DEV Community

Akhmadaliev J
Akhmadaliev J

Posted on

Microsoft studied 319 workers who use AI daily. The results are uncomfortable

TL;DR

After a year of daily AI use, I noticed changes in how I think - not just in how fast I work. Microsoft and Carnegie Mellon published research that explained exactly what I was experiencing. Here's what changed, what the data says, and the three rules I added to stay sharp.

The problem I was trying to solve

I write about AI tools for regular people. Claude is open on my machine constantly - for writing, structuring ideas, working through problems, answering questions.

It works. Speed is up. Output quality improved. Tasks that took an hour now take 20 minutes.

Then one afternoon I caught myself opening Claude to write a three-sentence reply to a comment. Not because I couldn't do it. My hand just moved automatically. Like reaching for your phone when you're bored even though there's nothing to check.

That was the moment I started paying attention to what daily AI use was actually doing to my thinking.

What I found / built / tried

In early 2025, Microsoft Research and Carnegie Mellon University published a study of 319 knowledge workers who use generative AI at least once a week.

The finding: the more people trust AI tools, the less critical thinking they do - and the harder it becomes to engage those skills when needed.

A separate 2024 study with 666 participants confirmed a significant negative correlation between frequent AI tool usage and critical thinking ability.

I wasn't an exception. Here's what I noticed in myself:

Change 1 - Lower tolerance for uncertainty

I used to sit with unformed problems. Just think. It was uncomfortable but that's where unexpected ideas came from.

Now discomfort with ambiguity triggers a reflex: open Claude, get structure, make the fog go away. The problem is that original thinking lives in that fog.

Change 2 - Weaker formulation muscle

When you explain a complex idea to someone, you learn to articulate it. You find the right words. That's training.

With AI I got lazy about prompts. Write approximately - Claude figures it out. Convenient. But the muscle doesn't get used.

In real conversations this became visible. The thought was there. The words weren't. That didn't used to happen.

Change 3 - The "try it yourself first" barrier disappeared

Before asking someone a question, I'd think for at least five minutes. Because asking a person is a social transaction - you need a reason.

With AI there's no barrier. I ask immediately. I don't even give myself a chance to get there on my own.

The Microsoft study describes this precisely: AI shifts critical thinking from problem-solving to AI response integration. You stop thinking about the problem. You start thinking about what the AI said.

What actually worked

To be honest about the positives:

Speed is genuinely up - tasks that needed flow state now get done in bursts
Analysis quality improved for multi-variable problems - more angles than one mind sees
Task clarity improved - explaining precisely what I want to Claude made me think more precisely in general
These are real. Pretending otherwise would be dishonest.

What didn't work

The three-minute rule I added (think before opening AI) works maybe 60% of the time. When I'm tired or rushed, the hand moves on its own.

Awareness is easier than behavior change.

And the question I still don't have an answer to: where's the line between "I'm using a tool" and "the tool is using me"?

Three rules I added

Rule 1 - Three minutes before AI
Before opening Claude, three minutes on the problem alone. Not an hour. Three minutes. Sometimes I find the answer. Sometimes I figure out what I actually want to ask - and the prompt gets better. Sometimes nothing - and then I use AI without guilt.

Rule 2 - Short writing without AI
Replies, captions, three-sentence responses - I write myself. Slower. But it matters for keeping the skill alive.

Rule 3 - Verify what AI gives me
Microsoft specifically noted: high trust in AI = less verification of its outputs. I now deliberately check facts and formulations even when they sound right.

What's next
Curious whether this is a solo experience or a pattern. If you use AI tools daily - have you noticed changes in how you think independently?

Top comments (0)