AI Engineer Search Track: 30-Year-Old BM25 Emerges as Backbone of Agentic Search

jobergum · x · 2026-09-17

AI Engineer World's Fair 2026 streamed its Search & Retrieval track, centered on the thesis that "the searcher changed before the search engine did" — machines are starting to search more than people do.

Talks include:

Key takeaway: BM25, a scoring function essentially unchanged for 30 years, is proving surprisingly effective for agent search, and search infrastructure is being rebuilt around machine users.

Original post →

More from coding & agent

coding & agent channel →