The AI middleman in therapy: when a third party appears between you and your therapist
Right now, a prompt is going viral that turns Claude into a therapist. You load CBT articles, describe your situation — and get surprisingly thoughtful questions. Not advice — questions. The kind a good therapist asks by the third or fourth session, once they’ve spotted the pattern.
People don’t want an AI therapist. They want their real therapist to be better prepared for the session.
That thought is the key to everything that happens next.
The problem nobody talks about
Psychotherapy is one of the most format-inefficient fields in medicine. Not in method quality — in logistics.
Think about it: you pay $100–200 for an hour, and a third of that time your therapist spends reconstructing context. What happened during the week. How you felt. What triggered your anxiety. Information you could have shared in advance — if there were a channel.
And between sessions — nothing. Seven days where you’re alone with your thoughts. No tools, no feedback, no one to ask: “How did that work situation we discussed go?”
Not a replacement — a layer
It’s important to be precise here. This is not about replacing the therapist — that’s a dead end, and here’s why:
Therapy is not a set of correct questions. It's a relationship. Transference, countertransference, therapeutic alliance — all built on the fact that a human being sitting across from you chose to dedicate this hour to you specifically. AI can simulate empathy but cannot experience it. And the client can feel the difference.
This is about something else — an agent-intermediary that works between sessions and makes each session dramatically more productive.
week without
support
50 min,
15 on context
▼ with AI layer ▼
daily
check-in
context gathering,
questions, tracking
50 min of
pure work
What the AI agent does
It doesn’t treat. It doesn’t diagnose. It doesn’t interpret. Here’s what it does:
1. Daily check-ins
Morning or evening — a short 3–5 minute conversation. “How was your day?” “Was there anything that caused anxiety?” “Did you manage to try the technique we discussed in session?” Not therapy — more like a journal with feedback.
2. Pre-session brief for the therapist
Before each session, the therapist receives a structured report: key events of the week, mood changes, sleep patterns, triggers, attempts to apply techniques. Not raw text — a digest prepared by the agent.
The therapist starts the session not with "tell me about your week" but with "I see that on Tuesday after the call with your mom there was an anxiety spike — let's explore what happened." 15 minutes of context become 15 minutes of work.
3. Homework and tracking
CBT therapists often assign homework: keep a thought diary, practice exposure, record automatic thoughts. The problem — 60–70% of clients don’t do them. An AI agent can gently remind, help fill out forms, and track progress.
4. Crisis routing
If during a daily check-in the agent detects markers of an acute crisis — it doesn’t try to “help.” It immediately notifies the therapist and offers the client crisis line contacts. This isn’t a replacement for emergency services — it’s an early warning system.
It already (almost) exists
The most telling signal is what’s happening right now. People take Claude, load CBT and ACT textbooks into it, and build a tool that doesn’t exist on the market. This is a classic signal: when users assemble a product from spare parts — it means demand exists, but supply doesn’t.
Uber didn't replace drivers — it became a layer between passenger and driver. Airbnb didn't replace hotels — it became a layer between guest and host. AI in therapy won't replace the therapist — it will become the layer that makes every session twice as valuable.
Ethics and risks
This isn’t a cloudless story. Here’s what could go wrong:
- Privacy. Daily mental health records are the most sensitive data imaginable. Who stores them? Who has access? What if an insurance company gets hold of them?
- False sense of safety. Clients may start treating the AI agent as a replacement for therapy and stop attending sessions.
- Regulatory vacuum. The FDA and its counterparts don’t yet have a clear framework for “AI therapist assistant.”
All of these problems are real. And all of them are solvable — provided the AI agent is clearly positioned as the therapist’s tool, not a standalone service.
By 2029, at least one major online therapy platform (BetterHelp, Talkspace, Headway, or equivalent) will launch an AI intermediary agent that works between sessions: conducting daily check-ins with the client, gathering context, and preparing a structured brief for the therapist before each session.
The product will be positioned not as an "AI therapist" but as "your therapist's assistant" — a tool that makes live sessions more productive. Therapists using such an agent will be able to serve 30–50% more clients at the same quality of care.