The Discipline of Scoping

Forward Deployed Engineering (FDE) is often misunderstood as simply saying "yes" to every customer request. In practice, this leads to "feature bloat" and wasted engineering effort. The core responsibility of an FDE is to act as a filter, not a conduit. When faced with urgent requests—such as a last-minute SAP integration to close a deal—the FDE must pause to validate the necessity of the request. This involves asking:

  • Is the urgency driven by the customer or by internal sales quotas?
  • Have all existing workarounds or manual processes been exhausted?
  • Does the customer have the technical capacity to use existing APIs instead of requiring a custom build?
  • Does this request provide value to other prospects or the broader product roadmap?

Failure to scope properly can lead to disastrous outcomes, such as building features for platforms the customer does not even use. Rigorous scoping ensures that engineering resources are focused on high-leverage work that actually moves the needle for the business.

Scaling with Tokens: The Agentic Pipeline

Once scoping discipline is established, the next imperative is to scale the FDE pipeline using AI agents. The goal is to treat the FDE lifecycle—from intake to implementation—as a factory that can be automated.

At Ramp, the team automated the request intake process using agents integrated with Notion. Instead of engineers manually reading and clarifying vague requests, an agent now:

  • Engages in back-and-forth questioning with the requester to gather necessary details.
  • Validates the request against product capabilities.
  • Produces a structured spec only when the request is sufficiently defined.

This automation reduced the latency of replies from days to seconds and saved approximately 20% of the time previously spent on manual scoping.

The Future: Combining Judgment and Automation

Moving forward, the FDE role will shift toward building and maintaining an "agent harness." This involves:

  • Evals and Rubrics: Implementing human-in-the-loop feedback to ensure agent outputs meet quality standards.
  • Context Injection: Feeding agents the right data, including historical request patterns, product documentation, and help articles, to ensure they possess the "tribal knowledge" of a product manager.
  • Judgment: While agents handle the volume, the FDE remains responsible for the final "taste" and strategic judgment.

The ultimate success of an FDE team depends on a dual strategy: using manual scoping to prevent building the wrong things, and using AI agents to build the right things at scale. Without scoping, agents simply become a "token-maxing slop cannon"; without agents, the team will be outpaced by more efficient, agent-native competitors.