Free course · 22 units

AI & LLMs → Agents

From tokens to production agents, about five minutes a lesson: read the cards, then answer on what you read.

  • 22 units
  • 63 lessons
  • 110 cards
  • 345 questions

Syllabus

What you'll learn

4 tiers, from foundations to advanced. Finish every unit and pass the final exam for the certificate.

  1. Foundations

    How AI models work

    Tokens, prompts and structured output

    • AI from zeroWhat a model is, how it learns and reads, prompts, made-up answers and agents, plus the words you'll see.
    • Tokens & next-token predictionWhat an LLM actually does, one token at a time.
    • Prompting & structured outputGet reliable, parseable answers.
    • Checkpoint: How AI models workMixed review of models, tokens, prompting and structured output.
  2. Core

    Build with models

    Embeddings, RAG, evals, tools and serving

    • Embeddings & RAGGround answers in your own documents.
    • Evals & guardrailsMeasure quality before you ship.
    • Tool use & the agent loopModels that call functions and act.
    • Local inference & servingRun models fast on your own hardware.
    • Checkpoint: CoreMixed review of RAG, evals, tools and local inference.
  3. Applied

    Ship agents

    Fine-tuning, harnesses, cost and MCP

    • Fine-tuning & adaptationWhen to change the weights, and how.
    • Production agentsDurable, observable, affordable agents.
    • Harnesses & context engineeringThe code around the model decides how well it works.
    • Routing, cost & local vs APIPay for the intelligence you need.
    • Agent memory & MCPWhat agents remember, and how they plug into tools.
    • Checkpoint: AppliedScenarios across fine-tuning, production agents, harnesses, cost and memory.
  4. Advanced

    Agent engineering in depth

    Agent evals, security, RAG and multi-agent

    • Agent evals & statisticsTell real improvements from noise.
    • Reward hacking & specification gamingWhen the agent hits the metric, not the goal.
    • Prompt injection & agent securityAgents read untrusted text. Plan for it.
    • Advanced RAG & retrieval evalsHybrid search, reranking and measuring retrieval.
    • Multi-agent orchestrationWhen to split work across agents, and how.
    • Eval designBuild evals that catch the failures that matter.
    • Checkpoint: AdvancedHard scenarios: agent reliability, reward hacking, injection, retrieval, orchestration and serving.

Try it now · 1 minute

Read one card. Answer three questions.

A real card from the Tokens & next-token prediction unit. No signup.

Lesson card · Tokens & next-token prediction

How a model learns

  1. Pre-training
  2. Instruction tuning
  3. Preference tuning
  4. Assistant

Three stages, from text predictor to assistant

Stage 1, pre-training: the model reads a huge pile of text and practises predicting the next token, billions of times.

Stage 2, instruction tuning: it trains on smaller sets of good question-and-answer examples, so it behaves like an assistant.

Stage 3, preference tuning: people rank answers, and the model is nudged toward the ones they prefer.

Most of the knowledge comes from stage 1. Most of the manners come from stages 2 and 3.

Key idea Pre-training gives knowledge; tuning gives behaviour.

New words pre-training: the first, biggest training stage: predicting the next token across a huge amount of text

New words instruction tuning: extra training on example questions and good answers so the model follows instructions

New words preference tuning: training that nudges the model toward answers people ranked higher

Glossary

114 new words, in plain English

Every new word is explained the first time it comes up. A few from this course:

.skip
test marker that skips a test, so its failure no longer shows (it.skip is the same thing)
cache
a fast store of recent results, so repeated requests don't redo the work
escalate
to pass a problem up to someone with more ability or authority
invoice
a bill sent to a customer listing what they owe
pruning
removing parts of a model to make it smaller and faster
server
a computer that runs programs and answers requests from other machines

See all 114 terms

FAQ

About this course

Is the AI & LLMs → Agents course free?

Every course, every lesson and the daily briefing are free forever. Paid plans add extras (no ads, unlimited AI); Elite adds a private course and never locks the core learning. An account is required, and after one sync the lessons work offline.

Do I need to know anything first?

No. It starts from zero and explains every new word the first time. A short placement check can skip what you already know.

How long does it take?

63 lessons of about 5 minutes, across 22 units. Go at your own pace. One lesson a day keeps your streak.

Do I get a certificate?

Yes, on every plan. Finish every unit, then pass the final exam (20 questions, 80% to pass) for a certificate you can check at /verify. It shows you finished the course, and it is not an accredited qualification.

Your first AI & LLMs → Agents lesson is five minutes away.

Not in the stores yet. Leave your email and we will tell you when it launches.

CEO Hero: launching soon

CEO Hero is not in the app stores yet. Leave your email for one note when it launches on Android.

Used only to tell you once when it launches.