Discover Before You Deploy
Blank's customer discovery phase is the practice of leaving the building to test whether your problem hypothesis is real. For AI-native founders, the temptation is to build a workflow first and find customers second, because the workflow is cheap and fast to assemble. Resist this. An agent that solves a problem no customer actually has will evaluate beautifully in your harness and fail quietly in the market. Build the eval around a real customer pain or the eval is measuring the wrong thing entirely.
Validation Is a Signal Gate, Not a Formality
Customer validation, in Blank's framing, means proving that a repeatable sales process exists before you invest in scaling it. The equivalent gate for an AI-native company is proving that your agent's outputs reliably earn user trust under real conditions. Evals run on synthetic cases tell you about coverage; they don't tell you whether a human in the loop will accept the output and act on it. That acceptance is the validation signal, and you have to watch for it deliberately.
Creation Phase Means Scaling What You Proved
Blank's customer creation step is where you spend money to generate demand, but only after validation. Founders shipping with agents face the same discipline: the moment your workflow is validated and the eval harness confirms consistent quality, you can graduate from a fixed path toward genuine agency where the branching is real. Scale what you proved. Do not expand the agent's autonomy and the customer base simultaneously, because you will not be able to isolate which variable is causing problems.
Company Building Requires Human Judgment at the Top
Blank's fourth step, building the company, is where process replaces heroics. In an autonomous company that process includes deciding which decisions agents own and which decisions humans own. Agents can execute at scale; they cannot set the goals, make the ethical tradeoffs, or absorb accountability when something goes wrong. The organizational design question Blank raised in 1995 is the same one you are answering now: who is responsible for learning, and how does that learning change what the company builds next?
