Field Notes · No. 14

Know Your Weight Class

AI is good at the work below the line. The work above it is the job.

The cover of DEPLOYED No. 05 on a deep oxblood background: Before we hand AI more of our work, we should know which part is ours. Ryan in a navy blazer on the right.

Last week I wrote about what the AI company hires for now. Anthropic measured how AI changed its own jobs, and one line stayed with me: the strategic 20% of the job becomes 90% of it.

I ended with a question. What part of your job is the 20% that needs you?

It is a hard question to answer on the spot. Most of us never mapped our work that way. We picked up AI, found it useful, and handed it whatever was in front of us. Some of that was busywork. Some of it was the work that makes us good at what we do.

So here are the two exercises I use to sort it out. They come from my book, and I have put both on this site as a three-minute audit.

Four questions before we trust it

The first exercise is about literacy. Before we decide what to hand off, we should know what we are handing it to.

  1. What is this thing? A system that predicts what sounds right. It has no built-in sense of what is true.
  2. Where is its edge? It is strong at some tasks and weak at others that look the same.
  3. How would I know if it’s wrong? Check it against a source, every time it matters.
  4. What is it doing to me? It changes how hard we think, how well we learn, and how right we feel.

Question three is the one I test people on. The audit shows a confident sentence: my book reached #3 in Amazon’s New Releases in Artificial Intelligence in July 2025. It sounds right. It reached #8. Here is the morning it happened. A sentence that sounds right is exactly what these systems are built to produce, and only a source catches the difference.

Question four is the quiet one. Researchers at Microsoft and Carnegie Mellon surveyed 319 knowledge workers about 936 real tasks they did with AI. The more confidence people had in the AI, the less critical thinking they reported doing. The more confidence they had in themselves, the more they did. The habit that helps is small: when the thinking is the point, draft your own answer first.

The Energy Map

The second exercise is about the work itself. Look back over the last two weeks. Your calendar and your sent messages will do most of the remembering. Then fill four boxes, one item each:

  1. What gives you energy?
  2. What drains it?
  3. What moves the needle?
  4. What just keeps you busy?

Boxes 1 and 3 are your ring. That is the work that makes you good at the job, and the work AI should amplify while you keep your hands on it.

Boxes 2 and 4 are below the line. That is your automation target. Pick one item and hand it off this week. One trade, then the next.

The two-minute test

Before handing anything off, ask one more question. Could you tell a good result from a bad one in about two minutes?

If yes, hand it off and check what comes back. If not, you are not delegating yet. Learn to check it first, or trade a different item.

This matters more than it sounds. In 2024, three MIT researchers pooled 106 experiments that compared people alone, AI alone, and the two together. On average, the pair did worse than whichever was better alone. When the person was the stronger partner, adding AI helped. When the AI was stronger, adding the person hurt, most likely because people could not tell which of its answers to trust.

The two-minute test is how we stay the stronger partner.

Take it in three minutes

I built the audit so anyone can run both exercises without the book. Four questions, the map, one trade. You leave with a card: the work to protect, the work to hand off, and one trade to make this week. There is no sign-up, and the card stays with you.

AI is good at the work below the line. The work above it is the job.

What is one thing below your line that you could hand off this week?