Rayyan Zahid · Claude Code Build Night · Bright Data SF · 20 Oct 2026

How should agents search?

Five talks, one question. Start with E, the combined talk. Each one starts from the basic terms, shows what we measured on 20 Oct 2026, and ends with something you can build in an evening with Claude Code.

E

Search in the age of agents · the main talk

All four sides at once: where an agent's facts come from, how to route and check them (the model's own memory agreed with search 27 of 27 times), how to be found by other people's agents, and your own library as a search engine.

Full talk
A

Search API Battle Royale

Eight search APIs, the same 85 questions. Who finds what, where they overlap, and when search matters at all.

Full talk
B

System 1 vs System 2 search

Quick search, deep research, and a router that decides. Claude's own confidence beat both trained routers.

Full talk
C

Self-improving RAG

Our research brain, a loop that learns from its misses, and a live demo of teaching it one question in twelve minutes.

Full talk
D

Agent-native search platforms

What search built for agents looks like. Four public APIs walked by hand, and how one site went from 52 to 100.

Full talk