A piece of electronics, almost working, one part missing. The repair cost more than a new one. That calculus reshaped how we treat physical objects, and it is now arriving for software.
During a live security incident, the AI you depend on can decline to help, because the work of a defender looks a great deal like the work of an attacker. That is an argument for owning a model you control, not for abandoning safety.
One of us has built close to twenty applications in the past year. Three are still in use. That ratio looks like waste, and the interesting question is whether it is. A case that the current building spree is a honeymoon phase, and what it is buying.
A flat $30 seat that includes frontier-model usage is a promotional price, whoever is paying the difference. Companies are adopting Copilot because it is cheap and easy to govern. Both of those are real advantages. Neither answers the question that arrives at renewal.
The fastest-growing open-source project in GitHub history vanished from the conversation within about a month. It did not vanish from the internet. What the big labs absorbed was not the software, and the difference matters if you are choosing tools.
The clearest finding from a year of AI productivity research is a gap: people feel dramatically faster and measure barely faster at all. What that feeling is made of, and how to tell which side of the gap you are on.
