- The 5-question diagnostic: does it involve text, is it repetitive, does quality vary, could a smart intern do it with good instructions, are the inputs structured — the more yes answers, the stronger the case
- The two-category map: 10x leverage tasks (drafting, summarizing, restructuring, brainstorming, getting unstuck) versus 10% improvement or worse (real judgment, original analysis, accountable decisions, deep relationships)
- Most people misclassify tasks in both directions — overestimating where AI helps, underestimating where it doesn't
- The "second pair of eyes" pattern: asking AI to critique your work rather than produce it — often more valuable, lower risk, and more sustainable than full delegation
- How to stress-test a use case before scaling it across the team
- Why honest failures are as important as wins — they reveal what the role actually requires
Map your own week against the 5-question diagnostic. List the three tasks where AI would have the highest leverage, and the three where it would actively get in the way.
Pick the strongest candidate from your list and stress-test it on a real piece of work this week. Bring the result — successful or not — to your team meeting.
Discuss the tasks where AI has earned a real place in our work — and the tasks where it hasn't.
Honest failures — tasks you thought would be a great fit and weren't. Those reveal what the role actually requires that AI can't provide. And concrete wins — high-leverage use cases described in enough detail that others can try them.
- What if my job doesn't have obvious AI use cases?
- Most people who say this haven't tried the diagnostic yet. Almost every knowledge-worker role has at least a few tasks that involve generating or restructuring text. Start there.
- What if the AI use case I found isn't approved yet?
- Flag it. That's exactly the signal the program needs. Don't route around the process — surface it so it can be evaluated.
Send us the highest-value use cases your team surfaced — and the failed attempts with a note on why they didn't work. A use case map with honest failures is more useful than a list of wins.