Learn AI agents
Start from zero or fill in the gaps. This is the map: what an agent is, the concepts that matter, a roadmap from beginner to advanced, and a glossary for every term you'll meet along the way.
What is an AI agent?
An AI agent is a program built around a language model that can pursue a goal by taking actions — not just answering one question. Give it an objective and it works in a loop: it looks at the current situation, decides what to do next, uses a tool or takes a step, sees the result, and repeats until the job is done.
That loop is the whole idea. A plain chatbot responds once. An agent can search, call an API, run code, read your documents, remember what happened earlier, and correct course when something fails. The concepts below are the pieces that make that possible.
Your learning path
Three stages, in order. You don't need to finish one before dipping into the next — but this is a sensible route from curious to capable.
Foundations
~1 hour- What a large language model (LLM) actually does
- Prompts, context windows, and tokens
- The difference between a chatbot and an agent
- The agent loop: observe → think → act → repeat
Core building blocks
~2–3 hours- Tool use and function calling
- Memory: short-term context vs. long-term stores
- Retrieval-augmented generation (RAG)
- Planning, reasoning, and breaking down tasks
Production & advanced
Ongoing- Multi-agent systems and orchestration
- Evaluation: measuring whether an agent works
- Guardrails, safety, and human-in-the-loop
- Cost, latency, and reliability in the real world
Core concepts
The building blocks that show up in almost every agent. Click any card to find related material across Butlerz.
Language models
The engine of every agent. An LLM predicts the next token given everything in its context, which is what lets it write, reason, and follow instructions. Understanding its strengths and limits is step one.
ExplorePrompting
How you instruct the model. Clear roles, examples, and constraints dramatically change the output. Good prompting is the highest-leverage skill in building agents.
ExploreTool use
Agents become useful when they can act — searching the web, calling an API, running code, or querying a database. Function calling lets the model choose a tool and supply the arguments.
ExploreMemory
Context windows are finite, so agents need memory: keeping the recent conversation in view, and storing important facts to recall later across sessions.
ExploreRetrieval (RAG)
Retrieval-augmented generation pulls relevant documents into the prompt so the agent can answer from your data instead of guessing — the backbone of most knowledge assistants.
ExplorePlanning & reasoning
Hard tasks need to be broken into steps. Techniques like chain-of-thought and plan-then-execute help an agent decide what to do next instead of answering in one shot.
ExploreMulti-agent systems
Some problems are better split across specialized agents — a researcher, a writer, a critic — coordinated by an orchestrator that routes work between them.
ExploreEvaluation
You can't improve what you can't measure. Evals — test cases, scoring, and comparisons — tell you whether a change made your agent better or worse.
ExploreSafety & guardrails
Agents that take actions need limits: input validation, permission checks, and human approval for risky steps. Guardrails keep an agent helpful without letting it go off the rails.
ExploreGlossary
The vocabulary of AI agents, in plain language. Keep this handy as you read.
Ready to build?
You've got the map. Now pick a direction — read deeper in the Library, follow a Playbook end to end, or grab a Template and start.