AI Training Session 2

Many people still think of AI as a chatbot — a text-in/text-out tool (on its own an LLM is a one-shot text generator). This session replaces that mental model with: LLM = brain, tools = hands, context = working memory, loop = the thing that cycles. Everything downstream (your own agent, the Master Prompt, skills) assumes this new model. Today we tour the inner harness — the bit Anthropic builds and ships ready-made: context, tools, artifacts, the three surfaces. Next week (Session 3) we move to the outer harness — the bit you build to make a generic agent into yours: skills, prompt craft, choosing your path. The session also tours all three Claude surfaces — Claude Code (terminal), claude.ai (browser), Claude desktop / Co-work (GUI + agentic mode) — using each one where it fits the topic, so attendees see the same engine through different channels. It builds concretely on the worked example — the Weekly Digest Agent — by demoing the full thing in 2.1. That demo (using Alex Okafor's persona and the week-of-noise input the cohort got attached to their homework) is the spine of the entire session. Vocabulary reveal: the session opens with the word agent (familiar) and at 2.5 names the same concept the way the industry now does — agent harness.