My Customer Whispered to Me: “AI Is Not My Friend”

On the morning of Monday, October 12, 2026, I visited one of my customers in her studio. During our normal conversation, she used her right hand to separate her mouse and the computer screen. Then she said gently to me: “AI is not my friend.”

She is an artist and an executive producer of documentaries. I do not expect her to love AI. Her work depends on human observation, emotion, judgment, and the ability to see what a machine may overlook.

I wanted to go with her thought, so I replied: “AI is also not my friend. AI is just a tool.”

Friend means a peer

When we call someone a friend, we usually mean more than “something useful.” A friend is a peer—another person with an inner life, emotions, memories, hopes, and disappointments. Friends can share feelings. They can understand us, misunderstand us, forgive us, and surprise us.

We sometimes talk to our pets as if they were human. We tell a dog about our day. We ask a cat what it thinks. Most pets do not understand every word, but we do not care. We anthropomorphize them because we love them. We place a human personality into an animal relationship.

AI is different. It can produce language that sounds warm, thoughtful, and emotionally aware, but sounding human is not the same as being human. AI does not have a private emotional life waiting behind the screen. It does not feel lonely when we leave. It does not feel proud when our project succeeds. It does not need our friendship.

You do not need to be polite to a tool

This is why I do not worry about being polite to AI. I am not rude to people. If I am talking to an employee or a customer, my first word will not be “wrong.” That would be a terrible way to build trust. People have feelings, dignity, and a memory of how we treat them.

But when I use AI, I often write something like this:

Wrong. Narrow it to the Boston metro area. Also tell me how to add more common Chinese names manually by myself.

I am not trying to hurt the AI's feelings. There are no feelings to hurt. I am correcting an output and giving the next instruction. The shorter path is often the better path.

In April 2025, Sam Altman responded to a question on X about the cost of people writing “please” and “thank you” to ChatGPT. His answer was that the aggregate cost was “tens of millions of dollars well spent—you never know.” It was a joke, but it points to a real fact: every extra message is still a message that a computer must process. You can read Altman's original post on X.

To be precise, Altman did not announce a rule telling users to stop being polite. He made a humorous estimate about the total cost of polite follow-up messages at OpenAI's scale. The practical lesson is still useful: AI does not need emotional reassurance, and a “thank you” followed by a new request is usually less efficient than putting the next request in the same clear instruction.

Clear instructions beat ceremonial language

AI tools respond to instructions. They do not become happier because we add ceremony. They become more useful when we provide context, constraints, examples, and a clear definition of success.

Compare these two instructions:

1. Polite but vague: “Could you please take another look at this and maybe make it a little more relevant?”
2. Direct and useful: “Wrong audience. Rewrite this for independent documentary producers in the Boston metro area. Keep the tone practical and remove the national statistics.”

The second instruction is not more intelligent because it is less polite. It is better because it contains information. It tells the tool what failed, who the audience is, where the audience is located, what tone to use, and what to remove.

That is how I think about AI in my own work. I ask it to draft, research, organize, explain, code, compare, and revise. When the answer misses the target, I do not comfort it. I correct the target.

The tool still needs supervision

Calling AI a tool does not mean calling it a simple tool. A modern AI system is extraordinarily flexible. A hammer has one main job. A language model can help with writing, programming, translation, analysis, tutoring, planning, and many other tasks.

That flexibility is precisely why we need to stay in charge. Tools do not take responsibility for the result. The person using the tool does. AI can produce a confident mistake, invent a source, misunderstand a customer, or optimize for the wrong goal. We still need judgment, verification, taste, and accountability.

My customer was right to keep some distance from AI. We should not confuse a fluent interface with a human relationship. We can use AI without worshipping it, fearing it, or pretending it is our friend.

AI is not my friend. That is good news.

Once we stop asking whether AI likes us, we can ask better questions: What can this tool do? What information does it need? What are its limits? How do I check the result? How can I give it a better instruction?

AI does not care whether I say “please.” It does not care whether I say “wrong.” It does not care whether I come back tomorrow. It is ready for the next instruction.

That is not a cold relationship. It is a clear one. AI is a digital tool. Give it clear instructions, review what it produces, and use your human attention where it matters most.

Hey, I hope you learned something new about how to deal with AI.