Learn

Build Your First AI Workflow

A short course, one idea per lesson. Each lesson opens with a picture, says what to do in a few plain points and ends with one exercise. Your progress stays in this browser; nothing is sent anywhere.

Start with lesson 1

10 lessons, each a picture, a short read and one exercise.

NotesStepCheckStepYoulookResultTry again(once)
One job per step
Each step does one small job, so each is easy to write and easy to fix.

The Lessons

  1. 1Lesson 1: What a workflow isA chain of small steps with checks between them, and the jobs where one good prompt is all you need.
  2. 2Lesson 2: Steps, one job eachSplit a job until each step does one thing you could check on its own, in the order the work happens.
  3. 3Lesson 3: Writing the promptFour parts in every step prompt: who the AI writes as, the task, the rules and what to return.
  4. 4Lesson 4: ChecksA plain question about a step’s result that can come back no, and what happens when it does.
  5. 5Lesson 5: Hand-offsWhat one step passes to the next, in the same shape every time, so nothing is left to guess.
  6. 6Lesson 6: Loops with a limitDraft, check, revise, with a cap on rounds, so a stubborn problem reaches you instead of looping on.
  7. 7Lesson 7: Split, angles and best of severalThree ways to use more than one pass: split a long file, read a draft from two sides, or keep the best of a few tries.
  8. 8Lesson 8: Where a person must lookSome things only a person can judge. Put them last, after the automatic checks, as one plain question.
  9. 9Lesson 9: Testing a workflowA workflow you have not tested is a guess. Build a test set, run it, score it, record it, fix one thing, run it again.
  10. 10Lesson 10: Adapting a templateRun a template as it is, then make it yours: swap the input, change the reader or add a check.