Engineering Manager Path
Who it's for
Engineering managers whose teams are already past "try Copilot" and need norms: when to spec, how to review agent PRs, how to keep seniors from drowning in review, and how to coach juniors who over-trust output.
Prerequisites
- A team of ≥3 with a shared repo and PR culture
- Support from leadership to change definition-of-done
- Baseline metrics: cycle time, revert rate, review latency
Ordered modules
- Responsibility Split — teach this explicitly in 1:1s
- Why Agents Fail — shared vocabulary for retros
- Docs as Agent Memory — team invests in repo docs or pays in review
- Writing Effective Specs — bar for "ready for agent"
- Reviewing AI PRs — calibrate the team on this playbook
- Quality Gates — reduce reliance on hero reviewers
- Large Tasks — stop stuffing epics into one session
- Scope and Context Failures — most common team failure
- Sprint plan template — make agent capacity explicit
- Multi-Agent Feature case study — when parallelism is worth it
Exercises
- Add "spec attached or explicitly waived" to the team's PR template; enforce for two sprints.
- Shadow-review five agent PRs with the AI PR checklist; publish anonymized findings.
- Measure review hours before/after introducing mandatory verify commands in CLAUDE.md.
- Run a team workshop on one failure mode using a real internal incident or near-miss.
Capstone
Ship a team working agreement: definition of ready (spec), definition of done (verify), review SLAs for agent PRs, and a coaching plan for over-trust. Present metrics from one sprint under the agreement.
Expected outcome
Your team shares a quality bar, review load is predictable, and juniors can explain when not to trust the agent — without you gatekeeping every merge.
Related
- Learning Paths
- CTO path — org policy above you
- Senior Engineer path — IC skills to coach toward
- Hallucination and False Confidence