By 2028, AI Will Take Over a Psychologist's Routine Work but Leave the Decisions to Them

It is a morning in 2028, twenty minutes before the first session. The client has filled out a mood diary. The system has collected the week's entries, flagged recurring themes and prepared a draft summary. The psychologist reads it, strikes two overly bold conclusions and changes the session plan. An hour from now, AI will remind the client about the exercise they chose together. Habr paints this picture of a working day. The author immediately adds a caveat: it is a scenario, not a forecast with a guarantee. But the direction of change it points to is fairly clear.

What the machine does in this day

In the scenario, AI does not conduct therapy. It performs the operations that currently eat up a specialist's time between sessions and before them:

• collecting and structuring the client's mood diary for the week;
• finding recurring themes and phrasings in the entries;
• a draft summary that is ready about 20 minutes before the session;
• a reminder to the client, an hour after the session, about the exercise chosen jointly with the psychologist.

All of this is work with text and schedules, which makes it the most direct use case for language models. It calls for memory, summarization and a careful tone. It does not require diagnostic judgment.

Where the line falls

The episode contains a detail that matters more than all the automation: the psychologist strikes two overly bold conclusions. A model given a week of entries tends to build coherent, confident explanations where the data supports only a cautious hypothesis. The specialist works as an editor: they cut what is excessive and keep what their knowledge of the client confirms.

This arrangement has a weak spot. A summary read twenty minutes before a session sets the frame. If the psychologist gets used to accepting the system's wording, bold conclusions will start migrating into the session plan unedited. The decision formally remains with the human, but in practice the model prepared it. The profession will have to develop a discipline for working with AI drafts, much as developers are already doing with code autocompletion.

An uncomfortable question about responsibility

The author names one more shift outright: automation brings the question of responsibility with it. Suppose the diary contained an alarming entry and the system left it out of the summary. Or, conversely, it raised a false alarm. Who is responsible: the psychologist who read the summary, the platform, or the model developer?

Regulators are already responding to adjacent cases. In 2025, the state of Illinois restricted the use of AI as a standalone psychotherapist. The Habr scenario sits on the other side of that line: AI works as an assistant under the supervision of a licensed specialist. But no one has yet established a clear allocation of responsibility for such an arrangement, either in law or in client contracts.

Data is a separate issue. A mood diary is among the most sensitive information that can be collected about a person. Where the entries are stored, which model processes them and whether they are used for fine-tuning determine whether such a tool can be adopted at all.

What a practicing specialist should do

There is still time before 2028 to prepare, without waiting for finished products.

• Start with tasks that involve no clinical judgment: reminders, structuring notes, exercise templates.
• Check every AI conclusion against the client's primary entries. The rule "strike two bold conclusions" is worth turning into a habit.
• Specify in advance in the client contract which data the system processes, where it is stored and who sees the summary.
• When choosing a service, ask about data storage and about opting out of training on your records, not just about summarization quality.

The profession, in the author's assessment, is not going anywhere. What changes is the content of the day: less mechanical work with records, more editing, verification and legal care.