Judgment · Working with AI

Knowing when. Knowing when not to.

The people who get the most out of AI are not the ones who use it most. They are the ones who use it on the right things. They have a sense of where AI adds value and where their own judgment, creativity, or relationship is the actual product. They are not using AI to avoid work they love. They are using it to get through work they don't.

This is the skill that does not come from a tutorial. It comes from knowing yourself. What work gives you energy? What work drains it? What would you fight to keep doing yourself? What would you happily hand off to anything that would take it?

Start there. That is not a philosophical question. It is a practical one.

When to use AI versus keep the work human

Hand off volume and structure. Keep accountability, taste, and work you would fight to keep.

Core principles

The dentist task test: is this work essential but dreaded? That is the right target. Nobody questions if it has to happen. They just hate doing it.
The pitchfork test: would you fight to keep doing this yourself? High fight means your judgment or joy is load-bearing. Low fight means it's probably a candidate.
AI for volume and structure. You for accountability and taste. Verification in between.
Protecting work you love is not resistance to AI. It is good judgment about where your humanity is the product.

Check yourself

What is the "pitchfork test" from this lesson?

Do this in Copilot

List ten real tasks from your job and work through the AI candidate analysis.

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

The judgment prompt

I'm going to list ten things I do regularly at work. For each one, I want you to help me think through whether AI could help, hurt, or shouldn't touch it. Here are my ten things: [list ten real tasks from your actual job]. For each one, tell me: is this a good AI candidate, a bad one, or somewhere in between, and why. Then ask me which one surprised me most and why.

Open Copilot →

Task triage

I'm going to list ten things I do regularly at work. For each one, I want you to help me think through whether AI could help, hurt, or shouldn't touch it. Here are my ten things: [list ten real tasks from your actual job]. For each one, tell me: is this a good AI candidate, a bad one, or somewhere in between, and why. Then ask me which one surprised me most and why.

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

The longer game

You just built five real skills, prompting, context, iteration, verification, judgment, and practiced each one on actual work in Copilot. Not a demo. Not a tutorial. Your email, your meetings, your documents.

Here's the significance of that. Research tracking the AI performance of teams across thousands of workers found that the relationship between AI usage and performance is basically flat when you look at the whole population. More usage doesn't automatically mean better outcomes. But when you split the data by one variable, whether people had real practice in real conditions, with feedback, on work that actually mattered, the lines diverged dramatically. The skill you just built isn't theoretical. It's practiced. That distinction is exactly what the data says separates the people who get compounding value from AI from the people who plateau at polished-looking mediocrity.