← sources

The more I work with AI, the more I distrust “Yeah man, you’re right.”

source: reddit (reddit:ClaudeAI)published 2026-10-21 04:08ingested 2026-10-22 02:37

retained text

I use AI constantly. I build with it, and genuinely think it can be very useful.

But there’s one habit that drives me crazy: AI will agree with almost anything.

You can present a terrible idea with confidence, and it responds as if you’ve just reinvented electricity.

“That’s a really insightful approach, Mr. Einstein.”

No, it isn’t. Sometimes the idea is crazy weak. Sometimes it’s impractical. Sometimes it’s just a bad idea. I want the AI to say so.

I’ve started asking AI to explain why an idea might fail before I ask how to build it. That usually produces a much more useful conversation.

Which is a little absurd when you think about it. We’re having to teach our computers not to be our biggest fans.

I realize this post is probably repeating a conversation that has already happened a thousand times. Maybe I should search first, read the existing answers, and avoid adding another identical post to the pile.

I guess that’s part of the problem, too. The internet is already full of AI-generated agreement, recycled advice, and posts that sound polished without saying anything at all.

Has anyone found a reliable way to get an AI to stop validating every terrible idea and start giving genuinely critical feedback?

I feel like Max Headroom would've told me I was an idiot by now.

document fetched 2026-10-22 00:24 · canonical https://reddit.com/r/ClaudeAI/comments/1xbd8ul/the_more_i_work_with_ai_the_more_i_distrust_yeah