In the explorers track, agents used retrieval — live search pulled facts the model did not know. This week you push one step further: the agent not only reads the world, it decides what to do next.
So retrieval was the input side. Decisions are the output side — picking a label, picking an order, picking which source to act on?
Exactly. Five decision shapes in five days. A single label out of three. A list of steps. A structured observation with urgency and summary. A ranking over search results. And a two-call refinement where the agent critiques its own draft.
And each of these is one function I write — no multi-file frameworks, just Python with a PydanticAI agent?
One function per lesson, one new decision shape per day. By Friday your code can classify, plan, observe, rank, and refine — the full decision vocabulary of an agent.
Literal["escalate","automate","ignore"] as result_type for closed-set decisionsresult_type=list[str] for plans the agent generates step by stepObservation(urgency, summary) for multi-field readsGoal: by Friday you can shape any agent decision as a typed call.
Create a free account to get started. Paid plans unlock all tracks.
In the explorers track, agents used retrieval — live search pulled facts the model did not know. This week you push one step further: the agent not only reads the world, it decides what to do next.
So retrieval was the input side. Decisions are the output side — picking a label, picking an order, picking which source to act on?
Exactly. Five decision shapes in five days. A single label out of three. A list of steps. A structured observation with urgency and summary. A ranking over search results. And a two-call refinement where the agent critiques its own draft.
And each of these is one function I write — no multi-file frameworks, just Python with a PydanticAI agent?
One function per lesson, one new decision shape per day. By Friday your code can classify, plan, observe, rank, and refine — the full decision vocabulary of an agent.
Literal["escalate","automate","ignore"] as result_type for closed-set decisionsresult_type=list[str] for plans the agent generates step by stepObservation(urgency, summary) for multi-field readsGoal: by Friday you can shape any agent decision as a typed call.