Treating Go-To-Market as an AI Engineering Problem
AI Engineergo watch the original →
the gist
Exa treats go-to-market as a data-driven engineering challenge by building internal agentic systems, including a personal AI clone and automated market intelligence dashboards, to scale a lean team.
Automating Go-To-Market Operations
Exa treats go-to-market (GTM) as a data problem, maintaining a live model of their addressable market that agents can query and act upon. The team utilizes two primary interfaces to manage this data: an ICP dashboard that classifies the total addressable market using Exa's own semantic embeddings to estimate annual spend, and "Request Lens," a signal-monitoring system that triggers alerts based on customer activity like signups or usage surges. By treating GTM as an AI engineering problem, the team maintains a lean operation where forward-deployed engineers (FDEs) both support sales deals and build the internal tooling that automates them.
Building Agentic Systems
To scale decision-making, the company deploys a dozen agents within Slack that have programmatic access to internal data. A notable implementation is "Jeffbot," an AI clone of the co-founder built during a one-week break using OpenAI's o3-mini (referenced as Opus 4.5 in the source). The agent was calibrated by analyzing 760 emails to mimic the co-founder's tone—specifically averaging 18 words per email and using "best" as a sign-off—and by creating evals from hundreds of past Slack and email decisions. While the co-founder grants the agent read and write access to all systems, the agent is restricted to drafting messages when used by other employees, ensuring security boundaries remain intact.
Principles for Agent-First Organizations
- API-First Architecture: Agents require programmatic access to data via interfaces like MCP or custom CLIs; without robust APIs, agentic automation is impossible.
- Hybrid UI Strategy: Not every interaction requires a generated chatbot interface. Consistent, crystallized UIs remain superior for repetitive, high-frequency tasks, while chat agents provide flexibility for ad-hoc queries.
- Arbitrary Customizability: The buy-versus-build debate is a false dichotomy. Whether purchasing SaaS or building internal tools, the primary requirement is that the system must be arbitrarily customizable to fit the company's specific workflows.