AI Madness and Illusion of Progress
5 March 2026
Over the past months I have been watching the same pattern repeat itself across teams and companies. Everyone is building AI prototypes. New demos every week, new tools, new features that promise to accelerate everything. And in many ways they do. It has never been easier to generate code, assemble interfaces, and produce working software in hours instead of weeks.
But something keeps bothering me. The thinking behind the problems has not improved nearly as much as the tools. Requirements are still vague. The understanding of user problems often remains shallow. Teams still jump quickly into solutions before fully understanding what they are solving. The difference now is that AI makes it possible to move faster without fixing any of that. So we produce more prototypes, more features, and more software, but not necessarily more value. In several conversations recently I have seen teams proudly demonstrate something they built in a few days, yet when you ask what real problem it solves, the answer becomes surprisingly unclear.
The implementation barrier has dropped dramatically, but the cognitive work that should happen before building has not caught up. If anything, the ease of generating output might make it easier to skip the hard thinking entirely. Instead of improving how we define problems, we are mostly accelerating the production of solutions.
I am writing this as a snapshot of how things look to me on 5 March 2026. I will revisit this in a few months to see if there are real signals of improvement. The real breakthrough with AI will not be when it helps us produce more code or faster prototypes. It will be when it helps us think more clearly about the problems that are actually worth solving.
5 March 2026
Over the past months I have been watching the same pattern repeat itself across teams and companies. Everyone is building AI prototypes. New demos every week, new tools, new features that promise to accelerate everything. And in many ways they do. It has never been easier to generate code, assemble interfaces, and produce working software in hours instead of weeks.
But something keeps bothering me. The thinking behind the problems has not improved nearly as much as the tools. Requirements are still vague. The understanding of user problems often remains shallow. Teams still jump quickly into solutions before fully understanding what they are solving. The difference now is that AI makes it possible to move faster without fixing any of that. So we produce more prototypes, more features, and more software, but not necessarily more value. In several conversations recently I have seen teams proudly demonstrate something they built in a few days, yet when you ask what real problem it solves, the answer becomes surprisingly unclear.
The implementation barrier has dropped dramatically, but the cognitive work that should happen before building has not caught up. If anything, the ease of generating output might make it easier to skip the hard thinking entirely. Instead of improving how we define problems, we are mostly accelerating the production of solutions.
I am writing this as a snapshot of how things look to me on 5 March 2026. I will revisit this in a few months to see if there are real signals of improvement. The real breakthrough with AI will not be when it helps us produce more code or faster prototypes. It will be when it helps us think more clearly about the problems that are actually worth solving.