Cohort Kit

Cohort Kit

Eight ninety-minute sessions that turn the tutorial, the workbench and the course into a programme for a team. Pre-reading before, decisions during, one artefact after.

The tutorial teaches the method to one reader. A cohort learns it differently: the arithmetic lands when two people get different answers, the gates land when somebody has to chair one, and the roles land when the room contains more than one. This kit is the facilitator's side of that. Every session follows the same shape, so the second one is easier to run than the first.

Who it is for. Anyone running this for a team: an engineering manager, a delivery lead, a trainer, a study group that wants structure. You do not need to have run an agentic project; you need to have read the pre-reading one week ahead of the room.

What a cohort needs. A projector for the workbench, one laptop per two people, the pre-reading done, and a real candidate project from the team's own backlog. The sessions keep returning to it. Without one, the exercises use SkyWays, which works but lands less hard.

---

The eight sessions

#	Session	The question it answers	Tracks of the tutorial
1	Kick-off	Why does software with a model in it need a different lifecycle?	Getting started · Fundamentals 1–2
2	Frame and spec	Is this an agent at all, and what does a spec it can be built from look like?	Fundamentals 3–5
3	Build, run and the loops	How is it built in proven slices, run, and kept honest?	Fundamentals 6–9
4	Methods decoded	Where do AI-DLC, AIDD, BMAD and spec-driven development actually sit?	Methods decoded
5	Plan and review	How is the work planned, boarded and reviewed when a model writes most of the code?	Running delivery 1–5
6	Prove and ship	How accurate must it be, how is that proven, and how does it reach production?	Running delivery 6–9
7	Govern and scale	What holds, what gates, what drifts, and how does an organisation adopt it?	Running delivery 10–13 · Teams and organisation
8	Practice and roles	What does each person in the room do on Monday?	By role · Practice · Interviews

One session a week is the intended pace. Two a week works for a team that is about to start a project; one a fortnight loses the thread. The twelve-week reading group on the Study Plans page is the course-shaped alternative, built on the modules rather than the tutorial; the two can run in parallel with the same people.

The shape every session shares

Minutes	Block	What happens
0–10	Recap and pre-reading check	One question per lesson, answered by a person not the facilitator. Anyone who has not read is paired, not excluded
10–35	The idea	The facilitator walks one picture from the tutorial and one page of the workbench. Not a lecture: the picture is on screen and the room says what it shows
35–65	The exercise	Pairs work a numbered exercise from Exercises and Answers or a workbench calculator on the team's own project. Answers are compared before they are checked
65–80	The decision	The room takes one decision the session was building to and writes it down in the template that step uses
80–90	Homework and close	The homework is one artefact, named. The close is one sentence each: what changed in how you would do it

The session template has this as a copyable run-sheet, with the facilitator notes that apply to every session.

Facilitating it

• Compare before you check. When two pairs have different numbers, put both on the board and ask the room which assumption differs. The answer key is for afterwards.
• Keep the candidate project in the room. Every exercise that can be done on the team's own backlog is done on it. The SkyWays figures are the fallback, not the default.
• Let the unwelcome answer stand. Several exercises resolve to "this is not an agent" or "this bar cannot be met". A cohort that learns to say so has learned the most expensive lesson first.
• Time-box the arithmetic, not the argument. The calculators take two minutes; the disagreement about the damage figure is the session.
• End on an artefact. A session that ends with a written decision, however small, is retained. One that ends with a discussion is not.

After the eighth session

The team has: an AI-fit verdict on a real candidate, a bar with its sample size, a spec with the decisions left to the builder marked, a bolt plan, a review policy by risk band, a two-number report and a gate they have chaired. That is the evidence pack for P0 and P1 of their own project, which is the point.

Write a field note about how it went. The sessions here are designed and not yet field-tested; the first note from a facilitator changes them more than another draft would.

---

Related: Study Plans for the calendar-shaped plans · Exercises and Answers for the exercise bank · Scenario Library for cases from other industries · The course companion when the cohort will also build