Build a Research Agent That Reads the Web For You

​Ask a question at nine, have a sourced brief by five past.
That part is genuinely easy now, which is exactly the problem.

​A research agent's failure mode is not that it can't find things. It's that it hands you six paragraphs where five are solid and one is confidently invented, and nothing in the formatting tells you which.

​Read enough of those and you stop trusting any of it — which puts you back to doing the reading yourself, having paid for the tool. So this session is about building the other half: an agent that searches, reads, and synthesizes, but where every claim is tied to a source you can click, and where it will tell you plainly that it could not verify something rather than smoothing over the gap.

​We wire in real web search live and point it at a question none of us have answered yet. Useful for competitor tracking, market research, and due diligence — anywhere the cost of a wrong sentence is higher than the cost of a slow one. **You'll leave with:** a reusable research agent and the prompt patterns that keep its output sourced. **Format:** 60 minutes building, 20 minutes questions, then whoever wants to stay and build sticks around.

This will be in person, seats are limited.