Ex-Twitter CEO Rebuilds Web Search for Agents, With Shapley-Value Payouts for Publishers
Training Data (Sequoia) · rss · 2026-08-25
Sequoia's Training Data podcast interviews Parallel Web Systems founder and CEO Parag Agrawal, the former Twitter CEO, on his bet to rebuild web search for AI agents.
- Core thesis: agents will query the web a thousand times more than humans ever have, and infrastructure built around human clicks is wrong for them; Parallel treats human click data as a bug and trains on agent feedback instead.
- Counterintuitive choices: shipping a search agent before a search engine, building the index incrementally, and a new Turbo product cutting agentic search to 200 milliseconds.
- The core problem is economic: the ad-supported internet collapses when agents show up instead of people. His fix uses Shapley values to pay content owners for the value their pages provide agents, with real dollars reaching publishers predicted within 12–24 months.
More from Companies & People
- ModelBest launches 'Forward Four' talent program with stock options for interns — 面壁智能 · 2026-08-25
- Why Anthropic might keep Mythos 5.1/2 internal — haider1 · 2026-08-25
- Opinion: China is teaching AI and robotics in primary schools — AryHHAry · 2026-08-25
- CirquarAI launches with world model as core tech route — 机器之心 · 2026-08-25
- Podcast: The L0–L3 Framework for Enterprise AI Proficiency and Non-Technical Builders — Practical AI · 2026-08-25
- Monthly Token Allocation Depleted in One Day: High Cost of AI Workflows — latticecut · 2026-08-25