GitHub Copilot agentic engineering guide: plan, build, verify and observe
GitHub published a guide to agentic engineering with Copilot across Plan, Build, Verify and Observe, citing 1.2 billion weekly commits and longer PR reviews.
GitHub published a developer guide on November 6 to what it calls “agentic engineering” — using GitHub Copilot across four stages of software work: Plan, Build, Verify and Observe. The post, by Nick Liffen and Philip Holleran, argues that faster code generation has shifted the bottleneck rather than removed it.
The company’s own numbers make the case. GitHub says it saw more than 1.2 billion commits in one week this October, over 400% more than a year earlier and more than GitHub received in all of 2025. The post then asks developers whether they feel 400% more productive.
Review is where the pressure shows, according to GitHub. It cites a Faros AI study finding that pull requests spent 441% longer in review when teams moved from low to high AI adoption.
What each stage involves
In the Plan stage, GitHub suggests using Copilot in Slack or Teams to gather requirements, having it draft GitHub Issues with acceptance criteria, and using GitHub Agentic Workflows to label issues and estimate complexity so low-complexity tasks can be assigned to Copilot. Plan Mode is pitched for comparing approaches before implementation starts.
For Build, the post describes session continuity across clients, cloud sandboxes that keep execution off your machine, and isolated repository copies per session in the Copilot app. It also mentions repository instructions, shared memory, and Copilot Dreaming, which reviews eligible sessions and pull request feedback to keep memories current. Copilot Fusion is described as letting you choose models and reasoning depth per task.
On Verify, GitHub recommends stacked pull requests to keep changes small, GitHub Code Quality to catch routine issues before human review, and Copilot code review with analysis depth matched to the change. It also lists AI Scan for vulnerabilities beyond CodeQL’s coverage, AI secret detection for unstructured secrets, license compliance checks, and a GitHub package firewall currently in public preview for npm. Agentic Autofix and Agent Merge handle repairs, failing checks and merge conflicts, and Copilot can merge pull requests once verification passes if you configure it.
In Observe, Copilot can pull from observability tools and business data through connectors such as Fabric to help investigate which requests or customer groups are affected, draft an issue with evidence and a regression test, and surface suggested improvements from Dreaming.
GitHub’s advice for getting started is to pick one piece of work already on your plate and put an agent on it, then expand over time. The topic was covered live at GitHub Universe 2026 in a session titled “Agentic Engineering: Past the Pilot.”
Related reading
Sources
Some links on this page may be affiliate links. If you buy through them we may earn a commission at no extra cost to you. See our affiliate disclosure.