A16Z Podcast Host | The a16z Show Summary

1. Key Themes

The Agent-Native Search Paradigm Is Fundamentally Different From Human Search

The core thesis of Exa is that agents are a completely different type of "creature" than human searchers, and building search optimized for humans (as Google has done) is the wrong architecture for the agentic era. Agents want comprehensiveness, complex semantic queries, controllability, and extreme customization — not the 10 blue links optimized for tired humans typing short keywords.

"The world of agents searching is just completely different from human searching. An agent doesn't just want 10 pieces of information. It wants everything." — Will Bryk [00:01:07]

"You should think of 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 [00:09:39]

Google's 20-Year Moat (Click Data) Is Largely Irrelevant for Agentic Search

One of the most non-obvious structural advantages for Exa: Google's most defensible asset — billions of human click signals used for PageRank — simply does not transfer to the agent use case. This levels the playing field dramatically for a startup.

"Human click data is great for humans when you want to find results that humans click on, which is obvious. However, agents, like, just don't, they don't benefit that much from click... it's interesting that, like, 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 [00:13:54]

"How have we, a team that, you know, 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 [00:14:49]

Search Is Infrastructure for the Agentic Economy — Potentially Larger Than Google Ads

Bryk argues that search will become like electricity — invisible but fundamental — and that agentic search will exceed Google's ad business in scale by the 2030s. The core logic: agents will make millions of searches per day per user, versus the 2-3 human searches per day currently.

"If you just play out the numbers, we think it will be bigger than Google Ads in 2030... Humans on average make, you know, a couple searches a day. But agents, when everyone has a personal assistant and every single software tool you use is going to be checking its work with retrieval... the number of searches is going to be... millions at some point. It's just going to be like the world will be filled with search in a way that the world is filled with electricity." — Will Bryk [00:35:01]

2. Contrarian Perspectives

LLMs Will Commoditize Faster Than Search

The prevailing assumption is that foundation models are the durable moat and that search/retrieval is a commodity layer. Bryk inverts this — arguing that most knowledge work doesn't require the smartest model, and open source models are rapidly closing the gap on capability. Search, by contrast, requires every extra "nine" of quality.

"I would argue that the LLMs are going to get commoditized or are getting commoditized faster than search is. And the reason is because you don't need to run, like, mythos over every cell on your Excel sheet when you're trying to find competitors. Like, most of knowledge work does not require the smartest model." — Will Bryk [00:16:51]

"Every extra nine of quality in search really matters." — Will Bryk [00:17:54]

Smaller Models + Retrieval Will Beat Larger Standalone Models — And This Becomes Obvious by End of 2026

Against the current trend of scaling to ever-larger frontier models, Bryk argues the winning architecture is tiny, hyper-intelligent models that are "knowledgeless" — offloading all factual retrieval to external search tools. This is more efficient, cheaper, and ultimately more performant.

"You should use a family of models of different sizes. The big model decides what to do, and it dishes out commands to the small models. And those small models can be way more accurate and reliable if they're using retrieval. So, retrieval helps small models act like big models in a cheap way." — Will Bryk [00:26:39]

"You could probably get to models that are like one billion, even less than a billion parameters that are extremely hyper-intelligent and completely unknowledgeable. Like, they just, it's like Einstein, like, who never saw the world." — Will Bryk [00:27:49]

"By end of 2026, it's very noticeable." — Will Bryk [00:28:44]

Political Polarization, Loneliness, and Other Social Problems Are Actually Search Problems in Disguise

Most people frame these as cultural or social media problems. Bryk reframes them as fundamentally information retrieval failures — and argues a perfect search engine would directly address them.

"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, you know, misleading in some way or straight up wrong. And if everyone had, like, information that was accurate and, like, controllable and, like, comprehensive, I think most reasonable people would be reasonable." — Will Bryk [00:19:23]

"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 [00:20:06]

The Benchmark Ecosystem for Retrieval Is Fundamentally Broken — and Nobody Talks About It

The retrieval/search space has the same benchmark-gaming problem as LLMs, but with fewer third-party benchmarks and less scrutiny. Customers who rely on published evals are being misled.

"The evals have been bench maxed in retrieval. There aren't too many evals in retrieval. That's one problem. There aren't that many standard third-party retrieval evals, and they've been, like, bench maxed. And they're not really actually good representations of agentic search." — Will Bryk [00:32:45]

3. Companies Identified

Exa

Description: AI-native search engine built specifically for agents and developer/enterprise use cases, offering semantic search, keyword search, and up to 10,000 results per query via API. Why Mentioned: Core subject of the episode; positioned as the emerging infrastructure layer for agentic search, already powering major AI products.

"With Exa, like you could search something and then get not just like 10 results or 100 results, but 1,000 results or 10,000." — Will Bryk [00:01:07]

Cognition / Devin

Description: AI coding agent company; Devin is their flagship autonomous software engineer. Why Mentioned: Named as a customer that tested Exa and found it meaningfully improved Devin's accuracy and reduced errors.

"When, you know, a bunch of coding agents try us, like Cognition, for example, we've talked about, tried us, like, 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 [00:24:52]

Thinking Machines (Tinker)

Description: AI infrastructure/tooling company referenced in context of RL research. Why Mentioned: Exa used their tooling for reinforcement learning experiments on search, referenced as a shout-out in the context of cutting-edge research.

"I think you used Tinker. So shout out to Thinking Machines." — Sarah Wang [00:29:34]

Hedge Coffee

Description: A coffee shop that opened next door to Exa's San Francisco office. Why Mentioned: Indirectly highlighted as a foot traffic catalyst that Exa leveraged for brand awareness and recruiting through their GOAT mascot.

"There was a coffee shop that opened up next door. And so many people, so, like, the mayor of SF kept posting about it... so many people are walking by. We're like, we got to, like, have them stop and look at what EXA is." — Will Bryk [00:41:55]

4. People Identified

Andrej Karpathy

Description: AI researcher and former Tesla AI Director / OpenAI co-founder; extremely influential voice in the ML community. Why Mentioned: Retweeted Exa's launch in November 2022, providing early signal/validation. Also cited for a prescient tweet about the trend toward smaller intelligence modules using tools.

"Andre Carbethy retweeted it. It was pretty popular on Twitter. It was like this new way to find information." — Will Bryk [00:07:12] "Andre Carpathy had a tweet about this a couple years ago, where it was, like, the trend is towards, like, smaller raw intelligence modules using tools. And that trend will, like, that's an important trend." — Will Bryk [00:27:18]

Jeff (Exa Co-Founder)

Description: Co-founder of Exa, former college roommate of Will Bryk. Why Mentioned: Built early crowdsourced search prototype with Will in college; foundational partner in Exa's creation.

"In college, I was roommates with Jeff and we were like, we could just build a better search using crowdsourcing the highest quality links." — Will Bryk [00:01:46]

Elon Musk

Description: CEO of SpaceX, Tesla, and X; serial entrepreneur and technologist. Why Mentioned: Will Bryk interned at SpaceX and credits Musk's leadership style — detail orientation, memetic naming, and mission framing — as direct influences on how he runs Exa.

"He's very detail-oriented. Like, he gets into details of everything... I do that, too, at Exa. Like, everything from, like, the algorithms, like, the VectorDB algorithm to, like, the office space." — Will Bryk [00:39:46] "He's very good at memetic, like, names and, like, inspiring through, like, memetic things. Like, you know, like, you realize that SpaceX's mission is not, like, improve rockets to get to orbit. It's, like, make humanity interplanetary." — Will Bryk [00:40:30]

5. Operating Insights

Let Engineers Self-Direct Their Work — Alignment of Passion and Business Need Is Often Emergent, Not Planned

Bryk's approach to talent retention and performance is unconventional: let people work on whatever excites them most, even if they lack direct experience. In a broad enough problem space (like search), almost any direction of improvement benefits the business. This creates retention, output quality, and serendipitous breakthroughs.

"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. And, like, he didn't really have experience in training models, but I was like, you're really smart. You'll learn it in, like, two weeks. And then you'll just do amazing work, and he's done amazing work." — Will Bryk [00:46:04]

Use Memetic Project Names as Internal Coordination Infrastructure

At scale, leadership can't be in every conversation. A well-chosen name for a project or initiative becomes a self-propagating coordination mechanism that keeps teams aligned on mission without requiring managerial overhead.

"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, like, the name, like, really, like, grounds the mission of that project... I could think for, like, a day of just about a name." — Will Bryk [00:41:26]

Use Physical Space and Unexpected Brand Touchpoints as a Low-Cost Recruiting and Brand Tool

Exa placed a fake GOAT mascot outside their office near a high-traffic coffee shop, turning casual foot traffic into brand awareness and actual hires. This is a replicable playbook for any startup in a high-foot-traffic urban location.

"We've hired people because of the GOAT." — Will Bryk [00:42:58]

Hire for "Fire in the Eye" / Passion Over Credentials — Increasingly Critical in an Agentic World

As agentic tools level the skills playing field, passion and agency become the primary predictor of output. Bryk interviews every single candidate personally and screens for this above all else.

"I look for passion. It's like someone's just like a fire in their eye. Like, do they really care about what we're doing?... The world goes to who is most passionate and who is most agentic. Because, like, you can now do anything... what matters most is, like, how much do you care about the end result." — Will Bryk [00:48:19]

6. Overlooked Insights

The Agentic Economy Requires a New Revenue-Sharing Model With Content Providers — And Whoever Solves This Owns the Distribution

Bryk briefly and almost casually describes a future where content providers earn more from the agentic economy than they do today from ad-driven web traffic — a complete restructuring of the internet's economic model. This is not just a philosophical point; it's a product and business model that Exa appears to be actively thinking about building. Whoever architects this distribution layer — the "toll road" between agents and content — could capture enormous value and also solve the web's "closing" problem preemptively.

"Instead of, like, you know, hundreds of billions of dollars going to one company, what if, you know, $50 billion a year went to one company and the other $150 billion went to all the content providers, right? Like, there's ways to distribute the value in this new agentic economy that are more favorable towards providers of content... it won't be $200 billion, it'll be a trillion dollars a year of value. So there's, if we could figure this out well, like, there's an opportunity for everyone to just, like, do amazingly." — Will Bryk [00:23:09]

RL Training on Better Search Tools Produces Meaningfully Superior AI Agents — An Underexplored Research Finding With Major Implications

Bryk mentions almost in passing that Exa ran an experiment reinforcement learning agents on Google-wrapped SERP results versus Exa's search — and the Exa-trained agents performed better AND used fewer search calls. This is a profound finding: the quality of the search tool you train your agent on shapes the intelligence of the agent itself. This means search infrastructure is not just a runtime dependency — it's a training dependency. Companies building foundation models or coding agents who ignore this are potentially capping their model quality.

"We simply RL'd on SERP. So, like, Google Wrapping versus Exa. And found that, you know, RLing on Exa does way better. Like, it both, like, uses fewer calls. So it's more efficient. And then it's, like, higher performance. And this makes sense because, again, like, Exa was designed for agents to use." — Will Bryk [00:30:04]