The Frontier

Will Bryk co-founded Exa to build a perfect search engine, believing it would have profound positive downstream effects across all industries and human life, a mission he has pursued since childhood.

Bryk argues Google is optimized for consumer use cases and human clicks, excelling at delivering quick answers but failing at deep, complex queries. He cites his own difficulty using Google to research the lived experience in the Roman Empire.

Exa's search engine was designed from day one to serve AI agents, which Bryk claims have fundamentally different needs than humans, including the ability to handle complex semantic queries and demand for comprehensive, not just top-10, results.

Bryk states Exa's small team built a superior search using LLMs, rendering Google's vast human click data advantage largely irrelevant for agentic search. He says this technological shift allowed them to bypass decades of incumbent optimization.

He claims that while it's easier to start an agentic search engine today, achieving perfection is far harder, as business use cases demand near-100% accuracy for high-stakes decisions like billion-dollar investments.

Bryk argues LLMs will commoditize faster than search. He asserts many knowledge work tasks are fundamentally search problems, and every extra 'nine' of search quality matters more than raw model intelligence for applications like go-to-market intelligence and recruiting.

He frames major societal issues like political polarization and loneliness as search problems, claiming accurate, comprehensive information would make reasonable people reasonable and help individuals find compatible communities.

For coding agents, Bryk states Exa provides fresher, higher-quality retrieval over technical documentation and SDKs, which reduces errors. He cites Devin by Cognition as a customer that found Exa's search made its agent more accurate.

Bryk says search retrieval can alleviate the 'token apocalypse' by letting smaller, cheaper LLMs use tools to act like larger models. Exa's research focuses on extracting only the most relevant document snippets to cut customer token costs.

He predicts the agentic search economy will be bigger than Google's ad business by the 2030s, based on extrapolating the number of LLM calls that will require search, which he believes will be orders of magnitude greater than human searches.

Bryk claims future bottlenecks will shift from model intelligence to infrastructure for handling massive query volumes and then to data access, as agents will need to query information not currently on the public web.

Exa operates like a research lab, applying LLM training techniques like pre-training, post-training, and reinforcement learning to search models. Bryk says serving over 5,000 diverse businesses pushes their research in many directions.

Bryk criticizes the current retrieval evaluation ecosystem, noting public benchmarks are maxed and don't reflect agent needs. He says sophisticated customers rely on their own A/B tests, which he claims favor Exa.

Bryk's leadership philosophy, influenced by his internship at SpaceX, emphasizes deep detail orientation, memetic mission naming, and ensuring employees work on projects they are passionate about to maintain excitement and high performance.

George Frazier argues AI agents need access to centralized business data to function effectively, creating a new imperative for data consolidation beyond traditional business intelligence.

Frazier notes some SaaS vendors like SAP are reacting to AI by restricting API access for agents, viewing them as a threat to their business models. He calls this a bad strategy that harms customers.

Frazier contends the fear that AI agents will reduce SaaS seat counts is overblown, as software costs are a minor part of business budgets and AI's primary use is to improve operations, not cut software spend.

FiveTran's open data infrastructure website scores vendors on data access policies, giving yellow ratings for charging egress fees and red for actively blocking access. Frazier says SAP has historically been bad while Salesforce was good but is getting 'squirrelly.'

Frazier believes the concept of 'data gravity' - that moving data is prohibitively expensive due to egress fees - is fake, arguing change data capture makes data movement manageable and that companies should centralize their data in a platform they control.

He advises companies to write data access guarantees into Master Service Agreements with large vendors, noting FiveTran provides model language for this. He says vendors often back down when challenged.

Frazier observes that early AI agent implementations often give agents human-like identities to slot into existing workflows, but sees this as an intermediate step before more integrated, closed-loop systems.

He argues most enterprise agent interactions will happen via APIs, not browser automation, because APIs are faster and consume fewer tokens. He notes Salesforce's CLI is comprehensive enough for agent use.

Frazier sees value in Model Context Protocol servers for agents, despite theoretical redundancy, because they solve practical problems like authentication and tool discoverability in real systems.

He disagrees with predictions of a 'SaaSpocalypse,' arguing the bigger threat is AI-native startups out-innovating incumbents. He notes established companies like FiveTran are accelerating, not slowing, despite market uncertainty.

Frazier says AI coding agents are an 'infinite supply of junior engineers' that help FiveTran troubleshoot its 750+ data connectors, leading to incremental quality improvements rather than replacing core engineering.

He states AI labs like OpenAI and Anthropic use FiveTran for typical data consolidation and analytics, building data foundations that look like any other modern company's, not exotic new systems.

Frazier argues AI is creating more demand for infrastructure, not commoditizing it, though it may threaten the highest 'consumption layer' of user-friendly platforms as agents can navigate more complex systems directly.

He claims PostgreSQL is outdated technology with a poor storage engine, and that undergraduates in database courses write better storage engines. He believes the world should create a new operational database instead of endlessly repackaging Postgres.

Frazier's main existential worry for FiveTran is AI coding agents becoming so good at writing data connectors that customers shift to DIY. His biggest excitement is AI creating a massive new set of use cases for organized data.

Benedict Evans argues AI's current product market fit is coding, where customer pull is absolute, versus widespread uncertainty about other daily use cases. He points to GitHub Copilot's revenue explosion as the concrete signal.