It’s Starting to Feel Like a Friend — My Conversations with AI

Radio waves. Patagonia. Jackets, a bed, QR codes, and collaboration between AI agents. When I look at my conversations with a chatbot, I see ordinary things to get done alongside questions that took me much further than I had intended to go.

Of course, the title is provocative. “Friend” is a metaphor here for how someone might experience that kind of conversation. A chatbot usually responds politely, patiently, and calmly. You can ask follow-up questions, change your mind, and return to the same subject. I’m convinced that many people will start treating it like a friend: telling it about themselves and asking things they would not ask any other person.

For me, that change starts with something very ordinary. More and more often, instead of typing a phrase into a search engine, I start a conversation. I can improve my own question, add a missing detail, or follow something that caught my attention in the answer. I often do this by voice, especially in the car. I speak as I think: sometimes imprecisely, sometimes trailing off, sometimes changing direction.

One answer leads to the next question

We are talking about radio waves. Satellite communication comes up, so I ask whether it works on the same principle. And at some point I say:

“No, really, it’s fascinating that humans came up with that, isn’t it?”

A moment later, I’m already wondering what other phenomena exist around us even though we do not notice them. What are we actually immersed in? I search for words, changing “magnetic energy” to “electromagnetic energy.” I am only beginning to name what has caught my attention.

It started with technology we use every day. It led to a question about the limits of what we know. How much of the world do I really understand? How many things do I take for granted even though I could not explain how they work?

In a conversation like that, an answer gives me the next place to start. I do not need to know from the outset where it will take me.

Sometimes it takes a conversation to know what I’m looking for

When choosing a bed and a futon, I add that some storage space would be useful, maybe a drawer. I ask about shops. Then I wonder whether a narrower width would be enough. Those successive details add up to a more precise description of what I need.

It is similar with jackets. First I am interested in specific models, then Polish manufacturers, cheaper options, and delivery. In the end, I change my approach: perhaps it would be better to find a few suitable shops instead of constantly looking for one product?

In these conversations, I change both the criteria and the way I search. I can respond to a suggestion and say: this still isn’t it; let’s try another way. That helps me put my own expectations into better words.

It is the same at work. When I ask about collaboration between multiple AI agents, I start broadly: what do I need to learn, how do others do it, are there ready-made solutions? Then I clarify that I probably already understand the basics of coordination. I am interested in how several agents can work in parallel.

That clarification changes the direction of the conversation. Without it, I might get a correct answer to a question that had stopped being my question.

You can say an idea out loud first

Before a project comes into being, you can talk it through. In one conversation, I was wondering whether, before building a product, it would be worth checking how teams use AI when creating software. I was forming that thought as I spoke.

That is where voice helps me. I do not have to write a polished instruction first. I can voice a rough thought, add a condition, and correct the direction. Only then do I decide whether I want to develop the subject.

That is why, in a conversation like this, I prefer one thought, an example, or an apt question. A long lecture can get ahead of the moment when I am still working out what I meant.

You can start with an instruction like this:

I want to talk through [topic], rather than immediately receive a full report.
Keep your answers short, one thought at a time, and leave room for me to ask follow-up questions.
If I am unclear, check that you understand my question correctly.
Do not assume that my premise is true simply because
I included it in the question.

When I ask for a summary, separate:
- what we established;
- what is a hypothesis or an idea;
- what needs to be checked in sources;
- what is worth returning to later.

Why it is easy to treat a chatbot like a friend

I think the sense that you can ask even an awkward question matters a great deal. You do not have to sound smart right away or have your thoughts in order. A calm answer can reduce the fear of being judged and encourage the next question.

That is why I believe that, for many people, these conversations will go far beyond choosing a jacket or explaining how a satellite works. They will concern doubts, shame, relationships, and matters that are difficult to say in front of another person. Someone might put them into words here for the first time.

So I understand why the word “friend” will keep coming back. At the same time, a polite tone alone does not prove a mutual friendship. The fact that I enjoy talking to a tool says above all something about my experience of that conversation.

It is worth knowing what I am looking for at a given moment: information, a counterargument, help putting a thought into words, or simply the chance to follow my curiosity. That determines the kind of answer I need.

A polite answer can still be wrong

In a conversation about Patagonia, I ask why people do not build there. That question already contains an assumption. Before I received a long explanation, it would be worth checking whether it is accurate and what exactly I mean.

A fluent answer can sound like confirmation that I framed the question well. Yet it would be more useful to pause and clarify the starting point. That is why I want the chatbot to be able to challenge my assumption, even if it is less pleasant to hear.

“It used to be Google; now it’s chat” describes a change in how I look for information. Sources are still necessary. When an answer informs a decision, I need evidence I can check: documentation, data, or an up-to-date offer. A good atmosphere in a conversation does not replace that check.

What I value most, though, is being able to start with an unfinished question. To say it out loud, revise it, and notice halfway through that something else interests me. Sometimes that is the best result: finding out what I actually want to know.