Skip to content
Playground

Big ideas. Small building blocks.

From a first prompt to an agent loop. Explore the learning path below.

Start with the five-minute text-predictor lesson below. The other nine entries are curriculum outlines, not working lessons.

Open playground

Start with four examples. Count, predict, then change the evidence.

Clayon's cartridge-shaped guide with a mint face and violet antennae
  1. 01

    Build a tiny text predictor

    Prompts

    Count next words, sample a token and see why frequent does not mean true.

  2. 02

    Temperature and length

    Prompts

    Same prompt, different answers.

  3. 03

    Tokens

    Prompts

    The model writes one token at a time.

  4. 04

    System prompt

    Prompts

    Instructions shape behaviour.

  5. 05

    Memory = the context window

    Memory and loops

    Model "remembers" only what is in the prompt.

  6. 06

    Chat loop

    Memory and loops

    A conversation is a loop over turns.

  7. 07

    Tool call

    Tools and agents

    Model asks for an action, code runs it.

  8. 08

    Agent loop (observe, decide, act)

    Tools and agents

    Agent repeats until goal.

  9. 09

    Plan then act

    Tools and agents

    A plan is a list of prompts.

  10. 10

    Embeddings and similarity

    Embeddings

    Words become numbers; close means similar.