Learn AI. Use AI. Build with AI.

Master AI.
Build with agents.

From your first prompt to intelligent AI workflows. Start from what a token is. Finish by designing agents that use real tools — and knowing exactly where they can go wrong.

Start learning Explore the platform

PromptA person asks for something

How it actually works

One request, nine stages

Scroll to follow a single request through an AI system — from the moment a person types it to the moment something happens in the real world.

Step 01 · Prompt
A person asks for something
Everything begins with a person and an intention. The prompt is the only lever you have on what follows — which is why how you write it changes everything downstream.
Step 02 · Tokens
Text is cut into pieces
Before the model reads anything, the text is split into tokens — chunks of characters, not words. Tokens are the unit of both the context limit and the bill.
Step 03 · Context
Instructions and history join in
Your prompt is not alone. System instructions, the conversation so far and any retrieved documents share one window. Everything competing for that space is context.
Step 04 · Model
Next-token prediction, repeated
The model computes a probability for every possible next token, one is chosen, and the loop runs again. No lookup, no database — a distribution, sampled over and over.
Step 05 · Agent
Deciding whether to act
An agent wraps the model in a loop: decide what to do, do it, look at the result, decide again. This is the step that turns answering into acting.
Step 06 · Tool
Requesting a capability
The agent does not run anything itself. It emits a structured request — a search, a file read, an API call — and your code decides whether to honour it. That gap is where every permission decision lives.
Step 07 · External system
APIs, MCP servers, databases
The tool reaches something real: an API, an MCP server, a database, a search index. This is the boundary where an AI system stops being a conversation and starts having consequences.
Step 08 · Result
Data returns to the loop
What comes back re-enters the context and is read like any other text — which is exactly why retrieved content must be treated as untrusted data, not as instructions.
Step 09 · Response
The person gets an answer, or an action happens
The loop closes: an answer, or a change in a real system. This is where verification matters most, because by now a mistake has left the model and entered the world.

The path

Ten stages, each built on the last

No jumping straight to agents. You will understand what a model is doing before you are asked to give one a tool and a goal.

The approach

Built to be understood, not skimmed

Explained, then shown

Every concept gets a plain explanation and then something you can poke at. Reading about tokens teaches less than watching your own sentence split into them.

Accurate, and dated

Concepts that last are written to last. Anything that can go stale carries its source and the date it was checked — because this platform teaches you to verify.

Honest about the limits

You will learn what these systems cannot do as clearly as what they can, including how to tell when confident output is simply wrong.

Stuck on a word?

24 terms, explained twice

One sentence for when you just need to keep reading, and the technical detail underneath for when you want it.

Open the glossary

Master AI.Build with agents.