The Signal · Column 053

Nobody Asked For A Focus Group Of Robots

Andrei van Noordt · 10 September 2026 · 2 min read

Steve Jobs famously refused to let focus groups tell him what to build, on the grounds that people do not know what they want until you show it to them. A wave of AI founders now believe they have solved the problem Jobs identified, using models to simulate customer panels and predict how real buyers will react. The pitch is speed. What took six weeks and a recruiting agency now takes an afternoon, and you can run it a hundred times.

I have no argument with the tooling. My argument is with what companies think it replaces. Jobs was not complaining that focus groups were slow or expensive. He was saying that aggregated customer opinion is the wrong input for the decisions that matter most, and that someone in the building has to hold a view the research cannot supply. Making the research instant does not fill that gap. It widens it, because now the volume of plausible-sounding evidence goes up while the number of people qualified to overrule it stays exactly where it was.

You can synthesise a thousand opinions in an afternoon; you still cannot synthesise the one person who knows which opinion to ignore.

This is the pattern I keep seeing repriced across the market. The work of gathering signal is collapsing in cost. The work of deciding which signal deserves to change a roadmap is not, and it never lived in the research function anyway. It lives in a handful of product people, data leads and founders whose taste was earned by shipping things that failed. When simulated evidence is abundant and free, the constraint moves upstream to whoever can read it sceptically. That person becomes more valuable, not less, and the companies hiring for prompt fluency instead of product judgement are staffing the wrong half of the problem.

There is a second-order risk worth naming. A model predicting customer reactions is trained on the past, which means it will reliably tell you that the thing nobody has built yet is a thing nobody wants. Run your roadmap through that filter without a human willing to override it and you will optimise yourself into a very well-validated version of what already exists.

For hiring leaders, the practical read is simple. Before you buy the tool that generates a hundred synthetic customers, make sure you employ the one person senior enough to tell all hundred of them they are wrong, and give that person the authority to do it out loud.