Is using AI every day making you worse at thinking?

When did you last write a first paragraph from scratch - no AI, no outline prompt, just a blank document and your own working memory? If you had to stop and think about that, the answer might tell you something worth examining.

A growing number of heavy AI users have started reporting an uncomfortable pattern: skills they used to have seem weaker. Writing from scratch feels harder. Holding a long chain of reasoning without offloading it somewhere feels less natural. Starting a task without reaching for an AI tool first feels vaguely wrong, like driving without GPS.

The instinct is to dismiss this as anxiety or technophobia. The research suggests otherwise.

What the studies are finding

A 2024 joint study from Microsoft and Carnegie Mellon University on AI's impact on critical thinking found that workers who relied on AI tools more heavily showed reduced engagement in their own independent reasoning. The more they trusted AI output, the less they verified it - and the less they worked through problems independently before asking the tool.

This is not a surprising finding when you situate it in the research on cognitive offloading. GPS navigation reduced people's ability to build internal spatial maps of unfamiliar cities. Calculators reduced mental arithmetic. Spell-check reduced spelling accuracy. When you consistently offload a cognitive task to a tool, the neural pathways that handle that task get less exercise. The skill atrophies. This is documented behavior, not speculation.

The question is not whether offloading causes atrophy. It does. The question is whether it matters.

When it matters and when it does not

Nobody cares that GPS reduced our ability to navigate by landmark. That skill is less useful now. Offloading it was a net gain. The same logic applies to many things AI can now do reliably - formatting, basic research, boilerplate text, routine code patterns.

Writing and reasoning sit in a different category. Not because they are noble or irreplaceable in themselves, but because of how they relate to evaluating AI output. If you cannot write a clear paragraph on your own, you lose something specific: the ability to judge whether the paragraph Claude or ChatGPT wrote is actually good, or just fluent and plausible-sounding. If you cannot reason through a technical problem independently, you cannot tell whether the AI solution is correct or merely confident.

The dependent skill underpins the quality check. That is the loop that makes the degradation more consequential here than it was with GPS.

How heavy users are navigating this

People who seem most thoughtful about this are not avoiding AI. They are being deliberate about which tasks they let it handle. A few patterns worth noting.

Some maintain what they call no-AI zones - specific task types they always do by hand. Personal emails. First-pass thinking on new problems. Initial outlines before any tool opens. The point is not to avoid efficiency. It is to keep specific cognitive muscles from going completely unused.

Others use AI for research and information gathering but force themselves to produce written outputs from scratch. They use it as a source, not a drafter.

A simpler version: before asking an AI tool, spend five minutes on the problem yourself first. Not to compete with the tool - you will not beat it on speed or completeness. But to keep the cognitive process active rather than passive.

None of this is anti-AI. It is the same logic as going to the gym when you own a car. You do not walk everywhere because cars are bad. You deliberately exercise capabilities you would otherwise lose.

The strongest counterargument

The most serious pushback on all of this: maybe the cognitive skills being degraded are not the ones that matter in a world where AI can do them better than you. Writing a first draft from scratch might be like doing long division when a calculator exists. The effort is not the point. The output is. If the AI version of your email is clearer and more effective than what you would write yourself, why does it matter that your unaided writing ability is declining?

There is a real argument that the answer is: it does not matter. If AI output consistently meets quality standards better than the output you would produce manually, then maintaining the manual skill is nostalgic rather than practical.

The accurate position is that no one fully knows yet. The research is early. Heavy AI use is less than three years old at scale. The long-term cognitive effects of consistently offloading complex reasoning will take time to study rigorously. We do not have the longitudinal data yet to say with confidence what is being lost and whether it matters.

A week-long experiment worth trying

For seven days, track every time you reach for an AI tool to start something - not to finish it, just to begin it. The first sentence of an email. The opening move on a problem. The initial structure of a plan. Notice the pattern.

Some of what you find will be completely fine to keep delegating. Some of it might be tasks where you would rather maintain the capability. The list will probably be shorter than you fear and longer than you expect.

Which cognitive tasks are you no longer comfortable starting without an AI tool first?