Inside the Agent
Understand every moving part
Section titled “Understand every moving part”A model proposes an action. Python checks the request, runs an allowed function, and sends the result back. Another model call follows. A few ordinary operations begin to look like an agent.
This course slows that sequence down. Each lesson introduces one idea, shows small examples, and follows concrete values through execution. Repeated snippets keep the explanation close to the code, so you can read in order or return to the part you need.
The subject is understanding and engineering LLM agents with ordinary Python: structured outputs, tools, loops, context, reliability, permissions, approval, evals, and the choice to use fewer model-driven components. The lessons use the OpenAI Responses API and Pydantic to make those boundaries visible.
You can study the full series without installing anything, creating an account, or obtaining an API key. This is a reading curriculum; the code supports the explanations.
A companion to Learn the Monolith. Created and curated by Max Braglia. About this course.
The curriculum
Section titled “The curriculum”Foundations and the loop
Section titled “Foundations and the loop”- 01 · The LLM Call, Revisited as an Agent Primitive
- 02 · Structured Outputs as Contracts
- 03 · Tool / Function Calling
- 04 · The Agent Loop
State, reliability, and control
Section titled “State, reliability, and control”Evaluation and architectural judgment
Section titled “Evaluation and architectural judgment”Keep Python in charge
Section titled “Keep Python in charge”By the end, you should be able to trace a tool call, explain where state lives, identify the checks needed before an action, and say what an eval actually proves. You should also be able to recognize a task that ordinary Python or one model call can already solve.
Start here explains the prerequisites, example conventions, and technical assumptions.