Roadmap

One path, in order, through everything that is written. It runs from the job itself to the markets you sell into, and each stage assumes the one before it.

This page lists only lessons that exist. Topics that are planned but unwritten are left out entirely rather than listed as gaps, so you can follow this top to bottom without hitting a dead link. It currently covers 59 lessons, about 30 hours of reading, and it will grow as topics are finished.

The working method matters more than the order. Read a lesson, run its program with your own numbers, and do the third exercise, which is always the one that applies the lesson to your own product. Skipping the programs halves the value, because what most lessons actually teach is that a number you assumed was right is wrong.

The path#

flowchart LR
  A[":i-user: <b>1. The job</b><br/><small>8 lessons</small>"] --> B[":i-users: <b>2. Users and problems</b><br/><small>4</small>"]
  B --> C[":i-wrench: <b>3. Building it</b><br/><small>6</small>"]
  C --> D[":i-compass: <b>4. Deciding and sequencing</b><br/><small>8</small>"]
  D --> E[":i-handshake: <b>5. The organisation</b><br/><small>6</small>"]
  E --> F[":i-gauge: <b>6. Reading numbers</b><br/><small>6</small>"]
  F --> G[":i-coins: <b>7. Money</b><br/><small>9</small>"]
  G --> H[":i-bot: <b>8. What a model changes</b><br/><small>5</small>"]
  H --> I[(":i-building-2: <b>9. The market</b><br/><small>7</small>")]
  class A,B neutral
  class C,E compute
  class D queue
  class F,G io
  class H compute
  class I memory

Nine stages, fifty nine lessons. The first six are the job with no model in it, which is deliberate: everything in stage eight is easier to judge once you can already tell a real problem from a loud one and read a number without being fooled by it.

Stage 1: The job#

8 lessons, about 4 hours. Start here even if you have done the job for years. The second topic is written for somebody arriving at it, and the first is written for everybody.

Before moving on: run the week audit from lesson 07 on your own last week. The preparation figure is the one to look at.

Stage 2: Users and problems#

4 lessons, about 2 hours. Who the product is for, and how to tell a real problem from a loud one.

Before moving on: find three workarounds among your own customers. Ask what they do in a spreadsheet that they wish they did in your product.

Stage 3: Building it#

6 lessons, about 3 hours. Getting something specified, designed and built, plus enough systems literacy to know which requests are cheap.

Before moving on: write only the out of scope paragraph for something in flight, and circulate just that. Count the surprised replies.

Stage 4: Deciding and sequencing#

8 lessons, about 4 hours. Choosing what to do, and ordering it once there is more work than capacity.

Before moving on: run the capacity program with your team’s real numbers. The share of the headline figure actually available for new work is usually a surprise.

Stage 5: The organisation#

6 lessons, about 3 hours. The machine around the product team, and where work dies between functions.

Before moving on: ask a support lead what happened to the last three things they escalated. Their answer is your closure score.

Stage 6: Reading numbers#

6 lessons, about 3 hours. Do this before any topic that quotes a figure at you. It prevents most of the errors the money and model stages would otherwise walk into.

Before moving on: apply the doubling test to every number on your main dashboard and count how many survive.

Stage 7: Money#

9 lessons, about 5 hours. What you charge, what it costs, and the vocabulary a review already assumes you have.

Before moving on: write the six answers for your own company from memory, then check each one and note how far off you were.

Stage 8: What a model changes#

5 lessons, about 3 hours. The product work that only exists because there is a model in the product.

Before moving on: price an AI feature you are considering, per task, including retries, guardrails and human escalation. Compare that to whatever was quoted in the proposal.

Stage 9: The market#

7 lessons, about 4 hours. The same job bent by the market it runs in. Read your own sector first, then one that is structurally opposite to it, which is where the contrast teaches most.

Before moving on: write the five structural facts for the market you actually sell into, and show them to somebody in sales.

If you are in a hurry#

Six lessons that carry most of the value, in this order. About three hours.

1. The Product Loop, because closure rate is the habit everything else depends on.
2. Finding a Problem Worth Solving, because a workaround is worth more than any feature request.
3. The Denominator Problem, because most metric arguments end here.
4. Scoping and the First Version, because shipping at the flattening point buys months.
5. AI Product Shapes, because it converts “is the model good enough” into arithmetic.
6. Cost Per Request, because the quoted cost of an AI feature is usually a fraction of the real one.

Then#

Use Workflows when you have to actually do something, References when a lesson stopped short of a question you now need answered, and Glossary when a term appears and you want the definition this course uses.