Build your first real agents.
You are comfortable using AI daily. Level 2 makes you a builder: someone who can build agents, RAG systems, MCP servers, and evals, and who knows what each workflow costs in dollars. Sixty to ninety days at four to eight hours a week.
What you can do when you finish
- You have built three real agents on your own stack, anchored on your own work
- You have deployed at least one MCP server
- You have built a RAG pipeline over your own documents
- You have an eval harness with regression testing, so changes cannot quietly break things
- You can state the cost of each workflow in dollars
- For companies: eval framework live, observability running, first agents in production
What is inside
Twelve core modules on the Individual track, from agent architectures through your first three agent builds, an MCP server, a RAG pipeline, evaluation, and cost engineering. Eight modules on the Company track, including agent governance and RAG at scale. Four public companion repositories back the build walkthroughs with working code, so you are never building from prose alone; the repos overview maps each one to its module.
You will want basic Python or TypeScript, comfort in a terminal, and a codebase or dataset of your own to anchor the examples on. Level 1's exit criteria are the assumed floor.
Where this meets the doctrine: the guide's Operator is the first agent most owners ever run. Level 2 is where you start building agents like it yourself. Every companion repo ships with propose-and-wait defaults, the same Zone 2 rule your Order 3 governance file sets: the AI drafts, a human approves, nothing sends itself.
How to start
- Confirm the Level 1 exit criteria hold; the level map in the repo has the checklist.
- Pick the thing you will anchor on: your own codebase, dataset, or business documents.
- Work the Individual track in order; the first agent build is module 02.
- Companies: stand up the eval framework early. It is what makes the rest safe to scale.
Read Level 2 on this site Start Level 2 on GitHub The whole book (PDF)