Weeks 1 through 3 gave your agents knowledge locked inside a model. Every agent answered from its training data. What's missing when you ask it to pull today's regulatory update or verify a live vendor policy?
It can't. The model was trained months ago. For anything current — market shifts, updated guidance, a competitor's latest pricing — you'd need the agent to actually search the web in real time.
That's the first upgrade this week. Swapping in a search-capable model means the agent queries the live web before answering. Day 24 is a straight web search. Day 25 chains that search into a structured extraction — so a research result becomes a clean Python dict your pipeline can consume. Days 26 and 27 introduce @agent.tool_plain: a Python function you register so the model can call it when it needs an exact answer instead of a guess. Day 28 is the capstone — search, extract, and format into a report in one function.
The agent is picking the tool — I just defined them. That's a completely different mental model from writing a prompt and hoping.
Exactly the shift. By Friday you have an agent that decides when to search and when to compute — and hands you a formatted output ready to send.
result_type=Fact@agent.tool_plain — one Python tool the agent can callGoal: agents decide when to search and when to compute.
Create a free account to get started. Paid plans unlock all tracks.
Weeks 1 through 3 gave your agents knowledge locked inside a model. Every agent answered from its training data. What's missing when you ask it to pull today's regulatory update or verify a live vendor policy?
It can't. The model was trained months ago. For anything current — market shifts, updated guidance, a competitor's latest pricing — you'd need the agent to actually search the web in real time.
That's the first upgrade this week. Swapping in a search-capable model means the agent queries the live web before answering. Day 24 is a straight web search. Day 25 chains that search into a structured extraction — so a research result becomes a clean Python dict your pipeline can consume. Days 26 and 27 introduce @agent.tool_plain: a Python function you register so the model can call it when it needs an exact answer instead of a guess. Day 28 is the capstone — search, extract, and format into a report in one function.
The agent is picking the tool — I just defined them. That's a completely different mental model from writing a prompt and hoping.
Exactly the shift. By Friday you have an agent that decides when to search and when to compute — and hands you a formatted output ready to send.
result_type=Fact@agent.tool_plain — one Python tool the agent can callGoal: agents decide when to search and when to compute.