Latent Demand Discovery Through Agent Usage
The Idea
In traditional product development, you imagine what users want, build it, and see if you were right. In agent-native development, you build a capable foundation, observe what users ask the agent to do, and formalize the patterns that emerge.
When users ask the agent for something and it succeeds, that’s signal. When they ask and it fails, that’s also signal — it reveals a gap in your tools or parity. The agent becomes a research instrument for understanding what your users actually need, run continuously by the users themselves.
Over time you can:
- Add domain tools for common patterns (faster, more reliable)
- Create dedicated prompts for frequent requests (more discoverable)
- Remove tools that aren’t being used (simpler system)
You stop guessing what features to build. You start cataloging what users already do.
Why It Matters
This is the same idea Boris Cherny used to discover Facebook Marketplace — Groups had no commerce feature, but 40% of posts in Groups were buy-sell activity. The behavior already existed; the product was built around it. Agent-native software collapses the loop: instead of waiting for users to hack your product, you watch them hack the agent in real time.
Related
- Latent Demand - How Facebook Marketplace Was Born — the original principle, applied to product behavior rather than agent usage
- Emergent Capability - Agents Do What You Didn’t Design For — the architectural property that makes latent demand visible
- Progressive Disclosure - Simple to Start, Endlessly Powerful — progressive disclosure is what gives users room to surface latent demand
- Brandon the Data Scientist Who Taught Himself — a non-engineer pulling Claude Code into a domain it wasn’t built for is latent demand in the wild
- From Primitives to Domain Tools — the discovered patterns are what justify graduating to domain tools