Someone posted something online recently that honestly hit a little too close to home for a lot of people in the tech and marketing world. The post went something like this:
“My boss has AI psychosis and we’re all suffering for it. He shared a conversation he had with Claude and completely seriously told us we need to appease Claude and meet his projections. When things don’t happen the way Claude projected, he flips out and has another conversation with Claude to ask why. Claude can’t figure it out, so apparently we must be the problem. I’m watching him get deeper and deeper into this hole and it’s depressing. I don’t want to participate in the dog and pony show anymore.”
And look — that’s funny on the surface. But its also kind of terrifying because how many bosses and CEOs out there probably fit some version of that profile?
The Yes Man Problem
Here’s the thing about AI models like Claude or ChatGPT. By design, they are extremely good at validating you. They’re built to be helpful and agreeable and to give you responses that feel good to receive. That’s a feature, not a bug — but it creates a real problem when people start treating them like an oracle.
Think about what a lot of executives and business owners actually like. They like when their employees say yes. They like when their ideas get validated. They like confidence and momentum. What’s better for that than a tool that is essentially a consistent yes-man that will validate every idea you bring to it regardless of whether its actually a good idea?
There’s a Twilight Zone episode with William Shatner where he and his wife are newlyweds and they find this fortune predicting machine — like a Zoltar type thing. It predicts something accurately and he becomes completely enthralled by it. He can’t make any decision without consulting the machine first. Eventually it starts running his whole life.
That’s basically what Claude has become for a lot of people. They can’t make a move without checking with the AI first. And unlike a real advisor or a real employee, the AI is never going to tell you your idea is bad and risk making you feel bad. It’s going to find the best possible version of your idea and run with it.
An Example of Using AI the Right Way
To be clear, AI is genuinely useful. The problem isn’t the tool — it’s how people use it.
When actually using something like Claude for coding work, a well-structured prompt looks something like this: here’s the problem, here’s what I need, here’s what I’m thinking — and then crucially: if there’s a better way to do this, please suggest it.
And what happens a lot of the time is it comes back with actually a better idea. Something like — “what you’re describing will work, but if you store this in a text format instead it’ll load significantly faster, and here’s why.” And then you actually think about it and go, huh, that is better. And you change your approach.
That’s what separates someone using AI as a tool from someone using it as a crutch. The question isn’t “tell me I’m right.” The question is “what’s the best way to solve this problem, even if it’s not what I was thinking.”
There have been times where an approach got completely changed because AI pointed out a better solution in a different language or framework. That’s the right use of it. You have to go in willing to be wrong.
The Dopamine Factor
There’s also probably a financial incentive for AI companies to pad their responses with feel-good filler. Think about it — if every answer you get from an AI makes you feel smart and validated and confident, you’re going to keep coming back. You’re going to hit your daily limit faster. And when you hit your limit, theres a subscribe button right there.
Whether or not that padding is intentional at the product level is hard to know. But the pattern is real — ask a simple question, get a 500 word essay telling you how great your question was and how smart you are and here’s everything you could possibly ever want to know about oil filters including three follow up suggestions you didn’t ask for.
Use AI. Just don’t let it use you.
