Know what you are working with

Know what AI can contribute before you decide what to trust it with.

AI can make an answer feel finished before the work is understood.

That is useful when you need somewhere to start. It is dangerous when polish is mistaken for proof.

This lesson gives you a simple way to decide what AI can start, what needs checking, and what should remain yours.

See it

Relay's team asks why handoffs keep slipping. A designer gives AI one sentence: “Explain why teams miss handoffs and propose a dashboard.”

What went in: A broad question with no research, product behavior, or constraints.

What AI produced: A confident explanation, a familiar dashboard, and several plausible metrics.

What the designer checked: The interview notes showed that ownership changed in private messages. The proposed metrics did not exist, and the dashboard would have monitored the problem rather than changed it.

The answer was coherent. It was not grounded.

Why it works

  • AI is strong at producing plausible language, patterns, and variations.
  • It cannot supply product truth that was never included or reliably accessible.
  • The more costly a wrong decision is, the more visible the evidence and human review must be.
  • A useful result names its assumptions instead of hiding them inside polish.

Try it

Try it · 6 min

Mark the trust boundary

Choose one design task on your desk. Split it into three columns: AI can start, check closely, and I decide. Put each part of the task in one column, then name the evidence that would let you move it safely.

KeepA one-page trust map for one real design task, including the evidence each AI-assisted step needs.

Check it

  • Did you separate generating an option from deciding whether it is right?
  • Did you name the source material behind any factual claim?
  • Is the highest-risk decision still assigned to a person?
  • Could a teammate see where the result is uncertain?
Human checkpoint

Where would a convincing but unsupported answer change the product decision?

Inspect the user impact, business consequence, reversibility, available source material, and who remains accountable for the choice.

Keep it

Save this beside the task:

TASK

AI CAN START
-

CHECK CLOSELY
- Output:
- Evidence required:
- Assumptions to expose:

I DECIDE
-

WHY THIS BOUNDARY EXISTS
-

Before you use this

Do not upload private research or unreleased material to an unapproved tool. Verify claims against the source, not against another generated answer. You own the boundary and the final decision.

Can you use it?

Question: What is the difference between a plausible answer and a grounded one?

Answer: A grounded answer can be traced to relevant material and still shows what remains uncertain.

Next

Now use the work itself to decide which AI changes deserve your attention.