Choose the right coding-agent tasks
Start with bounded, verifiable work and recognize tasks that need more human discovery first.
Strong candidates
The image highlights well-defined features, reproducible bug fixes, tests for existing behavior, targeted refactors and documentation updates. These tasks have a clear scope and evidence that can distinguish success from a plausible-looking change.
Tasks that need preparation
Ambiguous design decisions need discussion. Large architectural changes need decomposition. Performance work needs a measurable baseline and target. Work involving unavailable external systems needs an explicit access and testing plan.
These are reasons to clarify or split work, not claims that agents can never contribute.
Verify the environment's actual boundaries
The infographic describes an isolated, branch-based review workflow. Do not apply those permission claims to every Copilot surface or assume the agent can never merge or deploy. Effective permissions depend on the environment, credentials and policy.
Use least privilege, require review and protect important branches. Instructions and hooks complement these controls; they do not replace them.