Output was never the point, and now it's worthless
Cagan's central complaint about feature teams was that they measured themselves by what they shipped rather than what changed for the customer. Agents make this failure mode cheap to industrialize: a founder can now produce roadmap output at any volume they like. That means outcome is the only measure left with any information in it. The discipline transfers directly — define the result you want before you dispatch the work, and judge the work by whether the result arrived. An eval built before the agent runs is exactly this: the outcome stated precisely enough that a machine can be checked against it.
The four risks, redistributed
Inspired names four risks every product must retire: is it valuable, is it usable, is it feasible, is it viable. Agents have collapsed the cost of the feasibility question — a working prototype is hours away, not quarters. But value and viability are judgment calls about people and tradeoffs, and those stay with the founder. In an autonomous company, humans set the goals, weigh the tradeoffs, and own the accountability for what ships. The temptation is to let cheap feasibility answer the other three questions by default. It can't. Building it proves it can be built, nothing more.
Continuous discovery becomes continuous evaluation
Cagan insisted discovery run alongside delivery, always, not as a phase that ends. For a founder shipping with agents, the working translation is the feedback loop itself: tests, red cases, and structured results feeding back into how tasks are defined and tools are scoped. The failures you collect are your discovery artifacts — each red case is a small, cheap experiment revealing where the product or the agent's task definition is wrong. A team that only reviews green runs has quietly stopped doing discovery and gone back to admiring output.
Empowered teams, translated to empowered systems
Cagan's empowered team is handed a problem and trusted with the solution — but that trust is engineered, not wished for: clear missions, real context, accountability for results. This is the same design problem as reliable agentic work. Vague autonomy fails for agents the way vague empowerment fails for teams. What works is a clear task, tight tools, curated context, explicit permissions, and a loop that measures results. The founder's role is the one Cagan gave good product leaders: hold the problem, define the boundaries, and let the system find the path — then check the outcome, every time.
