The MVP Is Still an Experiment, Not a Demo
Ries defines the minimum viable product not as the smallest shippable thing but as the fastest way to test a specific belief about what customers want. With Software 3.0, prototyping collapses to hours, which means founders can run more experiments per week than was previously possible in a month. The risk is treating that throughput as progress. Shipping faster only compounds learning if you defined the question before you built the answer. Speed without a hypothesis is just noise at higher velocity.
Build–Measure–Learn When the Builder Is an Agent
The loop Ries describes — build, measure, learn — assumed humans were the bottleneck at the build stage. Agents remove that bottleneck, so the scarce skill shifts entirely to the measure and learn steps. Knowing what to ask the agent to build requires taste; recognizing whether the result actually answers your question requires judgment. Neither of those lives in the model. The founder who directs well and reads the output honestly is doing the hardest work in the loop.
Pivot or Persevere as a Judgment Call You Cannot Delegate
Ries frames pivot-or-persevere as a disciplined decision made on evidence, not frustration. In an agent-driven company, that decision belongs entirely to the human. Agents can surface usage data, generate variants, and run tests. They cannot weigh whether a pivot abandons a real insight or rescues the team from sunk-cost thinking. Setting the criteria for that call in advance — before the data comes back — is one of the clearest places where human judgment designs trust into the process rather than patching it in after a mistake.
