How Exa is Building the Perfect Search Engine | Deep Dives with a16z

a16z Deep Dives · 2026-06-04 · 49 分鐘
https://www.youtube.com/watch?v=k6OkMkDdbUY影片總結
Exa CEO Will Bryk argues agentic search will surpass Google by 2030, with 10,000-result retrieval and 20x lower costs
Exa cofounder and CEO Will Bryk says Google remains optimized for fast consumer answers, while AI agents need a fundamentally different search engine: complex semantic and keyword queries, controllable filters, comprehensive results, and outputs scaling from 10 to 10,000 documents. Exa began in 2021, launched its first engine in November 2022—two weeks before ChatGPT—and now serves more than 5,000 businesses. Bryk says retrieval lets smaller models perform like larger ones, cutting some customer costs by up to 20x, while Exa’s coding search helps Devin make fewer mistakes. He expects agentic search to exceed Google Ads in the 2030s, driven by millions of automated searches, though infrastructure, data access, and retrieval quality remain bottlenecks. Exa’s research applies pre-training, post-training, and reinforcement learning to search; tests found RL with Exa outperformed SERP while using fewer calls. Bryk also describes a culture built around passion, autonomy, and obsessive pursuit of perfect search.
章節
- 0:00Exa Origin Story: From College Search Engine to a 2021 Mission
- 2:35Beating Google With LLMs: Deep Queries, $10 Billion Optimization, and the 2022 Launch
- 9:00Search Built For Agents: Complex Queries, Controls, and Complete Biotech Results
- 12:11Search Built For Agents: 10,000 Results, Custom Latency, and Smaller Teams
- 15:47Search Built For Agents: 99.9999% Reliability for Billion-Dollar Decisions
- 19:03Search Built For Agents: Go-to-Market Intelligence, Polarization, and Flying-Car Communities
- 22:30Search Built For Agents: Coding Agents, Devin, and the Trillion-Dollar Agentic Economy
- 26:01Tokenpocalypse and Retrieval: Exa Claims 20x Lower Costs with Smaller Models
- 29:21RL for Better Search: Exa Outperforms SERP with Fewer Calls
- 32:23Benchmarks and Agentic TAM: Search Could Exceed Google Ads by 2030
- 36:52Benchmarks and Agentic TAM: Infrastructure and Unrecorded Data Are Bottlenecks
- 40:19Benchmarks and Agentic TAM: Elon-Inspired Leadership, the Exa Goat, and 10¹⁸
- 44:20Culture Hiring and Wrap: Exa’s Fun-First Culture and Passion-Driven Hiring
這是 Tier 1 公開摘要
每章重點、段落總結、心智圖由分享者控制是否公開。想看完整分析?自己提交一支。
同頻道的其他分析
Mintlify and the Transition From Human Docs to Agent Infrastructurea16z Deep DivesMintlify’s Han Wang and Hahnbee Lee explain eight pivots, a two-day prototype, and docs becoming AI infrastructure.
To Regulate AI Effectively, Focus on How It’s Useda16z Deep DivesMartin Casado argues AI laws should target illegal use, while US regulatory uncertainty pushes startups toward Chinese open-source models.
How Palantir Scaled: Why the Best Software Is Built Backwardsa16z Deep DivesPalantir’s Akshay Krishnaswamy explains FDEs, backward product building, Foundry, and avoiding the consultancy trap.
Inferact: Building the Infrastructure That Runs Modern AIa16z Deep DivesInferact founders explain vLLM’s rise, 500,000-GPU scale, and a universal open-source inference layer
AI Copilots Are a Dead End. Here's What Actually Works | Kavak CEOa16z Deep DivesKavak uses AI agents for 90–95% of customer interactions after a flat 2023, then grows four times.
相關主題的分析
2026/8/27(四)輝達財報再超標!追高意願低?美股為何原地踏步?【早晨財經速解讀】游庭皓的財經皓角英偉達營收年增106%、資料中心占92%,AI需求強勁但表外承諾與資料中心延誤成最大風險
Rebuilding Git for AI Agents and The Future of Developer Tools | Deep Dives with a16za16z Deep DivesGitButler rethinks Git for AI agents with parallel branches, agent-ready CLIs, and patch-based code review
EP313. 谷歌痛失 AI 大將、Fable 5 配方解密、Meta 士氣低落中、FSD 即將入台 | M觀點M觀點Google痛失兩名AI大將,Anthropic揭露Fable 5靠Agent Harness變強,Tesla FSD送件台灣
How AI Will Rewire Consumer Internet | Deep Dives with a16za16z Deep DivesJosh Elman predicts consumer AI will surpass productivity, with Siri, ChatGPT, Robinhood, and TikTok illustrating retention-driven growth.
EP322. 谷歌新模型不夠力、川普要查中國 AI、OpenAI 模型越獄 | M觀點M觀點Google Gemini 3.6 Flash落後GPT-5.6與Grok 4.5,Kimi K3蒸餾爭議延燒,OpenAI模型更越獄入侵Hugging Face