I started my career at the Ministry of Defence as the product manager of an analytics platform for the intelligence community, and as the Czech representative in a NATO working group. For a long time I treated it as an interesting but closed chapter. With the arrival of AI, I realized it may be the most relevant experience I have.
Intelligence analysis is, at its core, a craft that solves a single problem: how to turn a large amount of unreliable information into a good decision. And that is exactly the problem everyone working with AI is solving today.
Lesson 1: No source is a fact
An analyst never works with "the truth". They work with sources, each of which has its own reliability, motivation, and blind spots. Every report is graded twice: how credible is the source, and how credible is the information itself.
The output of a language model is exactly that kind of source. Fluent, confident, sometimes brilliant, sometimes off. Anyone who treats AI as an oracle makes the same mistake as an analyst who trusts a single source. Anyone who treats it as a source whose outputs are verified and calibrated gains enormous leverage.
Lesson 2: Uncertainty is stated out loud
In the intelligence craft, conclusions are formulated with confidence levels: "we assess with high probability", "we cannot rule out". It is not hedging - it is honesty toward the person who has to decide based on the conclusion.
Companies deploying AI today mostly lack this discipline. They either accept AI output blindly, or reject the entire technology after the first mistake. Yet the right answer is the same as in analysis: work with probability, design processes that expect errors, and know when human judgment is required.
AI did not abolish the need for critical thinking. It multiplied it tenfold.
Lesson 3: Signal takes work, noise produces itself
An analyst spends most of their time sorting: what is relevant, what is duplicate, what is deliberately planted disinformation. Noise generates itself; signal costs effort.
Today's AI ecosystem is a perfect analogy. Every week a new model, a new framework, a new "game changer". A company that tries to react to everything does nothing properly. Strategic work with AI starts with defining what counts as signal: which use cases have business value, which technologies are mature enough, and what is just conference noise.
Why I am writing this
Because I believe the AI era rewards a different profile of people than the previous one did. Less those who can build the technology, and more those who can evaluate what deserves attention, decide under uncertainty, and see things through. That is good news for anyone who has spent a career making decisions on incomplete information.
And if you are currently sorting out what in your AI backlog is signal and what is noise, write to me. That sorting is exactly what I enjoy most.
