You have been vouching for work you have never seen. the psychology lab is where you see it.

They arrive with a session, and you get it the way supervisors have always got it: they tell you. Twenty minutes of what she said and what they said back, honest and still a story, shaped by what they remember and by what they’d like you to think of them. Somewhere in it is the moment worth the hour, the place where she went quiet and they filled it, and you can’t tell whether it’s in the account or missing from it. So you supervise the account.

At the end of the year you sign a form saying this person can do the work. You’ve talked about their sessions every week, and you have never watched one.

In a single supervision hour, more than four in five trainees keep something back. Over a whole training, almost all of them do. And the mistake is not usually the thing they keep. Most often it is what they think of the supervision itself, and the clinical mistake comes further down the list. None of that is a flaw in them. It is what happens when the only way to bring a session is to describe it, from memory, to the person who also grades you.¹²

So the session that reaches you is the one that went fine. The one they lie awake over, the one you could actually help with, is the one you never hear about.

When supervisees were asked what two weeks of their supervision had actually consisted of, about one in ten had had a recording of a session reviewed with them, and three percent were observed live. The rest was the account, every week. Thirty years of studies have tried to show that supervision improves what happens to clients, and haven’t managed it. Nobody reads that as proof that supervision doesn’t matter. But one reason for it is obvious enough: it has been supervision of what the trainee could remember, by someone who wasn’t in the room.³⁴

Every other expert craft found a way to put the work itself on the table. Ours kept it off, to protect the client. A recording of therapy is the most sensitive object a clinician can hold, so almost nobody held one, and supervision went on running on memory.

psylab is built on one idea: a therapist should be able to look at what happened in the room, and nobody else should, unless the therapist decides otherwise. For supervision that means your supervisee looks first, alone, writes what they saw beside the moment, and then chooses what you see. That order can look like it works against you, but it is what lets the hard session be brought at all.

The session arrives with the moments already marked: their words on why this one, at the exchange it happened in. You open it together and the first twenty minutes come back. You have nothing to read the night before, because the preparation is theirs, and it always should have been. You won’t get through every moment, and you shouldn’t try. An hour holds a handful, properly. Supervision turns back into what it was supposed to be: two people looking at the same ninety seconds of a session and thinking about it together.

The therapists with the best results are the ones who go back over their own sessions and work on specific things. Deliberate practice needs one thing besides the material: someone who knows the work and can say what to try next. That person has always been you; what was missing was the session itself. A commentary from Nissen-Lie at the University of Oslo says as much: traditional supervision misses its chance to practise specific clinical skills. With the exchange in front of you both, the question stops being “what happened?” and becomes “what would you say now?”, and that one can be tried, and tried again.⁵⁶

If your supervisee shares it, you will also see how psylab read the session against the treatment model they trained in: where they clarified, where they reassured, and where an opening went by. It describes, and it stops there. What that moment means for this therapist, with this client, this year, is a conversation between the two of you, and nothing on the screen replaces it.

This is half a minute of a teaching session from 1965, published and shown in training ever since. The therapist asks one question, and the answer is about the client’s daughter. One track is the recording as it was made. The other is what a supervisor would hear if you kept the session for one: the same words at the same pace, in a voice that is not the one in the room, with the daughter’s name beeped. Switch between them while it plays.

The step runs on the way in, on every session, and there is no version of psylab without it. A name becomes language rather than a hole, so the note, the feedback and anything you ask about a session are written from a transcript that still reads like one. Keep a session for a supervisor and the voice is changed first. The original recording is deleted.

The reason almost nobody kept a recording is a good one, so it was the first thing we had to design for. psylab never asks who your client is. It doesn’t ask for a name, a date of birth or a personal number, because nothing in it needs them. A case has a title you choose, and we ask you to make it one that means something to you and nothing to anyone else.

What was said in the room is pseudonymised on its way in. Names, places, dates and the details that would point to a person are replaced, and the clinical meaning stays. It isn’t a setting, and there’s no version of psylab without it. The note, the feedback and anything you ask about a session are written from that text, so nothing after that step sees the original. We say pseudonymised rather than anonymised on purpose: it makes a client much harder to recognise, and we’d rather not claim more than that.

The recording itself isn’t kept. Audio is transcribed and then let go, and only the pseudonymised transcript stays. If you keep the audio so a supervisor can hear a session, the voice is changed first, and the original is deleted. That is what makes it safe to study your own work here, and to let AI help you with it: what’s on the page is the session, without the details that would identify a person.

We want it to be ordinary: they bring Tuesday’s session, you look at it together on Thursday, and the rest of the week goes on as before. Supervision that starts from what happened, not from what either of you can recall. And at the end of the year, a form you can sign having watched the work.

A profession that improves the way surgery and sport learned to, without giving up what makes it therapy: that what is said in the room belongs to the two people who were in it.

psylab is for supervisors who would rather see the session than hear about it, and for training programmes that would rather their supervisors could. It is for the supervisee too, the one with something they haven’t brought. Wondering whether to bring it is part of their job; guessing what it was shouldn’t be part of yours.

psylab is built by a small company in Oslo that hasn’t taken venture money. Raising tends to mean owing someone a rate of growth, and what a company owes shows up in the product: the price creeps up, and your data starts to look like an asset. We want to owe the people who use this.

So far psylab has been paid for out of our own pockets. From early access on, we want it paid for by the clinicians who use it, at one price for every therapist, held for the people already paying it. A few of them may come to own a small piece of the company. Nothing here depends on hitting a number by the end of a quarter.

We’re doing this because we think the quality of mental health care can go up, and that it starts with therapists being able to look at their own work. We also think AI can be used well here, on the clinician’s side, if the people building it are careful about what it’s for. psylab is our attempt to be careful.