Skip to content

Six modules · personal curriculum

Building things that have to be right

Agent harnesses, durable workflows, compiling intent into rules, proving what a system did, and the moment a human signs for what a machine proposed.

Start module 01
37 lessons · 55 checks · ~5h

How this works

Every check makes you commit to an answer and rate your confidence before it shows you anything. That is not a gimmick — it is the best-evidenced intervention in the material, it costs no measurable extra time, and rating confidence turns the progress page into a calibration chart. The curriculum uses the techniques it teaches.

Expect to be wrong in places you feel certain. Two of the design instincts this body of work started with turned out to be contradicted by field evidence, and the modules say so rather than quietly dropping them.

  1. 01

    The Harness

    What are you actually building?

    A model is a function from context to tokens. Everything that makes it useful is code somebody else wrote.

    7 lessons
    58 min
  2. 02

    Workflows and Agents

    Where does the harness actually run?

    Durable execution, and the real problem of composing a deterministic engine with a non-deterministic model.

    6 lessons
    49 min
  3. 03

    Determinism From Ambiguity

    How does fuzzy human intent become behaviour you can defend?

    Getting from a sentence someone said to a rule that is repeatable, inspectable and diffable.

    6 lessons
    56 min
  4. 04

    Provenance and Audit Trails

    How does a system prove what it did?

    Building for the stranger who arrives fourteen months later with adversarial intent.

    6 lessons
    54 min
  5. 05

    Building a Workflow Engine

    What decisions does building one actually force?

    The architecture, and two opposite answers to one question: who assembles the workflow, and what is the durable artifact.

    6 lessons
    53 min
  6. 06

    Who Signs

    What happens when a human has to sign for what a machine proposed?

    The applied case. Where the first five modules meet a domain that already has a written specification for what a valid signature requires.

    6 lessons
    58 min

Running alongside

Not part of the sequence and not a prerequisite. The six modules answer whether a thing is built correctly, which is a different question from whether it is worth building.

  • World-class Product Sense in PracticeShreyas Doshi · Live cohort · one week · ~18h instruction plus Q&A

    The six modules answer whether a thing is built correctly. This answers whether it is worth building. Covers product sense, taste and strategy, with the Motivation / Friction / Satisfaction / Nudge frame for predicting what will actually get used.

    Module 06. The Approval Layer research keeps landing on judgment calls the evidence cannot settle — whether the review step earns its cost, what a controller actually needs, which 17% is worth building. That is product sense, not engineering.