Twelve sessions. One real deliverable.
Scoping the Engagement
Scope with the business and lock the plan.
Building
Sprint the build; adjust as reality lands.
Training & Handoff
Test it, train them, and set it up in their accounts.
Buffer & Close
Held in reserve — then review, present and grade.
The pod math, out loud.
One senior + agent-equipped juniors is the forward-deployed firm’s only viable structure — and these five teams are pod dry-runs, with your instructor as the senior. This isn’t a class exercise shaped like a firm; it’s a firm shaped like a class.
Maister’s professional-services math: partner profit is the surplus on staff — firms above 8:1 staff-to-partner earn ≈ $491K per partner; below 4:1, ≈ $260K. Audit runs 16:1 on 22-year-olds (class 14’s gate is how).
Review capacity caps leverage: senior review runs 20–25% of junior build hours — at three juniors, ~17 hours a week of reviewing. A fourth forces rubber-stamping. Yes, our program runs hotter than 3:1 — which is exactly why first-pass review is peer review inside your team, and the instructor holds the client-visible gate.
The market bet against juniors. You’re the counter-bet.
Entry-level employment in AI-exposed occupations: −13% (Stanford/ADP). Big Tech new-grad hiring: down more than half since 2019; new grads ≈ 7% of hires (SignalFire). The bet: juniors + AI are replaceable.
Juniors + AI are leveraged — when someone senior actually reviews the work. The pod keeps the supervision machinery employers abandoned. That machinery is this movement. (The hiring wall from class 02 is the same bet, from the buy side.)
+34% for novice call-center agents with AI (NBER, 5,179 agents); the Copilot RCT ran 55.8% faster, biggest for the less experienced. The famous slowdowns hit experts on mature codebases — your build is greenfield precisely because 17–19 locked the spec.
Leverage without rubber-stamping.
1WIP limitsOne feature in flight per builder. Agents make starting free — finishing is the scarce resource. Reviews get 48 hours (class 14’s clock); when the reviewer is the bottleneck, WIP drops — you build less in parallel, never ship unreviewed.
2No solo client calls — earlyFor the first sprint weeks, client-facing conversations happen in pairs — your Echo (the team’s stakeholder owner) exists for this. Solo comes when trust is earned inside the team.
3Every feature carries its trailSpec · eval results · review notes travel with the work — the reviewer must be able to reconstruct why, not just what. (The audit standard’s spirit, at student scale.)
4Cold-review samplingOnce a sprint, the instructor pulls one already-merged piece per team — logged in the team channel. Not distrust — calibration. It keeps “reviewed” meaning something.
Reality lands mid-build. Good.
Somewhere in these weeks, two things happen on schedule: the build meets a fact nobody knew, and the owner — delighted — says “while you’re in there, just have the AI also…” You rehearsed this in class 13, and the artifact is the same one-page change-order form from the course repo — the owner signs the scope change, the instructor countersigns: accepted, deferred past the primary, or declined. Never absorbed.
Connect.AI