Production Hardening Path
Who it's for
Teams with agent-built features (or whole products) that work in demos and need to survive production: verification strategy, security review, deploy discipline, and readiness.
Prerequisites
- A staging environment and a real deploy path
- Ability to enforce CI gates
- Someone accountable for incidents (even if that's you)
Ordered modules
- Verification Strategy — what humans must own
- Reviewing AI Tests — agent tests lie differently
- Quality Gates — automate the boring gates
- Verification Checklists — operational checklists
- Reviewing AI PRs — last line before main
- Security review checklist — trust boundaries
- Deployment checklist — go-live hygiene
- Production Readiness review — subsystem bar
- Prototype to Production — demo → prod gap
- CI/CD Pipeline case study — gates in practice
- Production Debugging — when it still breaks
- Postmortem template — close the loop
Exercises
- Take one agent-merged feature; re-review with security + AI PR checklists; file gaps as tickets with owners.
- Add or tighten one CI gate that would have caught a real past failure (typecheck, boundary lint, smoke).
- Run a staging deploy using the deployment checklist; time how long smoke + watch takes; put that in the runbook.
- Draft production readiness for one service; get a no-go on purpose for one missing item, then fix it.
Capstone
Take one agent-built subsystem from "works on my machine" to production-ready: verification evidence, security checklist signed, deploy+rollback proven, readiness review go, and CLAUDE.md/DEPLOYMENT.md updated to match reality.
Expected outcome
You can articulate and enforce a production bar for agent-authored code — not "LGTM from the model," but gates, evidence, and rollback.