AI is in how you build. This playbook is for teams using AI tools to research, design, code and operate. Same seven phases as its companion, different question in each.
<aside> ⭐
The golden rule: AI accelerates the typing. You still own the thinking, and a human is always accountable for what ships.
</aside>
The evidence worth knowing first: DORA's research found AI makes individuals faster, but team delivery can get less stable at the same time. Not because of the AI. Because fast code generation tempts teams into big batches and thin review. This playbook is mostly the fix for that, phase by phase.
<aside> 📦
Full playbook on GitHub: step-by-step phases, exit criteria and templates. github.com/EmilieBuilds/AIProductPlaybook
</aside>
| Phase | In standard terms | What AI changes here |
|---|---|---|
| 1. Frame the problem 🔍 | Define | Research gets faster. Judgement stays human. |
| 2. Govern before you build 🛡️ | Plan | You need an AI-use policy before the first prompt. |
| 3. Design the solution ✏️ | Design | Prototypes get almost free. Try more of them, then throw them away. |
| 4. Build 🔨 | Build | The big one. Small batches, real review, and never merge code you don't understand. |
| 5. Evaluate 🧪 | Verify / Test | AI can write tests and review code. It cannot own quality, and it doesn't grade its own homework. |
| 6. Launch 🚀 | Release | Discipline unchanged. AI does the paperwork, humans sign it. |
| 7. Operate and improve 🔁 | Operate / Run | AI triages incidents. Humans decide. Measure whether AI is actually helping. |
📝 Team AI-use policy, written in Phase 2. One page: approved tools, green/amber/red data rules, code review rules, licensing stance, and who owns it. Reviewed quarterly, because the tools change monthly.
<aside> 🚦
Green, amber, red in one line: public info and your own code go in prompts freely 🟢, internal docs with care 🟠, and secrets or identifiable customer data never 🔴.
</aside>