We start by handing a piece of writing to an artificial intelligence. Then a folder. Then we give it access to our files, our tools, our ongoing projects.
It makes sense. Explaining the context again in every conversation soon gets tiresome. An assistant that already knows the work lets us get straight to what interests us.
So we entrust it with more.
At some point, hard to pin down, it stops being a tool we consult. It becomes the place we go through to use the others.
The difference starts to matter.
In my previous pieces, I have mostly been interested in what these technologies allow us to do, and the difficulties we face in making them our own. But that process works both ways. We learn to use a service. We also gradually organise our work around it.
An article by Tariq Krim on OpenAI and Anthropic raises this very question through their business models. In its introduction, he distinguishes a strategy focused on attention and consumer uses from another centred on the work of businesses. [1]
There is a concrete basis for the distinction: OpenAI is developing advertising in ChatGPT, while Anthropic says it intends to keep Claude free of ads. OpenAI states that ads remain separate from answers and must not influence them. These are commitments to examine over time, beyond how they are worded. [2] [3]
The boundaries are less tidy than a comparison table, though. Back in April 2026, OpenAI announced that businesses accounted for more than 40% of its revenue. Anthropic, meanwhile, is developing partnerships with consulting and services firms, including Accenture. Automating part of their work can also become a shared business opportunity. [4] [5]
The two companies therefore have different orientations without inhabiting entirely separate worlds.
For the user, another question emerges: what place will these services take between us and what we are trying to accomplish?
Imagine an assistant tasked with preparing a business trip. It could compare transport options, suggest a hotel, arrange meetings. The convenience would be real. Yet some of the choices would be made before we even looked at the options.
Which providers would be consulted? Which criteria would be used? Which possibilities would remain out of sight?
Delegating the search also means delegating some of the selection. That can be perfectly reasonable, provided we can understand and correct that selection.
The same issue arises at work. An assistant can save us time because it knows our documents, our rules and our habits. It is precisely this accumulated knowledge that can make replacing it expensive.
We can export files without being able to take a way of working elsewhere.
Sometimes we will have to rebuild connections, check results, spell out instructions that had gone unspoken. Nothing spectacular. A few weeks discovering that the supposedly “frictionless” tool had mostly moved the friction to the exit…
This dependence is not necessarily a mistake. We already depend on software, networks and people. Doing everything ourselves would be expensive, require skills and create other vulnerabilities. Running a local model, for example, does not remove the need for maintenance or verification.
But a dependence deserves to be chosen with an understanding of what it involves. The subscription price tells us nothing about the cost of an interruption, a price rise or a change in how the service works.
A fairly simple experiment can teach us more than a long debate about the best model: take an important task and do it with another tool. See what carries over easily, what is missing and what means starting all over again.
This small exercise also helps us see what we have actually learned. Have we understood our work better? Or have we mainly become very good at doing it in one particular environment?
The two can look much the same for a long time.
There is much to gain from these assistants. To keep a sense of proportion, we will sometimes need to take the time to work without the one we have grown used to. Not to prove that we can do without it. To know exactly what we have entrusted to it.