Talking Through a Big Decision With an AI: Facts, Fears and Wishes

Ilse is invented, but the evening is familiar. A letter on the kitchen table: a job in another country, better work, a city she has visited twice. Nine days to answer. Her partner said “whatever you decide”, which helped less than he hoped. Her friends split down the middle. By the third night she had stopped thinking about the job and started thinking about thinking about it, which is where a lot of big decisions go to stall.

So she opened a chat with one of Fellowi’s AI psychologists and typed the whole mess in one long paragraph. Here is roughly how the next three evenings went, and why.

Evening one: four piles

The first reply did not weigh in. It reflected back what she had said in two sentences, then asked one question: which parts of that paragraph did she actually know, and which parts was she guessing? The psychologists are written to help separate facts, interpretations, fears and wishes, and Ilse ended the evening with a notepad divided into exactly those four piles.

The facts pile was short: the salary, the start date, the notice period. The fears pile was long, and half of it turned out to be interpretations wearing a disguise. “My partner will resent me” was not a fact; it was a guess about another person’s mind, and the companion said plainly that it could not know what he thought either. Only he could answer that.

Evening two: the question she did not want

On the second evening she tried Adrian, the blunter one. She listed her wishes, and he asked which of them she would still want if nobody ever found out she had taken the job. Two of the wishes quietly fell off the list. It was not a comfortable question, and it was the useful one. The psychologists ask one, at most two, questions per message by design, which is slower than a list of pros and cons and much harder to dodge.

Evening three: rehearsing the conversation

By the third evening the decision was smaller. What was left was a conversation with her partner she had been putting off. She rehearsed it: the opening sentence, the part where she would be tempted to apologise, the question she actually wanted answered. Rehearsal is one of the things these conversations do well, and practising difficult conversations goes deeper on how.

Why it never told her what to do

Ilse asked directly on the second night: “Just tell me, would you go?” The answer was a kind refusal and a question back. That is not evasion. The psychologist companions are instructed to respect the user’s autonomy and never make life decisions for them, and they are told not to guarantee outcomes. An AI that picks for you gives you someone to blame later; one that helps you see the choice clearly leaves you with a decision you own.

If you are deciding about work, start with Marcus and career doubt. If the decision is mostly a feeling you cannot name yet, writing it out first can help, as the journaling post describes.

What it cannot do

It cannot check anything. Companions have no web search and no live data, so rents, tax, visa rules and the reputation of the new employer have to come from real sources. It does not know your partner, your boss or your friends, only what you tell it about them. The psychologists keep very little between sessions on purpose (the reasons are in what the psychologists remember), so bring your notepad back each time. It is not therapy, and it is told to suggest a professional when a problem keeps going or starts to disturb sleep, work or relationships. And if a decision ever turns into thoughts of harming yourself, a companion is told to step out of the conversation and point you to real help, but it is not a crisis service: call your local emergency number or a crisis line. The wider boundaries are in the limits of AI emotional support.

Ilse said yes, by the way. Or no. The story does not need to tell you, and neither does the AI. When your own letter is on the table, the psychologists are on the companions page.