The fifth session in the series, and a different shape from the first four. Rather than teaching something new, this one condenses everything the earlier sessions covered into a single deck people can keep open while they work. Every claim in it was checked against Anthropic’s documentation on 30 July 2026.
Audience: anyone who attended some of workshops #1–#4, or none of them.
Six parts
- How it works — it predicts the next chunk of text, everything is measured in tokens, it all has to fit in one window, and the whole conversation is re-sent every turn.
- What you’re driving — the four models, how much autonomy to grant, how hard it should think, and when to reach for Cowork instead of Chat.
- Context, in practice — why more context is not always better, what to do when a conversation gets long, the handover prompt, and the instruction stack your prompt sits on top of.
- Talking to it — treat it like a brilliant new colleague who knows nothing about your work; six techniques ordered by payoff.
- Trusting it — a model has no way to know that it doesn’t know. Four defences, cheapest first, and the situations that need a second thought.
- Giving it hands — MCP as the kitchen and a connector as the keys to it, what each connector costs you before you type a word, and how a skill gets deployed.
The one that changes behaviour
If a room only takes one part away, it’s context. Everything downstream of it — prompt quality, hallucination, the cost of a long conversation — is really a context problem wearing a different hat.