Beyond chat

Week 1 of 6 · 4 seeds

Stop asking, start delegating

Seed 1 of 4 · 30 min

Write the job description

You would never hire someone and just say 'help me'. That is what most people say to AI every day.

  1. Pick one job you actually want handed over. Not a question, a job. The weekly numbers, the follow-ups, the first draft of every proposal.
  2. Write it out like you are hiring. What it owns, what a good result looks like, what it must never do, and when it should come back and ask you.
  3. Paste that in as the setup, then give it the job with no other instructions and see what comes back.
  4. Fix the job description, not the answer. That is the whole skill.

What sprouts The difference between a tool you operate and a job somebody else holds.

Seed 2 of 4 · 20 min

Say what good looks like

Vague standards are why the output feels generic. It is matching a standard you never set.

  1. Take that same job and write three lines on what a good result contains and what a bad one looks like.
  2. Add one real example of a good one you have done yourself.
  3. Run the job again with the standard attached and compare the two versions side by side.

What sprouts Output you would actually send, because you finally said what sending-worthy means.

Seed 3 of 4 · 15 min

Let it ask you things

A colleague who never asks a question is a colleague who is guessing.

  1. Add one line to your setup: before you start, ask me up to five questions if anything is unclear.
  2. Give it a genuinely underspecified job and let it interview you.
  3. Notice which questions it asks. Those are the gaps in how you brief people too.

What sprouts A back and forth instead of a vending machine, and a much better brief at the end of it.

Seed 4 of 4 · 20 min

The handover test

If you cannot hand it to a person, you cannot hand it to an agent either.

  1. Take your written job and send it to someone who does not do your job.
  2. Ask them to do the task from your description alone, no questions.
  3. Whatever they get wrong is what the agent will get wrong. Fix that, not the model.

What sprouts Proof that agent work is delegation work, and delegation is a thing you already know how to get wrong.

That is week 1

Worth talking through with someone who has done it?

The seeds are the easy part. Deciding what to keep, what to drop and what to try next in your own situation is where people get stuck on their own. I do short calls for exactly this, and the first one is free.