Designing With AI

July 2026

Part 2 of 2 — Part 1, on the discipline behind the tool, is A Tool That Never Gives the Same Answer Twice.

1. Mark the maker

Let it be known, at a glance, which decision came from a machine and which came from a human. A distinct color, a distinct label — attribution isn’t decoration, it’s how someone calibrates how much to trust the thing in front of them.

2. Show your reasoning

Every output earns a “why,” grounded in actuals. Genericness is the tell that nothing real is behind it.

3. State it honestly

Say how sure you are, in words the data actually supports. “High confidence” that isn’t backed by anything is the same lie as a fake “+7%” — precision and confidence are one discipline, not two.

4. Leave the door open

Every step can be skipped. Every flow can be exited. Every action can be walked back. A system that forces a path through itself is optimizing for its own completion, not the person using it.

5. Guide, don’t command

Suggest. Don’t instruct. The tone of an output should sound like someone who respects that the human still decides — not someone who expects to be obeyed.

6. Close the loop

Don’t just act and vanish. Say what happens next — “we’ll check back in later” — so a completed action feels acknowledged, not dropped into a void the moment it’s done.

7. Less, louder

Show fewer things, and make the important one unmistakable. Two competing outputs, both shouting, teach a person to ignore both. One clear output, quietly followed by a second, teaches them to trust the first.

8. Show the seams

Trust isn’t earned by hiding how the system works — it’s earned by making the seams visible enough that no one has to wonder.