Sarah Wang | AI + a16z Summary

Podcast: AI + a16z | Participants: Sarah Wang (a16z), Will Bryk (CEO, Exa)

1. Key Themes

Search Is Being Rebuilt From Scratch for a Non-Human Customer

The fundamental insight driving Exa is that agents are categorically different users than humans — and the entire architecture of search must reflect that. Google was optimized around human click behavior, which becomes largely irrelevant when the end user is an agent.

"The world of agents searching is just completely different from human searching. I guess you make the analogy of like agents to humans is like humans to sloths... agents as these like crazy creatures that have like infinite, like time is meaningless for them. They just want to like make complex queries very fast and like analyze it really fast." — Will Bryk [00:09:33]

"Human click data is great for humans, when you want to find results that humans click on... However, agents, like, just don't, they don't benefit that much from click." — Will Bryk [00:13:57]

Agentic Search Will Dwarf Google Search in TAM

Bryk makes a bold and specific prediction: agentic search will be a bigger business than Google Search by the early 2030s. The math is simple — humans make a few searches per day; agents will make millions. This reframes search not as a legacy category being disrupted, but as an entirely new market being created.

"If you follow that trend, even conservatively, you get to a massive TAM for agentic search in the 20s... the number of searches is going to be, we say thousands because that is like understandable and groffable to people, but really it's going to be millions... the world will be filled with search in a way that the world is filled with electricity." — Will Bryk [00:35:05]

"We think it will be bigger than Google Ads in 2030." — Will Bryk [00:35:35]

Retrieval Is the Solution to the Tokenpocalypse

As token costs balloon (Uber overspending, ServiceNow burning their annual budget early), retrieval-augmented small models are the structural fix. Exa positions itself not just as a search tool but as cost infrastructure for the agentic economy.

"Retrieval can help solve the tokenpocalypse because we should not be using gigantic models for every task... retrieval helps small models act like big models in a cheap way... we could save like 20x on cost for customers compared to other providers by being very efficient in what information does the agent actually see." — Will Bryk [00:26:38]

"Andre Carpathy had a tweet about this... the trend is towards smaller raw intelligence modules using tools... those weights should be focused only on like intelligent processing." — Will Bryk [00:27:38]

2. Contrarian Perspectives

Google's 20-Year Data Moat Is Largely Irrelevant for the Next Era of Search

Conventional wisdom says Google's click data and user signal gives it an insurmountable moat. Bryk disagrees — and has been saying so for years despite pushback.

"I remember saying this like years ago. People thought that was crazy. But it turns out to be right. Like human click data is great for humans... However, agents just don't benefit that much from click... all that click data that Google has accumulated just doesn't really matter for agents, and so it's a whole new ballgame." — Will Bryk [00:13:57]

LLMs Will Commoditize Faster Than Search

Most market observers treat foundation models as the durable value layer and assume search/retrieval gets commoditized underneath. Bryk inverts this entirely.

"I would argue that the LLMs are going to get commoditized or are getting commoditized faster than search is... most of knowledge work does not require the smartest model... a lot of knowledge work is actually a search problem, not only an intelligence problem." — Will Bryk [00:17:22]

Political Polarization and Loneliness Are Search Problems

This is a genuinely non-obvious framing that redefines the addressable scope of what a search company is solving. Most people see these as social, political, or psychological problems.

"Political polarization. I would argue that's a search problem because there are people out there who want to understand the world, but they're getting fed information that's just like misleading in some way or straight up wrong... Loneliness is a search problem... A lot of people are feeling lonely in modern society. Well, it's because they're not finding people to hang out with or to be in relationships with." — Will Bryk [00:19:45]

Building a Search Engine Today Is Both Easier AND Harder Than Beating Google

The intuitive take is "it's hard to beat Google." Bryk argues the reality is more nuanced — it's actually easier in certain dimensions because Google's advantages don't apply, but harder because the quality bar for agentic use cases is much higher than for consumer use cases.

"It's easier and harder to build a search engine for AI agents... you don't need all the click data... you could have a re-ranker that you just call an LLM and you have one engineer working on it... But it's also harder... billion-dollar investments are on the line, this has to be perfect... these new surgeons for agents need 99.99, 99.9999." — Will Bryk [00:13:27, 00:15:25]

Benchmark Maxing Has Made Retrieval Evals Meaningless

At a moment when many companies are competing on published benchmarks, Bryk openly states that retrieval evals are not trustworthy indicators of real performance — including likely his own competitors'.

"The evals have been bench maxed in retrieval. There aren't too many evals in retrieval... they've been bench maxed and they're not really actually good representations of agentic search. Like what agents actually need... customers can't really know what is true, which is sad." — Will Bryk [00:33:06]

3. Companies Identified

Exa (formerly Metaphor) Developer-facing semantic search engine rebuilt from the ground up for AI agents. Powers agentic search with support for complex queries, high result volume (up to 10,000 results), and token-efficient retrieval.

Why mentioned: Core company being discussed; powering Devin (Cognition), recommended by Claude itself, serving 5,000+ businesses.

"With Exa, you could search something and then get not just like 10 results or 100 results, but 1,000 results or 10,000." — Will Bryk [00:11:31]

Cognition / Devin AI coding agent company. One of Exa's high-profile customers.

Why mentioned: Validated Exa's search quality by running tests and finding meaningful accuracy improvements.

"A bunch of coding agents try us like Cognition, for example, we power Devin now and they've just found when they tested it that it just makes Devin way better, way more accurate, make way fewer mistakes." — Will Bryk [00:25:10]

Thinking Machines (Tinker) AI research infrastructure company.

Why mentioned: Exa used their tools in RL research on search tools, comparing SERP/Google-wrapping vs. Exa as the RL environment.

"There's a lot of efficiencies there, et cetera. I wanted to just flag that because it was an interesting finding. I think you used Tinker, so shout out to Thinking Machines." — Sarah Wang [00:29:07]

SpaceX Rocket and space transportation company founded by Elon Musk.

Why mentioned: Will Bryk interned there and directly witnessed the first barge landing; the experience shaped his leadership philosophy and mission-driven thinking.

"For example, I saw the first landing on the barge... just seeing that, like people coming together to do something magical. Like that was very inspiring and made a big impact on me." — Will Bryk [00:39:31]

4. People Identified

Will Bryk Co-founder and CEO of Exa. Former SpaceX intern. Has been obsessed with search and information quality since childhood, started building a mini search engine in college with co-founder Jeff, and pivoted toward neural/transformer-based search in 2021.

Why mentioned: Central guest; built a sub-100-person company that is outperforming Google in specific search domains for agents.

"How have we, a team that has always been below 100 people, been able to build a search engine that's better than Google in all sorts of ways? Well, it's because LLMs unlock new types of techniques." — Will Bryk [00:14:26]

Andrej Karpathy AI researcher, former Tesla AI director and OpenAI co-founder.

Why mentioned: Retweeted Exa's launch in November 2022 giving them early credibility; has also published thinking that aligns with Exa's thesis on small models + retrieval tools.

"Andre Carbethy retweeted it. It's pretty popular on Twitter. It was like this new way to find information. It was the first time people were like, holy cow, it's possible to find things beyond Google." — Will Bryk [00:07:06] "Andre Carpathy had a tweet about this... the trend is towards smaller raw intelligence modules using tools." — Will Bryk [00:27:38]

5. Operating Insights

The "Ego Flying Above the Company" Management Framework

Bryk articulates a specific and scalable approach to detail-orientation as a CEO — don't be in the weeds constantly, but maintain the right to dive deep selectively, then resurface. This solves the classic tension between founder detail obsession and the need to delegate at scale.

"I like the ego metaphor of like I'm an ego flying above the company. And then when I see some detail that I think should be important to fix, I go dive down and go into it and then come back up." — Will Bryk [00:40:31]

Let People Work on What They Want — Especially When Any Direction Improves the Business

Bryk gives engineers near-total latitude to switch domains (e.g., from vector DB to model training) based on personal excitement. He notes that in a well-defined mission space, individual passion and business need tend to naturally align — and AI tools fill the competency gaps.

"We had someone who built the vector database who was like, hey, I want to go train models. I was like, okay, just go do it... he's done amazing work. So I make sure everyone's working on exactly what they want to work on. And luckily in search, in any direction we improve things, it will be good for the business." — Will Bryk [00:46:25]

Naming Projects as a Force Multiplier for Culture at Scale

As companies grow, leaders can't be in every conversation. Bryk deliberately invests in naming projects with memorable, mission-aligned memes so those names carry the culture into conversations he's not part of.

"A name is really important because it's, when you have a company of a certain size, people are constantly communicating in ways you're not part of those conversations and the name really grounds the mission of that project... I could think for like a day just about a name." — Will Bryk [00:41:31]

Use Physical Space as a Recruiting and Brand Tool

The goat story is actually a deliberate (if playful) strategy: the office is positioned next to a popular coffee shop, and Bryk has used a fake goat with Exa branding to stop foot traffic, generate conversation, and even convert passersby into hires.

"We've already hired people because of the goat... people stop by. Like it's so beautiful. Even on Saturday and Sunday, like people come by and I go and I say hi to them. I talk to them. I tell them about Exa." — Will Bryk [00:43:00]

6. Overlooked Insights

Go-To-Market Intelligence Is Exa's Stealth Second Product — and They're Dogfooding It

Buried in the conversation is a signal that Exa is building out a specific "go-to-market intelligence" product — finding companies to sell to and people to hire. This is a massive, underserved B2B market (think ZoomInfo, Apollo, LinkedIn Sales Navigator) but with a fundamentally superior retrieval architecture underneath. Crucially, they are using it internally, which means they are building with real production feedback.

"That's something Exa is leaning very deeply into, like go-to-market intelligence because we care a lot about it. It's very exciting. It's also very useful to use internally. We have companies to sell to and we have people to hire. So it's been great to dog-food it and give us some advantage." — Will Bryk [00:18:49]

This is significant because the GTM intelligence market (sales prospecting, recruiting, competitive intelligence) is a multi-billion dollar category where incumbents have stale data and poor semantic search. Exa's architecture is purpose-built to solve exactly this — comprehensive retrieval, complex queries, 10,000+ results — and they are quietly moving into it while most observers focus only on their developer API.

The Infrastructure Bottleneck Before the Data Bottleneck — and the Investment Opportunity Hidden Inside It

Bryk briefly mentions that the first bottleneck to agentic search scaling isn't data or intelligence — it's raw search infrastructure capable of handling 100x-1000x query volume. He specifically calls out high-throughput vector databases as an area of active exploration. This is a non-obvious infrastructure investment thesis hiding inside a search conversation.

"Initially the bottleneck is going to be actually the infrastructure, which is kind of interesting. No one realizes, but if you actually get 10x, 100x, 1,000x more searches on Google, the infrastructure to handle that is insanely large. It just hasn't been built yet. So we're really excited to explore all sorts of cool new vector databases that have super high throughput." — Will Bryk [00:36:34]

This implies a coming infrastructure buildout in vector database and high-throughput retrieval systems that will parallel the GPU/compute buildout we saw with LLM training — but for inference-time retrieval at agentic scale. Companies building for this specific use case (high-throughput, low-latency vector search at massive scale) may be significantly undervalued relative to where demand is heading.