There is a failure mode that is especially lethal for small agent companies with engaged customers: following every customer into their specific use case until you have built three different products instead of one.
The pattern is predictable. Customer A wants deterministic workflow agents — parsing alerts, executing playbooks, chaining discrete steps with structured outputs. Customer B wants staff augmentation agents — research that hands off to PRD writing that hands off to ticket creation that hands off to Slack notifications. Customer C wants something else entirely. Each request is reasonable. Each customer is engaged and willing to collaborate. And if you follow each one, you end up with a product that is three things and none of them well.
There is a Chinese proverb that captures the discipline required: pluck a feather from each passing goose, but follow no goose absolutely.
The mechanism for maintaining this discipline is a structured design partner program. Not informal conversations. Not ad hoc feature requests in Slack channels. A real program with regular cadence — biweekly syncs, prototype reviews, shared context on what other partners are asking for. The program serves two functions: it gives you a venue to learn what customers actually need (as opposed to what they say they want), and it forces you to synthesize across multiple customers rather than optimizing for any single one.
The synthesis is the hard part. When one customer wants workflow-driven agents with deterministic steps and another wants autonomous agents that take initiative and hand off to humans only when blocked, the temptation is to build both. The discipline is to find the architectural primitive that serves both — and to know that you will not find it by building in a vacuum. I've argued elsewhere that provocatypes over roadmaps is the method that surfaces the underlying primitive when the product shape is genuinely unknown. The design partner program is the venue. The provocatype is the artifact you bring to it.
Related Essays
Provocatypes Over Roadmaps
A provocatype is a prototype built to provoke conversation, not to ship. When the product shape is genuinely unknown, provocatypes surface assumptions that requirements gathering misses.
The Long Game in Agent Companies
Agent companies are marathons you sprint through. Frameworks commoditize, monetization pins to infrastructure, and proximity to real builders is the moat.
The Human Is No Longer the Integration Layer
The ERP is not going away. The CRM is not going away. The landing page is not going away. But the human as the operating system between them is. The agent takes that job now.
Key takeaways
- The lethal failure mode for small agent companies is following every engaged customer until you have built three different products instead of one.
- A structured design partner program — biweekly cadence, prototype reviews, shared context — beats ad hoc requests in Slack.
- The discipline is in synthesizing across partners to find the architectural primitive that serves them all, not in optimizing for any single one.
FAQ
Isn't following customers the whole point of customer development?
Following individual customers is fine. Following all of them simultaneously is fatal — you build divergent product surfaces that share no underlying primitive. The discipline is synthesis across customers, not service to each one.
How do you say no to an engaged customer without losing them?
You say yes to the underlying need and no to the specific implementation. The phrasing is "we are not going to build that, but we are going to build the primitive that lets you build it." Engaged customers respect this — disengaged customers were never going to stay anyway.