AI Training Session 3

The Session 2 retrospective showed the cohort retained the agent-loop mechanics. What was deferred to the close was what comes next. This session is what comes next: personalisation. Three things happen: 1. The structural pivot — chatbot persona (your homework) + generic agent loop (S2) → personalised agent. This requires two things the chatbot didn't: longer-term memory across sessions, and the ability to learn / adapt. 2. The Chief of Staff frame — Miessler's "what would I do if I had 1000 employees?" + Atomic Habits compounding. You're hiring employee #1. Your first agent is hard; your second takes an afternoon; by your fifth you're getting compounding returns. 3. The 7-layer roadmap — Identity, Context, Connectors, Skills, Automations, Memory, Verification & Learning. The single visual map of what you're building over the rest of the course. The session ends by showing the `ai-training-starter` project file structure as the concrete artefact participants will download next week and customise. Plus three homework deliverables that feed directly into the build session: SOUL.md + USER.md (from persona.md), GOALS.md, and a list of skill ideas.