Building Good AI Judgment Over Time · AI Judgment

Calibrated trust, not blind faith or avoidance.

Over-trust: confident prose without review. Under-trust: verify so much efficiency disappears. Aim for calibrated trust.

Calibrated trust over time

Note what you verified last week, confidence grows from understanding, not assumption.

Core principles

Build intuition by using the tool, noticing failure modes, and learning from mistakes.
Three automatic questions: What must I verify? What standard does my audience require? Does this sound like me?
MillerKnoll: Copilot in tenant; organizational data not used to train public models; permissions follow the user.
Review before you send, especially when output sounds confident.

Check yourself

What are the three automatic questions this lesson recommends asking before using any AI output?

Do this in Copilot

Write three personal rules for AI use (when you always verify, what you never paste, when you decline AI). Share one with a colleague.

Paste this into Copilot Chat and work through it before moving on.

Personal AI rules

Help me draft three personal rules for using Copilot at work: one about verification, one about data I will not put in prompts, and one about when I will not use AI at all. Keep each rule one sentence and actionable.

Open Copilot →

Personal rules

Help me draft three personal rules for using Copilot at work: one about verification, one about data I will not put in prompts, and one about when I will not use AI at all. Keep each rule one sentence and actionable.

Did you run this in Copilot? Mark complete when you have tried it.

The longer game

You just learned that AI gets things wrong in predictable ways, and that targeted verification, scaled to what's actually at stake, is the professional standard. That's the floor.

Here's what's above it. Research tracking over 92,000 workers found that the fastest-declining human performance metric right now isn't productivity or stress. It's mattering, the sense that what you contribute actually counts. And the data shows what happens when that drops: people stop bringing their judgment to AI outputs and start just forwarding them. Polished. Finished-looking. Useless. The antidote isn't more verification. It's being someone whose expertise is irreplaceable enough that the output without your judgment isn't the same thing as the output with it. The judgment you just built keeps you from getting embarrassed. What's on the other side of it keeps you from becoming optional.