What an agent actually is
The loop, the state, the tools. Where the magic is and where it is not.
First to openAI Agents
From a single tool call to a system that plans, acts, checks its own work and knows when to stop. Built the way production teams build them.
CourseComing soonIntermediateWriting now
What you will be able to do
- Design an agent loop with explicit state, tools and stop conditions
- Choose between a single agent, a router and a team of agents, and say why
- Add memory, retries and human approval without losing control of cost
- Evaluate an agent on tasks, not vibes, and ship it behind a service
Planned modules
7 modules · order may change
Tools and structured output
Schemas, validation, and what to do when the model returns almost-valid JSON.
PlannedPlanning and control
Plan-then-act, reflection, budgets, and stop conditions that hold.
PlannedMemory
Working memory, episodic memory, and when a database beats a prompt.
PlannedMulti-agent systems
Routers, specialists and supervisors. When two agents are worse than one.
PlannedEvaluation and safety
Task suites, traces, guardrails, and the incident you will have anyway.
PlannedShipping
FastAPI service, queues, observability, and the cost line.
PlannedTell me when it opens
One email from the letter when it goes live. Unsubscribe any time.
The announcement comes through the letter: one email when part 1 goes live, free like every first part. Unsubscribe any time from the email or your account.