The Experiment Is Still the Unit of Progress
Ries argued that output — features shipped, lines written, screens designed — is vanity unless it teaches you something. That logic holds when the builder and the editor are the same person directing agents in natural language. Collapsing prototype time to hours doesn't change what you owe yourself: a clear hypothesis before you build, a real signal after. The bottleneck moved from typing to thinking, so the thinking has to be sharper, not skipped.
MVPs Are Cheaper to Run, Harder to Read
An MVP was always about minimum effort for maximum learning, not minimum quality for maximum speed. When software becomes language-directed, the minimum gets very small, very fast. The risk flips: founders now face a surplus of testable surfaces and a shortage of careful reading. Deciding what a result actually means — whether users are engaging or just tolerating — is judgment work that no agent performs on your behalf.
Pivot or Persevere Requires a Human in the Chair
Ries framed pivot-or-persevere as a structured meeting, not a gut call — evidence reviewed against the original hypothesis, then a decision owned by someone accountable. In an AI-native company, agents can instrument behavior, surface anomalies, and summarize patterns. Setting the goals those metrics serve, making the tradeoffs when signals conflict, and owning the call — that remains irreducibly human. Autonomous systems still need humans to define what winning looks like.
