Summary
Jeffrey Wang, co-founder of Exa — the search engine built for AI agents and powering tools like Cursor and Cognition — makes the case that go-to-market is fundamentally a data problem, and that engineers should treat it as such. Exa is positioned as a web-access layer for agents: providing real-time external data that agents cannot otherwise reach. The talk weaves together Exa’s product identity with concrete examples of their own internal GTM infrastructure.
The most distinctive part of the talk is JeffBot, Wang’s attempt to build an AI clone of himself using Opus 4.5. He analyzed 760 of his own emails to extract voice characteristics (18-word average length, ending with “best” rather than “sincerely”), modeled hundreds of past decisions to create a calibrated decision-making framework with accompanying evals, and gave the agent read/write access to every system he has access to as CEO. The result is a company-wide tool that anyone at Exa can use to draft Slack messages or decisions in Wang’s voice.
Wang also describes two other internal GTM primitives: an ICP discovery system that uses Exa’s own search to generate large lists of potential customers, and Request Lens, a signal system that alerts the team any time a significant customer event occurs — a signup, a usage spike, a dropoff. The broader argument is that a dozen specialized agents inside Slack, each with access to internal company data, is now a practical and affordable GTM infrastructure for a lean team.
📺 Source: AI Engineer · Published August 26, 2026
🏷️ Format: Workflow Case Study
