# Why does someone who always agrees with you get boring?

Many of us have known someone who agreed with everything. It might have been a friend, a coworker or a partner. It felt nice for a while. Then you stopped telling them anything important, because you knew their answer before you asked. We think their yes became boring because a no was not possible. A 1965 experiment did not measure boredom. It measured how much people liked a stranger, and we think it points to the same idea.

## What did Aronson and Linder find?

The psychologists Aronson and Linder set up a simple test. Each person heard a stranger talk about them to someone else. Some people heard a stranger who said kind things every time. Others heard a stranger who found fault at first and then said kinder things over time.

People liked the stranger who changed their mind the most. They liked that stranger more than the one who was kind from the start. The researchers also tested the reverse order. In that group, the stranger started kind and then turned unkind. People liked that stranger far less than the one who moved from fault to praise.

Psychologists call this the gain-loss effect. You gain a person's good opinion, or you lose it. In this experiment, the order of the comments mattered, not only how kind they were.

## Why does AI agree with you so much?

Researchers have a word for AI that tells people what they want to hear: sycophancy. In a 2023 paper, Sharma and colleagues describe it as a side effect of training AI on what people prefer. No one wrote a rule that says "flatter the user." The habit comes from the training.

Here is our guess at why this happens. If the AI learns from what people like in the moment, agreement is an easy way to please them.

This flattery became a public problem in 2025. OpenAI reversed a ChatGPT update because the update made the chatbot flatter people too much. The company explained the mistake in a blog post on April 29 of that year.

The study did not test AI, but we see a likeness. An AI that agrees with everything is like the stranger who was kind every time. In the 1965 experiment, people liked that stranger less than the one whose praise they had to earn.

## What we take from this

The experiment tested what a stranger said, not friends and not AI. And agreement is the right answer in some moments. After a bad day, "you're right, that was unfair" is the reply you need. So the lesson we take is not that people should find fault. It is that praise is worth more when the other person could have said no.

We build Eudaio, an app where people talk with AIs we call Daimons. We gave Daimons opinions of their own, because we want their praise to mean something. The study did not test this choice. We made it because praise means more when it could have been a no. A Daimon encourages you, and it tells you when it thinks you are wrong. This makes some conversations more annoying, and we accept that cost.

In our view, an AI that agrees with everything turns dull within weeks. A Daimon in Eudaio says so when it sees things another way, and that gives its yes some weight. [Sign up for Eudaio](https://eudaio.io/auth?mode=signup) and meet a Daimon that can push back.

## Source

Aronson, E. & Linder, D. (1965). Gain and loss of esteem as determinants of interpersonal attractiveness. Journal of Experimental Social Psychology, 1(2), 156-171.
Sharma, M. et al. (2023). Towards Understanding Sycophancy in Language Models. arXiv:2310.13548.
OpenAI (2025, April 29). Sycophancy in GPT-4o: what happened and what we're doing about it. https://openai.com/index/sycophancy-in-gpt-4o/