The learning method

Six stages, on repeat

Every Agentsable program runs on this cycle. Open any stage for the reasoning behind it.

The cycle

One loop, six stages

  1. 01

    Learn how it works

    Short explanations of the mechanism. Enough to reason with.

  2. 02

    Watch real work

    Someone builds it in full, dead ends included.

  3. 03

    Build it yourself

    Same session, blank file, your own problem where possible.

  4. 04

    Break it and diagnose

    Your build fails. You write down why before you fix it.

  5. 05

    Get specific feedback

    A reviewer responds to your decisions, not the polish.

  6. 06

    Improve and repeat

    Revise with what you learned, then take a harder problem.

Stage 04, up close

Breaking things is the lesson

Your build will fail, on purpose. Naming why it failed is what makes the next build better.

The presenter at her desk beside a shape sorter, the agents hovering over the slots they belong in
Every agent needs the right slot. Finding why one did not fit is the work.

In detail

Why each stage is there

01 Learn how it works

Mechanism before interface: what a context window costs, why retrieval changes how things fail, what an agent loop is doing when it decides. About 15 minutes of theory per hour of work.

02 Watch real work

Edited tutorials cut exactly what you need: the wrong turn, the misread error, the choice between two good designs. We keep those in.

03 Build it yourself

Watching feels like learning until you try. Closing that gap in the same sitting is the most important choice in the method.

04 Break it and diagnose

A learner who can name the cause can reuse the lesson. A learner who only has a working fix cannot.

05 Get specific feedback

Was the human checkpoint in the right place? Was the failure named? Was the check cheap enough to run? Polish is not graded.

06 Improve and repeat

The cycle is the product. A program is a sequence of problems fed through it. Programs change. The cycle does not.

What we can't show yet

Learner work is coming

Builds, reviews and failure write ups will appear here as people agree to share them. Until then, read this page as our standard, not as proof.

Practise the cycle with others

Four weeks, live on Zoom every Wednesday, starting November 18, 2026.

Reserve your seat

Super Early Bird: $100 for the first 30 seats.