With AI

§Read Part II again and notice what every method has in common. Design Thinking's "build to think" was radical because building was slow, so thinking with rough artifacts saved time. Lean's minimum viable product was minimal because a full one would cost too much to risk. Agile's increments were small because a big one that turned out wrong wasted months of engineers. The Double Diamond's second diamond was long because exploring solutions meant making them.

§Every one of them was, underneath the philosophy, a budget for a scarce resource. The resource was the ability to make software.

§ 17.1Which assumptions no longer hold#

§The ones about scarcity. In 2025 a person with no technical background could describe a product and have a working version the same afternoon, and the year's fastest-growing companies sold exactly that. When a prototype costs a prompt, the minimum viable product is no longer a way to spend less on building; it has to be redefined as a way to learn more per test, or it becomes an excuse for building anything. When increments are cheap, the discipline is not keeping them small but keeping them attached to a question. When the second diamond can hold twenty prototypes instead of two, the close, the choice, becomes the entire difficulty.

§ 17.2Which hold harder#

§The ones about judgment. Empathy as the starting point holds harder, because a team that can build anything will build the first thing it imagines unless it has looked at a real person first. The distinction between problem and solution holds harder, because solutions are now so cheap to produce that a team can be deep in the solution space before it has finished a sentence about the problem. Validated learning as the only measure of progress holds harder, because shipping has stopped being evidence of anything.

§And the closes of the Double Diamond hold hardest of all. When the open phases are nearly free, the whole value of discovery concentrates in the convergent ones: the problem definition that says which of a hundred observations matter, and the choice that says which of twenty prototypes ships. Closing means leaving things out. The AI era made leaving things out the only expensive part of the process.

§ 17.3What this does to the pillars and the models#

§Context adaptability matters more, because the market for most AI products is nascent and the discovery it needs is the slow observational kind, exactly when the tools tempt a team toward the fast building kind. Continuous learning matters more, because the loop turns faster and the picture goes stale sooner. Team-wide discovery is easier, since one person now holds more of the slices, and the connection is still the work.

§First principles matter more, because the best practices of software were written for a cost structure that no longer exists. Systems thinking matters more, because one element of every AI product's system, the model, is improving on someone else's schedule. Managing uncertainty is unchanged in kind and compressed in time. And the conversion of information into insight is the one thing no model does for you, because a model can summarize a hundred transcripts and cannot tell you which sentence changed what you believe.

§ 17.4What did not change#

§A stranger still has a problem you can photograph, and you still have to go and look at it. Twelve conversations still produce one insight, and someone still has to find it. A team still has to close, and closing still means telling people the thing they built is not going in.

§Product Sprint, the next whitepaper, is what that looks like as a practice, with the building step nearly free and everything around it exactly as hard as it was.