Practice: AI won't replace you. But it will expose you.

Self-check

Six quick questions. Answer each in your head (or on paper) before opening the collapsible. Active retrieval beats rereading by a wide margin, so the effort of recalling is where the learning actually happens.

1. What does it mean to call AI an “amplifier”?

Show answer

An amplifier makes what’s already there louder. AI makes your existing clarity, judgment, and taste more productive, and it makes your confusion, sloppiness, and guesswork more productive too. It does not supply what isn’t there. Clear input, clearer output. Silent in, silent out.

2. Name the four components of your human delta.

Show answer

Judgment, context, taste, and trust.

3. Which of these is a good use of AI, and why?

• (a) “Should I fire this supplier?”
• (b) “Draft three options for the difficult email to this supplier, and I’ll pick one.”

Show answer

(b) is the right call. (a) delegates judgment, which is yours, because “fire the supplier” depends on context and relationships AI doesn’t have access to. (b) delegates mechanics (drafting the language) and keeps the judgment with you (choosing which framing fits the situation).

4. Fill in the blank: “An AI can generate okay for free. ______ requires someone with taste to shape it.”

Show answer

Right. The distinction is “okay vs. right.” AI defaults to okay. You bring the taste that turns okay into right.

5. Why is “my first AI output was bad, so AI doesn’t work for me” a trap?

Show answer

You don’t write off a colleague because they were wrong once. You iterate with them, sharpen the ask, and try again. AI fluency works the same way. The first few attempts are clunky, and you get better by iterating, not by quitting after one try. One bad output is a data point, not a verdict.

6. A friend says, “I’ll start using AI once I really understand it.” What would you tell them?

Show answer

That’s backwards. You understand AI by using it on real work. The first two weeks are clunky. Fluency comes from doing. Reading about AI without using it produces opinions, not skill.

Try it yourself: the 10-minute audit

This is the whole lesson in practice form. You’re going to take one real task from your week and split it into mechanical vs. judgment parts, then pick the one piece you’ll try delegating to AI.

Side effects: none. This exercise is pen and paper (or a blank doc). No API calls, no file changes, no money spent. In the next lesson, we open Clawless and try delegating the item you circle here.

Steps:

1. Pick one recurring task from your week. Something you do at least every other week, that takes you more than thirty minutes. Examples: a weekly status update, a customer email thread, a research summary, a vendor comparison, onboarding notes, a board prep doc.

2. Draw two columns on a blank page or doc. Label the left column Mechanical. Label the right column Judgment.

3. List every part of the task, and for each one ask: if a competent assistant who doesn’t know our context tried this, would the result be right? Or would it be technically correct but wrong for this situation? Right for any context = Mechanical. Requires your specific context to get right = Judgment.

4. Circle one item from the Mechanical column. This is the one piece you will try delegating to AI this week.

5. Write one sentence describing what “success” would look like for that delegation. Example: “Success is an eight-bullet summary of the twelve-person Slack thread that I can skim in under a minute and trust.”

Expected outcome: you finish the ten minutes with three things in hand.

• A list of the parts that make up one of your real tasks
• A clear split between what’s mechanical and what’s judgment
• One specific item with a one-sentence success criterion, ready to try in the next lesson

That’s the concrete bridge out of “AI is a vague thing happening on the internet” and into “AI is something I’m going to use on this specific piece of work, this week.”

Flashcards

Ten cards in a format you can paste into Anki (or any other spaced-repetition tool) with the front/back convention below. One concept per card. Review once a day for a week, then on the intervals your tool suggests.

Q: What is the “human delta”? A: The set of things you bring to your work that AI cannot generate on its own at your level of consequence. Four components: judgment, context, taste, trust.

Q: What is the amplifier metaphor for AI? A: AI is an amplifier. It makes what’s already there louder. Clear input, clearer output. Silent in, silent out.

Q: Define judgment (in the human delta sense). A: Choosing between correct-looking options.

Q: Define context (in the human delta sense). A: Knowing what this particular situation needs, not what situations like it usually need.

Q: Define taste (in the human delta sense). A: Recognizing the difference between “okay” and “right.”

Q: Define trust (in the human delta sense). A: Being accountable to the people who rely on you. Trust doesn’t scale past the human who owns it.

Q: What is the correct delegation rule? A: Delegate mechanics, keep judgment. “Draft three options, I’ll pick one,” not “what should I do?”

Q: What do you compare AI output to? A: Your own baseline in the same time, not a perfect human. The baseline is real. Perfect is a moving target.

Q: What is the real sentence underneath “AI won’t replace you”? A: “But it will expose you.” It exposes whether you can think clearly, decide well, and apply judgment when the mechanical layer falls away.

Q: What’s the first concrete move after this lesson? A: Split one recurring task into mechanical and judgment columns. Circle one mechanical item to delegate to AI this week, and write one sentence describing what success would look like.