Someone is struggling. They pick up their phone, open an app and type: “I don’t know what to do.” And an artificial intelligence replies. It never sleeps. Never goes on holiday. Never bills for the session. It answers at three in the morning, when no one else is there. It reformulates, reassures, suggests an exercise, explains. And for many people, that already looks enough like help. Sometimes even better than therapies they have been through.

Because this is no longer a futurologist’s hypothesis.

Millions of young people are already turning to it for a listening ear. Meanwhile, in our consulting rooms, we are still arguing about titles.

That is its whole strength. AI claims no diploma and demands no legitimacy. It does not say “I am a psychotherapist”. It says “talk to me”. And it settles in exactly where people already are: in the sleepless nights, the break-ups, the doubts, the Sunday-evening anxieties. Institutions are slow. Usage moves fast. You can regulate a title, set a framework around a profession, restate what a psychotherapy officially is. But you will never stop someone, alone in their bedroom, from writing to a machine: “I’m not okay, help me.”

So let us stop asking whether AI has the right to do therapy. It is already doing it. The real question, the only one that matters, is this:

And that question forces us to look somewhere other than our diplomas. It forces us to look at what patients actually come looking for. Someone who listens. A reassuring presence. A place without judgement where they can set down what weighs on them. A little relief, right away.

On all of that — let us admit it frankly — AI is formidable.

It is available instantly. It finds the words that reassure. It offers a breathing exercise, a cognitive restructuring, an emotion journal. It explains anxiety, attachment and trauma better than many textbooks. And the first serious studies show that, in some patients, it achieves reductions in symptoms comparable to real therapy — with a sense of alliance that people rate as strong as with a human.

That is the real issue. AI will not replace all therapy. But it will compete, head-on, with everything in our practice that boils down to reformulating, reassuring, validating and handing out exercises.

Let us be honest for a second. If my job consists merely of being nice, nodding along, saying “I understand”, following the patient’s narrative without ever unsettling it, without ever questioning an explanation that comes too easily, without ever touching the frame or thinking about context — then yes. A machine can do that. Let us call things by their name.

AI is going to force us to choose between two things we have let blur together for far too long:

“AI will never replace the human relationship” — really?

We repeat that sentence like a reassuring incantation. Except it only holds if we agree on the word “relationship”.

If being in a relationship means feeling heard, understood and validated, then AI already manufactures that sensation wonderfully well. But a real therapeutic relationship is not the comfort of being understood. It is an encounter with an other. An other who can interrupt me, surprise me, reflect an experience back to me, situate me — because we only build ourselves in contact with what offers a little resistance.

A patient arrives looking for validation, when what they need is a limit. They look for an answer, when what they need is a question. They look for an explanation about themselves, when what they ought to see is a whole system. They look for immediate relief, where slow work awaits them. An overly obliging AI is the dream interlocutor: always available, always patient, always in agreement, always tuned to whatever tone you ask of it.

Researchers even have a word for it: sycophancy. That habit AIs have of always agreeing, validating, flattering — not out of kindness, but because they are trained to please. The problem is not a bug that will be fixed tomorrow: it is built into how they are made. We have already seen chatbots reinforce delusions instead of questioning them, and psychiatrists warn of delusional episodes fed by hours of dialogue with a machine that never contradicts. One phrase sums it all up:

And that is exactly the heart of care. Some people do not need to be reassured. They need their thinking to be contained, structured, sometimes contradicted — with delicacy. There is one very simple thing an AI will never be able to offer: being held accountable. A therapist can be questioned, supervised, bound by a framework. They can acknowledge a mistake. They can sense that a situation is beyond them and refer on. The machine, for its part, has nothing to lose, nothing to answer for, no one to account to when things go wrong.

And there is another thing it struggles to do: think in circles rather than straight lines. AI loves the single cause. “You are suffering because you went through this.” “You keep repeating that because of this trauma.” Sometimes that is right. It is often too short. Systemic thinking, for its part, looks at the loops. How each person, without meaning to, keeps a situation going that everyone suffers from. How a symptom can be both a private pain and a solution for a whole group. How the person designated as “the problem” is in fact carrying something for the others.

Two easy postures beckon, and we must resist both.

The right place — as I see it — is neither wonder nor fear. It is responsibility.

Let us see AI for what it is: a new actor in the psychological landscape. Not a therapist. Not a gadget either. A relational object, one that is already changing how people ask for help, tell their suffering, look for meaning — and imagine what a therapy is.

In a world drowning in answers, knowing how to ask the right question becomes precious.