Overview

A chatbot answers questions; an agent pursues goals — the difference is a loop, and by the end of this day you'll have built one and understood every line. You'll write system prompts as job descriptions (role, tools, goals, constraints, workflows), wire up the ReACT loop so the model decides its own next move, and give agents real tools and memory with RAG (FAISS, pgvector) — and see that it's search, not magic. Then you'll build your own agent framework in about 140 lines, so every SDK stops being a black box, and orchestrate a small swarm of agents that outwork you. No hand-waving: you leave able to build agents that think, act, and learn, from scratch.

Who this is for & what you'll need

🔵 Solo codingYou should be able to write in some language on your own.
🖥️ A few installsYou'll install a couple of tools — we'll help you in class.

Where This Class Leads

What you leave able to do

  • construct a retrieval system
  • construct an agent that uses tools
  • govern a swarm by information hierarchy
  • re architect an agent as a stateless reducer
  • reconcile agents from different toolchains into one run
  • select among agent architectures
  • synthesise a self extending agent behind a review gate

How you show it. A diff the agent authored to its own prompt or toolset, the benchmark run that accepted it, and a logged rejection with the reason recorded.

Join Us

Want to schedule this class for your team?

Contact us: liz@themultiverse.school