On 26 August the European Central Bank published its figures on the use of artificial intelligence at work. They come from the Consumer Expectations Survey, which every month interviews around twenty thousand people across eleven euro area countries, and they show that in two years the share of workers using it has gone from 26 to 52 per cent. Those who use it keep it open about three days a week.

So far this is the headline everyone ran: half of Europe’s offices now work with artificial intelligence.

Then there are the other two lines in the same document. The efficiency gain for the euro area economy as a whole is 3.8 per cent. The productivity growth the ECB estimates is 0.35 percentage points a year.

Half of all workers use a technology that everyone agrees is changing the world, and the economy moves by three tenths of a point. The reason has more to do with firms than with models.

[GRAPH 1: `Grafica/grafici/uscita-05_dalle-ore-ai-decimi_EN.png`]

The three hours that become forty minutes. The median user reports saving three hours a week, or 7.7 per cent of their working time. That sounds like a lot. But 7.7 applies to those who use it and genuinely gain from it, and they are fewer than they seem: 48.8 per cent of workers say they use it and save time. Fewer than half. Spread that saving across everyone who works and 7.7 turns into 3.8.

And the ECB adds a condition none of the Italian write-ups carried, which is the most important of all: those hours become productivity only if someone fills them with work. If the freed time goes into meetings, internal queues or waiting, it enters nobody’s accounts. It does not depend on the person using the tool. It depends on whoever organises the work.

Where the gain is real. The document ranks tasks, and the result is instructive. People who use artificial intelligence to write or debug code save almost eight hours a week, a full working day. Eight per cent of workers do this. Data analysis and the automation of routine tasks are in the same range.

The most common uses are different: looking things up, writing text, editing what has been written. They are also the ones that save the least.

[GRAPH 2: `Grafica/grafici/uscita-05_ore-per-compito_EN.png`]

Anyone who read last Friday’s piece on this blog will recognise the pattern, because it is the same one. The visible part of artificial intelligence, the words, moves fast and counts for little. The part that moves the numbers sits inside processes, where nobody is looking. Now there is a measurement for it too.

Then there is what Altman said, and how he said it matters. On 23 August Sam Altman was a guest on David Senra’s Founders podcast. At one point Senra puts to him something Tobi Lütke, the founder of Shopify, had told him: that 2026 will be remembered as the year every business was up for grabs, and that artificial intelligence would redesign existing software starting with his own. Then he asks Altman whether he agrees.

Altman says he does not. On the timing.

> I thought when we got to GPT-4, which was back in 2023, that very quickly after that there was > going to be much more disruption, software businesses up for grabs right away, than it turned > out to be. And the thing that I think I was wrong about, in terms of the speed, is the economy > just has so much inertia. People keep doing the same things they’re doing. They keep buying from > the same company, keep sort of wanting to use their tools in the same way.

The Italian press took this for a retreat. But Altman is not apologising, and the part the write-ups cut is the part that says the most:

> I think it’s actually a positive in many ways, and it’s going to make this big transition in > front of us go smoother and slower. I’m grateful for it. But I think it means we’ve all been too > ambitious on timelines, even with this incredible technology.

The man selling the technology says his customers’ inertia suits him. That is a statement against the interest of the person making it, which is why it carries more weight than ten forecasts.

The two sources say the same thing from opposite ends. One measures, the other admits. The ECB counts the hours of twenty thousand people a month and finds three tenths of a point. Altman looks at the market he knows better than anyone and finds the same slowness. Neither claims artificial intelligence is a bubble, and neither says it is eating jobs. Both point to where the bottleneck is, and it is the same one: the time an organisation needs to rebuild the way it works.

How long this slowness will last nobody knows, and the numbers admit it. There is a line in the ECB document that did not make it into the write-ups. Existing estimates of the productivity growth artificial intelligence will bring over ten years range from 0.1 to 3.4 per cent. Between the low and the high there is a factor of thirty-four. Anyone who tells you they know what it will be worth is guessing, and that goes for those promising paradise as much as for those announcing the end of work.

There is one last figure that says more to a business owner than all the others. Of those who do not use it, 41 per cent say they are simply not interested, and among managers the share of the uninterested is 34 per cent. A third of the people who decide how work is done inside a company have no appetite for looking at the tool meant to save those hours.

In the interview Altman explains why he considers Lütke six to eight months ahead of every other chief executive, and the reason has nothing to do with vision: he uses the tools himself, writes software, tries the models, redesigns his own workflows in person. Then he starts a sentence about most company bosses and leaves it there.

The whole point sits between those two numbers. The hours really are freed, and a cheap machine frees them. They turn into output only if someone, inside the firm, decides what to do with them. That someone is not in San Francisco and not in Brussels.