How do you coach your team to use AI effectively without over-reliance or blind trust?

Question

Accepted Answer

The goal is to grow **judgment, not dependence.** I want engineers who use AI as leverage and still get sharper over time — not ones who can't ship without it and can't tell when it's wrong. Coaching is about building that habit deliberately.

## Teach the two skills that matter

- **Prompting** — how to give context, constrain the output, and iterate. A good prompt is half the result.
- **Verification** — how to check the output: run it, test it, read the docs, reason about correctness. This is the skill that separates leverage from liability.

I pair these explicitly: every use of AI ends in verification, never "looks right, merge."

## Set clear norms

- **Never merge code you don't understand.** This is the single non-negotiable rule.
- **Use AI for the right tasks** — drafts, boilerplate, exploration, test scaffolding.
- **Discourage copy-paste of unverified output**, especially in critical or unfamiliar areas.
- **Be transparent** about where AI was used, so review can focus appropriately.

## Build a feedback loop

- **Share patterns** — a channel or doc for prompts and workflows that worked.
- **Review together** when AI produces something subtle — turn a near-miss into a team lesson.
- **Watch for over-reliance** — if a junior can't explain their own PR, that's a coaching moment, not a gotcha.

## Grow judgment, not just output

The risk with juniors is that AI lets them produce code faster than they build understanding. I counter that by insisting they can explain and defend everything they submit. The tool accelerates work; it must not replace learning.

## Why it matters

A team that trusts AI blindly ships confident bugs and stops growing. A team that uses it well moves faster *and* gets stronger. The lead's coaching is what decides which team you end up with.