Sunny Rekhi separates enterprise AI deployment into configuring an agent's behavior and building reusable product capabilities. Instructions, tone and human handoff rules must fit each customer, while recurring requests should shape the shared platform.
Sunny Rekhi describes a shift from generalist agent engineers to specialized agent builders and software engineers as Decagon grows. The talk emphasizes agreeing on measurable outcomes early, assigning people with relevant industry experience and demonstrating value before expanding a deployment.
Sunny Rekhi argues that cheaper AI-assisted coding makes restraint more important: rapidly generated patches can create brittle systems. Repeated integrations should become self-service features, and field teams should advise customers about priorities rather than simply executing every request.
Watch on YouTube
