Get the app

Working With AI

Specify, delegate, verify and own AI-assisted work — asks that come back right, checks sized to the stake, workflows a colleague can run. No code.

28 lessons

Download on the App Store
Explore with AI:ChatGPTClaudePerplexity

What you'll learn

  1. Unit 1

    The Handoff Is the Work

    6 lessons
    1. One braided idea: the model never met you, so the words in the box are the whole job. First the gap — a real ask and the reply it produced, with the sentence the learner never wrote pointed at — then the stranger's read: which part would a capable temp with no context have to guess. A load-bearing image (R5) anchors both halves: the ask on the left, the reply on the right, and the empty space between them where the intent lived.
    2. The learner acts by tapping, in a spec table whose rows are the things a reply can invent (who it is for, what week, what tone, what to leave out), exactly the rows the original ask never supplied, and the material responds by lighting the supplied rows green and revealing what the model filled into each missing one.Exercise
    3. One idea: before you delegate, ask whether you could tell a wrong answer from a right one in a minute. Six real requests from one working day, sorted live; the reveal is that the uncheckable ones are exactly the ones people delegate most eagerly.
    4. Which handoff failsChallenge
    5. The learner acts by tapping, in an inbox table of eight requests, the ones they could verify in a minute, and the material responds by revealing for each tapped request the one-minute check that exists — or, for a wrong tap, that no source, reproduction or recognisable fact is in hand.Exercise
    6. The handoff reviewUnit review
  2. Unit 2

    Say What Done Looks Like

    7 lessons
    1. The gap from unit 1 gets a shape: four things a reply has to know, and an ask that supplies all four leaves nothing to invent. One real ask rebuilt in four moves, the reply changing after each.
    2. The learner acts by tapping, in a four-row spec table, the row each sentence of a real ask fills, and the material responds by lighting the row and revealing which rows the finished ask still leaves for the model to guess.Exercise
    3. Two ideas braided: an example fixes tone, length and shape at once where adjectives fail, and a fence says what to leave out. One brief for a product description, tried with adjectives and then with one pasted example plus one fence.
    4. Spot the missing partChallenge
    5. The learner acts by tapping, in a table of reply passages, the passages a stated fence would remove, and the material responds by striking them and revealing the length the fenced reply comes to.Exercise
    6. The learner acts by assembling, on a four-row ask board, the sentence for each row from candidate lines that include near-misses, and the material responds by accepting the row and revealing the reply fragment that row now controls.Exercise
    7. Say what done reviewUnit review
  3. Unit 3

    Give It What You Know

    5 lessons
    1. The model has none of your files, your history or this week. One ask about a real situation, and the reply's confident wrongness traced fact by fact to the thing the learner knew and never pasted.
    2. The move that makes an answer checkable: supply the document and ask about it, then read the answer against the document. One contract clause, asked from memory and then from the pasted text, and the difference in what can be checked.
    3. What was it givenChallenge
    4. Permission to say 'not in the text' changes which answers you get. The same pasted document, the same five questions, with and without the line — and the two answers that flip from invented to honest.
    5. What you know reviewUnit review
  4. Unit 4

    Read the Answer Like a Reviewer

    5 lessons
    1. A reply read twice: once as a reader, once as a reviewer with three questions — what does it claim, what is costly if wrong, what do I check first. The second reading finds the sentence the first one slid past.
    2. The most specific detail is the most convincing and the most often invented. A reply about a regulation with a date, an article number and a quoted sentence — and which of the three survives a check.
    3. Which claim firstChallenge
    4. An ask that carries its own answer gets that answer back. 'Why is our churn caused by pricing?' returns three reasons pricing causes churn. Asked the other way, it returns three reasons it is not. The reply mirrored the frame; the learner learns to ask the frame away.
    5. Reviewer reviewUnit review
  5. Unit 5

    Iterate, Don't Restart

    5 lessons
    1. A correction is three parts: the defect named, the good part kept, one change asked. A reply that is 80% right and the two corrections that follow — one throws the 80% away.
    2. Two edits in one turn produce a reply nobody can attribute. The same draft improved in one two-edit turn and in two one-edit turns; only the second tells you what worked.
    3. Which correction worksChallenge
    4. A long thread keeps steering by its own wrong turns. Eight turns into a draft, every correction makes it worse; the new-chat rule — start over with the good parts pasted in — and the reply that comes back.
    5. Iterate reviewUnit review

Explore more roadmaps

Psychology
A working understanding of psychology — what the science actually claims about how minds, brains, and behavior fit together. The trunk gives you the orientation a literate professional needs (how psychology knows what it knows, the brain behind behavior, learning, memory, decision-making, emotion, personality, social influence, development, pathology, and what therapy does); clusters dive into research methods, perception, cognition mechanics, social influence tactics, the field guide of disorders, applied behavioral science, and the psychology of everyday life.
Personal Finance
The financial life skills nobody actually taught you — money psychology, budgeting, debt, credit, taxes, insurance, retirement accounts, and the math of financial independence. Walk the trunk and you have a system; pull on a cluster when something specific matters.
Natural Science
A working understanding of the natural world — what science is, how matter and energy behave, how life persists and changes, and how Earth and the cosmos fit together. The trunk gives you a single coherent throughline across physics, chemistry, biology, and earth/space science; clusters dive into experimental design, mechanics, thermodynamics, electromagnetism, modern physics, genetics, ecology, and climate.
Philosophy
A guided tour through philosophical thinking — how to reason carefully, evaluate arguments, navigate ethics, and engage with the great questions about knowledge, reality, mind, and society. The trunk gives you orientation across the whole field; clusters dive deep on logic, ethics, epistemology, metaphysics, mind, politics, and the history of ideas.
Law
Legal literacy — how law actually works. The trunk gives you the operating system: where law comes from, how courts and precedent work, contracts, torts, property, criminal vs civil, and constitutional structure. Clusters dive into reading cases, contract mechanics, IP, your rights with police, and the startup legal toolkit.
Programming
A practical track that takes you from zero to shipping working software. The trunk gives you the throughline — how to think in code, write a program, and put it in front of users — and clusters dive deep into thinking tactics, Python craft, engineering for change, real-world shipping, and working alongside AI assistants without letting them put garbage in your codebase.

Frequently asked questions

What does the Working With AI roadmap cover?

5 units, starting with The Handoff Is the Work, Say What Done Looks Like, Give It What You Know and Read the Answer Like a Reviewer, and 1 more.

Who is the Working With AI roadmap for?

Anyone who already hands work to an AI model at their job and wants it to come back right — and to catch it when it does not.

How do I start the Working With AI roadmap?

Get the Mochivia app on the App Store and open this roadmap there.

Start Working With AI

Download on the App Store