Practical AI

Practical AI

  • Nine parts · about four hours

This module covers the craft of using and building with models. Parts 1 and 2 are how nearly everyone meets AI: the chat window, and the AI inside software they already use. Parts 3 to 9 are what to do when that is no longer enough: the instruction files, skills and plugins that shape an agent, what happens when one runs, models that take more than text, connecting a model to tools and systems, answering from your own documents, and keeping the result safe and measured. It names products and file formats throughout, and those date quickly. It is where the guide starts, and it needs no knowledge of how a model works. Its first part lists the seven things about a model’s behaviour that the rest leans on. Its part 9 ends with a design review checklist and the module’s own two reference sections. The whole module is about four hours of reading.

  1. Part 1: AI chat

    After it you can: Get far more out of a chat assistant, and know when you have outgrown it

    30 minutes

  2. Part 2: AI in apps

    After it you can: Judge the AI built into the software you already use, and what it may see

    25 minutes

  3. Part 3: Intent, AGENTS.md and other instruction files

    After it you can: Write the standing instructions that tell a model what you want, and know which file does what

    35 minutes

  4. Part 4: Skills, agents and plugins

    After it you can: Package know-how so an agent can use it, and choose between a skill, an agent definition and a plugin

    25 minutes

  5. Part 5: Agentic work

    After it you can: Brief an agent, keep it on course over a long task, and decide where it should run

    30 minutes

  6. Part 6: Multi-modal models

    After it you can: Use models that read images, audio and video as well as text, and know what they miss

    30 minutes

  7. Part 7: MCP servers and tool use

    After it you can: Connect a model to your own systems, and decide what it may do there

    25 minutes

  8. Part 8: Retrieval

    After it you can: Have a model answer from your own documents, and diagnose it when it answers badly

    25 minutes

  9. Part 9: Guardrails and evals

    After it you can: Contain what a model can do, and measure whether it is doing it well

    25 minutes