I have worked with data for more than thirty years: as a search-engine developer, in web analytics, in research, and today as a CDTO. In that time, a lot has become cheaper: first access to information, then measuring and computing, and with AI now producing itself. One thing never became cheaper: judgment, knowing which question matters, what is worth doing, and what is actually true. Each of these waves tempts us to mistake an abundance of information for understanding. This page is about that difference.
Understanding Data
We measure more than we used to and often understand less. Once we are convinced of an idea, we mostly notice the data that proves us right. AI systems reinforce this, because they agree so convincingly. The more useful question is usually: what speaks against it?
Whether data is any good depends not on its volume but on how it came about. How was it measured? What wasn’t captured at all? Which assumptions sit inside a metric? In my research I have shown that even the most common web-analytics tool is hardly used in any serious way: the data is there, the handling of it is not.
In the end, what counts is not the dashboard but what you do with it. The hard part isn’t the analysis, it’s the step to a decision. That’s where most data projects fail, not on the technology.
Creating Value with AI
AI is, for now, the most radical step in this development: it lowers not only the cost of measuring, but of producing itself. Suddenly almost anything is possible, and that is exactly the danger. Where effort used to force us to think, AI delivers a convincing result instantly. It has never been so easy to be convincingly wrong at scale.
So I don’t believe AI takes thinking away from us; it shifts it. Less time spent executing, more spent deciding: which problem is even worth it? Which result is the right one? Was the effort worth it? These aren’t technical questions, and no model asks them for me. Whoever doesn’t ask them mostly automates their own mistakes.
That, to me, is the bigger picture: what decides who gets ahead with AI isn’t the next model, but the ability to judge. The same ability that was already scarce with data. As agents begin to divide tasks among themselves, we no longer manage only people, but machines too. The real skill stays human: knowing what is worth doing, and standing behind the result.