The empty column

An opinion piece in the Swedish business daily Dagens industri last week contained a sentence I have not been able to let go of. When leadership calculates automating away its marketing department, one column is exact. The other, what you give up, is left empty, because nobody can put a number on it in advance.

The piece was about Nike. The company scrapped its brand communication, cut out the retailers and moved everything to its own web shop. Every decision could be defended in a spreadsheet. In early October Nike left the S&P 100 after eighteen years in the index, with around 230 billion dollars of market value wiped out. Now the company is spending around five billion dollars to buy back what it once optimised away.

For Nike, the empty column could eventually be filled in. It is the most expensive way to find out what it held.

Same calculation, different room

The piece is about marketing departments. I hear the same calculation in another room.

Replace the development team with a couple of people who are sharp with agents. The savings can be worked out in an afternoon: this many positions, this much in consulting fees, this much already this year. The column next to it is empty. And it is empty for the same reason as at Nike. What you give up does not show until it is missing.

I think I know what it holds. And it is not what most people think.

It is not the doing

The usual objection is that the agents cannot do the job. That the code gets worse, that the model hallucinates, that it lacks context. That is true today and it gets less true with every new model. Whoever builds their resistance on it has an argument with a best-before date.

I do not build mine there. The doing can be automated and I do it myself. AI is part of everything I deliver: the code, the data collection, the reports. Building got cheap. Building is not what the empty column holds.

It is the judgment

What it holds is judgment. And judgment is three things that look like one.

The first is knowing what matters. One of the most senior people I have worked with can spend a whole day thinking and then check in four lines that go straight into production. Three hundred lines from someone else can be noise. In a spreadsheet that counts positions, the two weigh the same.

The second is being able to say no. To look at a conclusion that seems reasonable and know that it is wrong, because you have been there before. A model gives the statistically most likely answer. Someone who has been wrong and paid for it sees when the likely answer does not hold in this particular case.

The third is being able to be held accountable. When something breaks in production on a Tuesday night, someone has to understand why it was built that way and stand behind it. That is not a capability a model can grow into. A computer can never be held accountable. And that argument does not age with the next version.

All three live in people. None of them shows up in a budget.

I am the calculation

I want to be honest here, because I stand on both sides of this calculation. I deliver with a fraction of the team and the time it used to take. On paper I am exactly that pair who are sharp with agents.

But it works for a reason that is not in the spreadsheet. The judgment came along. Thirty years of building systems, of being wrong and paying for it, is what lets me have AI write and still know what should be delivered. AI amplifies what you have. It does not create what you lack.

Remove the team and keep the judgment, and the calculation can hold. Remove the judgment along with the team, and you save money this year and take on a debt at an unknown rate.

It cannot be bought back

Nike can buy back its retailers and its stories. It is expensive, but they are for sale. The retailers were still there.

The judgment in a team is not still there, waiting. It lives in people who have moved on. The knowledge of why the system looks the way it does, which shortcuts were tried and why the decision in March went the way it did, leaves with them. It is not in the repository. It went home.

That is why the empty column is more dangerous here than at Nike. There, it could be filled in afterwards with a number. In a development team, it gets filled in with a system nobody understands anymore.

What actually follows

This is not an argument against efficiency. It is an argument for knowing what you are making efficient.

Automate the doing without hesitation. That is where the gain is, and it is real. But before the team shrinks, someone needs to be able to say who carries the judgment in it and whether it comes along. If it does not, the calculation has an empty column. It always gets filled in eventually, just not by the person who did the counting.

The savings show in the spreadsheet. The judgment shows only when it is missing.

Source: "Gör inte om Nikes räknefel" ("Don't repeat Nike's miscalculation"), opinion piece in Dagens industri, 9 October 2026.

See also: Give me the insight, not the decision (series 34), A Computer Can Never Be Held Accountable (series 49) and Building got cheap. The bottleneck moved. (series 38).