GitHub's guide to agentic engineering past code generation

GitHub's argument in this guide, written by Nick Liffen and Philip Holleran, is that writing code is no longer the slow part of software work, so agents have to help with the stages around it. GitHub says it saw more than 1.2 billion commits in one week this October, over 400% more than a year earlier, and it cites a Faros AI study that found pull requests spent 441% longer in review when teams went from low to high AI adoption (GitHub on X). The guide treats that review queue as the cost of faster code generation, then walks through four stages of the work, Plan, Build, Verify and Observe, naming the Copilot features GitHub offers for each.

Verify gets the most concrete treatment, because that is where the 441% shows up. The advice is to shorten the wait from both ends. Authors split a large change into stacked pull requests, a chain of small dependent pull requests that can each be reviewed on its own, and automated checks clear routine problems before a person opens the diff. The tools named for that second part are GitHub Code Quality, which checks changes against coding practices and the organization's standards; Copilot code review, which applies lighter or deeper analysis depending on the change; AI Scan, which looks for vulnerabilities in languages and files beyond CodeQL's coverage; license checks on new dependencies; and Agent Merge, which addresses failing checks, review comments and merge conflicts. If a team configures it, Copilot will merge the pull request once the verification steps pass.

The other stages follow the same pattern. For planning, Copilot in Slack or Teams pulls scattered context together and turns agreed requirements into GitHub Issues with acceptance criteria, Agentic Workflows label issues and estimate their complexity so simple tasks can go straight to Copilot, and Plan Mode produces a plan the developer approves before implementation starts. For building, Copilot runs tasks in cloud sandboxes, the Copilot app gives each session its own isolated copy of the repository, and repository instructions and shared memory carry project conventions between tasks; a feature called Dreaming reviews past sessions and pull request feedback to keep those memories current. For production problems, connectors to observability tools let Copilot help find which requests or customers a release affected and draft an issue with a regression test. The Copilot app and a deeper Max tier of Copilot code review were presented at GitHub Universe in late October (Universe announcement). The guide is a product walkthrough rather than a measured result: it does not report whether teams using these features got their review time back.