What does a standard development workflow look like once you adopt AI?

Question

IT Interview Practice · Accepted Answer

AI doesn't remove the software lifecycle — it compresses each stage and shifts your job from writing code to specifying, reviewing, and verifying it. Keep the same phases (spec → design → implement → review → test → ship) but make the human gates explicit.

treat AI as a fast, eager junior pair — it drafts quickly but needs direction and checking. The unit of work becomes at every stage, not "type it all yourself".

**Mental model:** treat AI as a fast, eager junior pair — it drafts quickly but needs direction and checking. The unit of work becomes *"specify → generate → verify"* at every stage, not "type it all yourself".

## The workflow

```text
Spec ──▶ Design ──▶ Implement ──▶ Review ──▶ Test ──▶ Ship
 AI       AI         AI            HUMAN      AI        HUMAN
 draft    options    code+tests    gate       gen+run   gate
```

- **Spec first.** AI is only as good as the spec — write a clear problem statement + acceptance criteria before generating. Garbage spec → confident garbage code.
- **Design.** Ask AI for 2–3 approaches with trade-offs; a human picks.
- **Implement.** AI drafts code + tests; you steer in small, reviewable steps.
- **Review (human gate, non-negotiable).** Someone who understands the code approves it — AI output is reviewed **more**, not less.
- **Test.** AI helps generate/expand tests; CI runs them.
- **Ship.** A human owns the deploy/rollback decision.

The key shift: the bottleneck moves from *writing* to **reviewing and verifying** — so invest in specs, tests, and review, not raw typing speed.

## Interview lens

The interviewer is checking whether you keep engineering discipline or just "let AI write it". **Weak** answer: "AI writes the code so we ship faster." **Strong** answer: same lifecycle, AI accelerates each phase, humans own the gates (spec, review, ship), and **verification is the new bottleneck**.

## Pitfalls

Skipping the spec produces confident nonsense; skipping review builds quality/security debt; measure **outcomes** (working software, fewer incidents), not lines of code.

The workflow

Interview lens

Pitfalls

Related questions