from low to high AI adoption.
For planning, GitHub points to Copilot in Slack or Teams, issues generated with acceptance criteria, and agentic workflows that label issues and estimate complexity so simple tasks can go straight to Copilot. Plan Mode compares approaches and waits for human approval before implementation begins.
On the build side, cloud sandboxes keep tasks running after a laptop is closed, and each session gets an isolated copy of the repository. Repository instructions and shared memory carry project conventions, while a feature called Dreaming reviews past sessions and PR feedback to propose memory updates. Model and reasoning choices are handled through Copilot Fusion, and enterprise managed settings spread company policy across Copilot surfaces.