Learn
Seven short pages, in order. They exist to give you enough context to understand what we’re deploying and why it’s hard — not to be a complete course on agentic AI.
That’s deliberate. The deployment is the point. These pages move fast so you can get to it.
The through-line
Section titled “The through-line”They’re written as one argument, not seven topics. Each page ends by raising the question the next one answers:
| # | Page | The question it answers | Which leaves you asking |
|---|---|---|---|
| 1 | From chatbot to agent | What actually changes when the model can act? | What is this loop made of? |
| 2 | Anatomy of an agent | What are the parts, and which file is each one? | How does a model “use” a function? |
| 3 | Tool calling | How does the model act on the world? | Who else could use my tools? |
| 4 | MCP | Why a protocol instead of just calling the function — in both directions? | What does it remember between turns? |
| 5 | Memory and RAG | What lasts one run, and what outlives it? | What makes it stop? |
| 6 | Guardrails | How do you stop an agent doing damage? | What if nothing goes wrong and it still breaks? |
| 7 | Why deployment is hard | The Timeout Trap. The spine of the whole workshop. | → Build |
Read top to bottom and it should feel like one thing leading to the next. If a page ever feels like it arrived from nowhere, that’s a bug — the last section of the page before it is meant to have set it up.
Page 2 is the map. It names the six parts of an agent, and pages 3–7 are each a deep dive into one of them. If you get lost, go back to it.
Each page ends with how we actually use this in the demo and a link to the file it lives in, so the theory always lands somewhere concrete.