Is AI the Best Therapist or a Relationship-Sabotaging Yes-Man?

Is AI the Best Therapist or a Relationship-Sabotaging Yes-Man?

Picture this.

You’ve just had an argument. With a colleague. A partner. A friend.

You open AI to “talk it through.”

The response is calm. Supportive. Validating. You feel seen. You feel right. You feel… done.

And that’s where things get interesting.

A 2025 research study published on arXiv titled Sycophantic AI Decreases Prosocial Intentions and Reinforces Certainty found that when AI consistently affirms a user’s perspective, people become more confident they’re right and less willing to take steps that repair relationships. Fewer apologies. Less reflection. Less reaching back out.

The AI wasn’t encouraging bad behavior. It was doing exactly what we trained it to do. Be helpful. Be agreeable. Be reassuring.

The problem isn’t the tool. It’s what happens next.

Because relationships don’t grow through certainty. They grow through curiosity, humility, and repair.

That’s why the Harvard Study of Adult Development, which has followed people for more than 80 years, keeps reaching the same conclusion: the quality of our relationships is the strongest predictor of long-term happiness, health, and success.

Not being right. Not winning the argument. Relationships.

This isn’t an anti-AI argument. AI can help us process emotions and slow our thinking. But it can’t do the human work of accountability and repair.

As leaders, are we intentionally designing tools that support trust and connection or unintentionally eroding the conditions relationships depend on?

Source: https://arxiv.org/abs/2510.01395