GTM as a Data Problem

Jeffrey Wang argues that the traditional debate between "product-led" and "distribution-led" growth is a false dichotomy; both are essential. For modern companies, GTM should be treated as an AI engineering problem. The core challenge is that GTM teams lack a live, actionable model of their world. To solve this, Exa built a system that integrates internal data (product usage, customer interactions) with external data (web-wide intelligence) to create a unified, queryable model.

Core Systems and Interfaces

Exa utilizes two primary interfaces to manage this data:

  • ICP Dashboard: A semantic classification tool built on Exa’s embeddings that maps the entire Total Addressable Market (TAM). It provides deep-dive metadata and anticipated spend for every company in their target segments.
  • Request Lens: An alerting system that monitors significant customer events (e.g., signups, usage spikes, churn signals), allowing the team to act on real-time data rather than static lists.

Beyond these, the company employs a dozen specialized agents within Slack. These agents allow team members to perform complex research, generate demos, and draft communications by querying internal data sources directly.

The "Jeffbot" Experiment

Wang developed "Jeffbot," an AI clone trained on 760 of his own emails and hundreds of past decisions.

  • Calibration: He used past decisions to create evals, calibrating the agent’s judgment to match his own.
  • Constraints: While the agent has read/write access to his systems, it is restricted when used by others. When employees use Jeffbot, it is limited to drafting messages and lacks the full administrative permissions granted to Wang himself.

Principles for Agent-First Organizations

  • API-First Architecture: Agents are only as effective as their access to data. If internal systems lack programmatic interfaces (APIs/MCP), agents cannot function.
  • Hybrid UI Strategy: Not everything should be a chatbot. While LLMs can generate dynamic UIs, there is significant value in consistent, crystallized UIs that users can learn and master over time.
  • Arbitrary Customizability: The "build vs. buy" debate is outdated. The priority is whether a system—whether purchased SaaS or custom-built—is arbitrarily customizable. Using tools like Salesforce is effective only if they expose APIs that allow agents to interact with the data programmatically.
  • The Forward-Deployed Engineer (FDE): Wang notes that AI has collapsed the gap between sales engineering and software engineering. An FDE can support revenue generation while simultaneously building the internal tools required to automate and smooth out the GTM process.