The measurement blind spot: assisted output hides the capability gap (Brown 96->48). "What is your in-person score?" News-pegged.

An economics professor at Brown University recently ran, almost by accident, one of the most useful experiments of the AI era.

He set his class a take-home midterm exam. The average came back at 96 out of 100, the highest he had ever seen.

Suspicious rather than flattered, he set the next exam in person, in a room, with no tools. The average fell to 48.

The previous work looked too good to be true. Under conditions closer to what the students could do unaided, much of the apparent mastery disappeared, giving the lowest average in the course’s history.

Let us leave aside the question of cheating, which is contested and not really the point here. What the second exam exposed was the gap between assisted output and underlying capability.

And the striking thing is not that a gap existed. It is that he only found it because he had an instrument that could measure it.

Most organisations do not.

You have deployed AI. The deliverables look polished. The decks are sharper. The analysis arrives faster. The emails sound better. The code appears sooner.

But nothing in your workflow separates work your people can produce, defend and adapt from work the tool produced for them. The output can look the same either way, which is precisely the problem.

A team can appear more capable while quietly becoming less so, and you may have no signal until the tool is unavailable, or a client asks a question the model cannot answer, and someone has to think on their feet.

This is not an argument against the tools. It is an argument for knowing what your people can actually do without it. Because a capability you cannot see is one you cannot rely on, plan around, or charge for.

So the question the professor could answer about his class is the one many leaders cannot yet answer about their teams.

Strip away the AI tomorrow, and what is your organisation’s in-person score?