Column

Who owns this November stress week?

AI only pays off if we dare to let go of the old. But how much should we build on a capability whose terms someone else can set according to their interests?

Black-and-white portrait of Petri Lattu: a bearded man looking upward with a finger at his temple, beaded bracelets on his wrist.

AI should finally start delivering productivity. That requires us to dare to let go of something old. This is everyday bread and butter by now.

But how much do we dare to build on a capability that someone else can define according to their own interests?

The decisive moment of the AI transformation may look very ordinary. Someone in a meeting asks whether the old system can now be shut down.

The room goes quiet.

The new AI-era operating model has been thought through and the tech platform has already been bought. People have been trained, pilots run and successes announced. Still, the old is needed just in case. The new one’s results are checked the old way, and both operating models have their own owners, budgets and steering groups.

The future has arrived.

It’s just billing you.

We may be at exactly this point in the AI transformation. The AI curious phase is starting to turn into the AI serious phase. We know these tools can do amazing things. Now we have to decide how to build operations on them that also have to work in an ordinary rainy November week, when the kids are rushing to their hobbies and you are rushing even more.

What happens to the old way of working then?

It was recently reported that Accenture capped centrally funded Claude use at 29 dollars per employee per month. This is not an absolute ceiling on all AI use: heavier use has to be funded by a project or team. But the theme is interesting.

After encouraging people to use AI, the question becomes who pays and what the money produces.

Fair enough.

Then again, looking at a 29-dollar bill does not yet tell you whether the use is expensive. A thousand dollars a month can be an excellent investment if it creates value. Even cheap use can be pointless if it mostly produces more stuff for others to check.

So we need to rethink the cost of the whole piece of work. Including the time of the people who figure out what the AI actually did. Not exactly easy, and even harder given how fast the AI world moves.

The problem of parallel worlds

At this question, the problem of parallel worlds starts to bare its teeth.

If the new operating model is layered on top of the old one, a company can get more output without its ability to perform improving accordingly. One person works faster, another checks more and a third reconciles the two. Everyone is busy, and the deck about the productivity leap gets finished in record time.

Limited parallel use is often sensible. It builds evidence before the switch. But a transition needs conditions for when it is over. Otherwise the temporary becomes a new permanent layer of the organisation. If AI speeds up organisations and we live in a constant change frenzy, who is thinking about where the money comes from, now and tomorrow?

How many “transformation meetings” do we have, compared with “here’s what we could do” meetings?

That is why trust is the economic precondition of the AI transformation.

We need to be able to show where the new way works well enough. How are errors spotted? Who is responsible for fixing them? What does the work cost, with all its steps? When does a human check improve the result, and when does it just repeat what has already been done?

These need answers at scale, not just from the point of view of one unit, before the old can be let go. Trust also comes from being able to define where the system is not yet trusted.

A company that solves this before others can gain a significant head start. It can transform its operations and its market while its competitor maintains two worlds.

When the competitor succeeds too

But let’s take this thought a little further.

What happens when the competitor succeeds too? And then the next one?

Let’s assume the productive efficiency achievable with AI starts to level off at some point. Model development may keep going at a breakneck pace, but the same improvements reach everyone else so quickly that they do not create a lasting lead.

Using AI becomes the price of entry.

Everyone has a capable system. Everyone produces fast. Everyone’s costs go down. Everyone’s strategy promises to be customer-centric, and AI has helped make that case very convincingly.

How do you win then?

Part of the competition shifts to what a company knows how to aim for. Who understands the customer’s situation? Who is trusted? Who has access to the customer? Who dares to leave undone what everyone else can now do almost for free?

In my previous column I looked at defining problems and solutions when there are more ideas and outputs on offer than we have time to process. Now the same question grows to the size of a company. Abundance makes choosing more important than ever.

But the story also has a less romantic possibility.

The benefit may end up with those who control customer relationships, distribution and the infrastructure everyone uses. More efficient companies compete with each other on price while the capability provider bills every one of them.

The means of production of the new era are available to everyone. For a monthly fee. Within usage limits. On the terms in force at the time.

So the differences in capability between companies may narrow while the gap between those who own capability and those who use it grows. A lot depends on who decides on capacity availability and the daily price, and on what the ethical cost of that capacity is.

When someone else sets the terms

At this point, the original question about trust soon looks quite different.

We can verify that the system works well today. But do we dare to build so much on it that we can no longer manage without it?

Ukraine’s dependence on satellite connections shows this question in an exceptionally harsh environment. Ukraine recently negotiated the removal of a speed limit on the military Starshield network. Conflicting claims were made about who was responsible for the limits, involving Elon Musk, and Trump denied that his administration had interfered with the system. Rumours are flying in every direction.

Alternatives are being built too. The EU’s IRIS² and the Canadian-backed Telesat Lightspeed are projects launched earlier to strengthen independent satellite capacity. Finland is in the game as well: ICEYE and Nokia recently announced a joint project that competes directly with Starlink. Their significance is not hard to see in this situation.

At the same time, the US ambassador to the EU, Andrew Puzder, has warned of countermeasures if the EU restricts US satellite companies’ access to some frequencies. The issue is a proposal to reserve frequencies for European operators that would let phones connect directly to satellites.

This will no doubt be debated a lot more, and new information will probably come to light.

The basic problem of dependence is clear all the same. The terms of a capability essential to your own operations may be for someone else to decide.

A company’s situation is of course different from that of a state at war. But at worst the mechanism is the same. Once the old ways of working have been dismantled, the new supplier’s interest in you may extend well beyond the billing model.

And then we get to ourselves

My favourite topic: what thinking with AI is.

Can we already ask the next question, as AI becomes more and more important geopolitically and in trade policy: how do we keep control over the capability we as individuals build in collaboration with it?

Think of an AI companion you have worked with for a long time.

Around it has built up the history of your projects, half-finished ideas, your own concepts and ways of working that have proved effective. You have learned to use it, and the system remembers how best to work with you. Why not live with you, too?

Switching providers can then mean much more than creating a new password. Exporting your conversations to a file will not necessarily carry your ways of working, memory and context over to another environment in a form that lets co-thinking continue seamlessly.

Part of your own agency has been built into a service you do not control yourself. This may give rise to one of the next political turns in the AI debate.

What if part of the opposition to AI turns into a demand to bring AI under our own control?

This may be an odd thought. AI is opposed for many reasons, and not all opposition can be explained by one theme. But if people feel they are losing influence over their own work and lives, an even better AI demo is unlikely to fix that.

The ability to influence, and ownership, might.

Then the demand could be surprisingly simple: I want to use this capability for my own purposes and keep what I have built with it.

Could the right to a personal AI companion start to look like a civil right?

This does not yet mean a right to unlimited compute or to automating any activity whatsoever. In practice, it could be the right to move your own history and memory, switch providers and keep reasonable continuity when a service changes or shuts down.

And the right to an alternative that is actually usable. A local model is of little help to someone who cannot afford it or does not have the skills to run it.

European and national alternatives can be part of the answer. But a server’s location alone does not tell you how much control an individual has. National independence and the independence of the citizen are questions of different scale, and both need answers.

This is where trust, competitiveness and ownership come together.

One lever for us is probably understanding our own market position, and understanding that international technology companies cannot easily do without, say, European customers.

Another thing I will come back to in the future: what if our AI superpower turns out to be using it best, not just building regional alternatives?

So, to finish

A company needs enough trust to transform its operations. A person needs it to build new ways of thinking and doing. For both, trust is strengthened by knowing that the terms of the collaboration can be assessed, the supplier can be changed and the capability that has been built can be kept.

So we should build the ability to leave on our own terms. Terms with teeth.

That may sound like a contradictory goal in the middle of a massive adoption race. But the easier it is to leave a relationship if needed, the more reliably you can build on it.

So, in the middle of November’s stress week, before the old system is switched off, one thing should be settled.

Who gets to switch off the new one?


The writer believes they stay upbeat in autumn too, and at exactly that time of year are more excited than worried – often to others’ annoyance. That said, on the day this column was finished they are also taking their son to robotics school, where Linus is learning to build, and hopefully also to keep in charge of, tomorrow’s new housemates.

Petri Lattu

Innovation director and AI strategist who writes about AI's impact on society and about cognitive sovereignty: how people can think with AI without handing their thinking over to it.

More from this writer →

Read next