Six starts. None of them finish you.

Track 01 · Judgment

Reviewing what you didn't write

For anyone now signing off model output

Fluent text reads as correct. That's the whole problem. This track builds a repeatable method for checking work you didn't produce, fast enough to be practical at the volume the job now demands.

• Week 1Error signatures in your own domain — what wrong output actually looks like when it's confident.
• Week 2Sampling under time pressure: what to check when you can't check everything, and how to choose.
• Week 3Adversarial review — finding the failure you'd have missed, on cases seeded with real errors.
• Week 4Building a review protocol your team can run without you, and testing whether it catches what you catch.
• ArtifactA written review protocol plus a scored record of your catches and misses on the seeded set.

Track 02 · Specification

Writing the brief

For people whose work arrives as tickets

Delegation to a model fails in the same place delegation to a person fails: an ask that sounded clear and wasn't. This track is about turning vague requests into constraints and acceptance tests that survive contact with someone else doing the work.

• Week 1Interrogating the ask — finding the unstated constraint before you start.
• Week 2Acceptance tests for non-engineers: what "done" means in a form someone can check.
• Week 3Decomposition — splitting work so failure shows up early and cheaply.
• Week 4Running a real delegation, to a model and to a person, and comparing where each broke.
• ArtifactA specification for a live piece of your own work, with the delegation results attached.

Track 03 · Systems

Building your own tooling

For non-engineers who keep hitting the same wall

Most workflow problems don't need a platform team. They need one person willing to assemble something small and maintain it. This track covers doing that responsibly — including recognising when the answer is to not build it.

• Week 1Mapping your workflow to find the step worth automating, which is rarely the annoying one.
• Week 2Assembling the thing — scripts, agents, and the joins between tools you already pay for.
• Week 3Failure modes: what happens when it breaks while you're on leave, and who is responsible.
• Week 4Handover — documenting it so it outlives your interest in it.
• ArtifactA working tool in use by at least one other person, with its handover document.

Track 04 · Evidence

Measuring whether it worked

For anyone asked to prove a change paid off

Almost every AI rollout is justified with a number nobody can defend. This track is evaluation design for people who aren't researchers: enough rigour to be honest, little enough overhead to actually get done.

• Week 1Baselines — measuring the before, which is usually already too late.
• Week 2Choosing a metric that can't be gamed by the thing you're measuring.
• Week 3Sample size and noise, without the statistics course.
• Week 4Reporting a result that includes what didn't work, and surviving the questions.
• ArtifactAn evaluation of a real change in your own workplace, written up for a sceptical reader.

Track 05 · Risk

Owning the decision

For people who answer to clients, boards or regulators

A model can produce the recommendation. It cannot be accountable for it. This track is about what accountability requires once the work is partly machine-made — disclosure, defensibility, and the failure modes that end up somewhere formal.

• Week 1Where liability actually lands when AI-assisted work goes wrong.
• Week 2Disclosure — what to tell clients and colleagues about how the work was produced.
• Week 3Building a record you could show a regulator, without drowning the workflow in it.
• Week 4Distributing judgment so the team isn't one person deep. Case studies from finance, health and legal work.
• ArtifactA disclosure and record-keeping policy for your own function, stress-tested against a failure scenario.

Track 06 · Persuasion

Explaining it upward

For people trying to change how work gets done

Being right about a workflow change is the easy half. The hard half is moving people who will be judged on the outcome and had no part in your reasoning.

• Week 1Reading the room you're pitching into: what each person is actually protecting.
• Week 2Making the case with evidence that survives a hostile question.
• Week 3Handling the objection you can't answer, without losing the room.
• Week 4Live pitch to a panel of practitioners, recorded and reviewed.
• ArtifactA recorded pitch and the written proposal behind it, with reviewer notes.