Discovery Before Deployment
Blank's customer development model runs in parallel with product development — you build and learn simultaneously rather than sequencing one after the other. For a founder shipping with agents today, the temptation is to let the system run, gather data, and infer what customers want from usage patterns. That skips the discovery phase entirely. An eval harness can tell you whether your agent completed a task correctly, but it cannot tell you whether the task was worth doing. That question requires a founder in a room — or on a call — watching a real person struggle with a real problem.
Validation Is a Gate, Not a Milestone
Blank separates discovery from validation deliberately: you first find a problem worth solving, then confirm enough people share it to justify building. Translated forward, this means your agentic workflows should not be designed for scale until the underlying customer hypothesis has been validated by humans. Building a sophisticated orchestration layer on top of an unvalidated assumption is expensive in a new way — agents can execute wrong things at speed and volume that manual processes never could. Validate the judgment call before you automate it.
Creation Requires a Repeatable Signal
The third phase in Blank's framework is customer creation — moving from early adopters to a broader market — and it only works if you have a repeatable story about why customers buy. In an AI-native company, that story includes trust. Customers need to understand what the agents are doing on their behalf and feel that a human is accountable for the outcome. A founder who cannot explain the system's behavior to a skeptical early customer has not yet earned the right to scale it, no matter how clean the eval scores look.
Earn the Right, Then Scale the System
Blank's final phase, company building, is about converting a validated, repeatable process into an organization. The AI-native equivalent is designing the full stack — agents, workflows, evals, and human checkpoints — only after the founder has done the customer work personally. The eval harness measures task reliability; it does not measure whether you found the right task. That prior judgment belongs to the founder, made through the unglamorous, unautomated work Blank described decades before anyone used the word agent.
