Beyond code generation: GitHub's guide to agentic engineering
GitHub says faster code generation moved the bottleneck to planning, review and production debugging, and proposes an agentic workflow across four stages. The post pairs striking numbers, 1.2 billion commits in one week and 441% longer pull request review times under heavy AI use, with concrete Copilot features for each stage.
Key points
- GitHub measured over 1.2 billion commits in a single week this October, more than double-digit growth over a year earlier and more than all of 2025.
- A Faros AI study cited in the post found pull requests wait 441% longer in review when teams shift from low to high AI adoption.
- The suggested workflow spans four stages: Plan, Build, Verify and Observe, each with its own Copilot surface.
- Plan Mode lets you compare approaches and approve a plan before Copilot starts implementing; GitHub Agentic Workflows can label and triage issues automatically.
- Copilot runs tasks in cloud sandboxes with isolated repository copies, and session continuity carries the same conversation across clients and devices.
- On the review side: stacked pull requests, Copilot code review with adjustable depth, AI Scan and AI secret detection, Agentic Autofix and Agent Merge.
GitHub argues that faster code generation has shifted the bottleneck to planning, review and production debugging. The post cites 1.2 billion commits in a single week in October — up over 400% year on year — and a Faros AI finding that pull requests sat in review 441% longer as teams moved from low to high AI adoption. Its answer is an agentic workflow across four stages: Plan, Build, Verify and Observe.
Concretely, that means using Copilot in Slack or Teams to gather requirements, Plan Mode to agree on an approach, cloud sandboxes and session continuity to keep tasks running, stacked pull requests and Copilot code review to shrink diffs, and connectors to observability tools for post-release investigation. Each feature is an argument that judgment, not typing, is now the scarce resource.
Real-world use
- An API change where the design doc, compatibility notes and an old pull request live in three different tools that all need to be reconciled first.
- A team whose review board is clogged by large refactor PRs mixing logic changes, formatting and generated tests.
- An on-call engineer chasing intermittent timeouts the morning after a release, correlating logs, release history and customer reports.
- A developer who starts long agent tasks and needs them to keep running while they are in meetings or on another machine.
Why it matters
If review and investigation, not typing, are where your time actually goes, optimizing for generation speed alone will not make you more productive; the workflow around the code is the part worth automating next.