What the evidence actually says
Your raw cognitive capacity does not shrink from using AI. Intelligence is not a muscle that wastes away when you delegate a task. That framing is wrong, and worth putting down immediately.
What does change is practice. Cognitive offloading research shows that when you consistently hand a mental task to an external tool, your ability to perform that task without the tool weakens over time. This is not a theory. GPS studies have demonstrated measurable reductions in spatial memory and navigation ability in regular GPS users compared to those who navigate without it.
The same principle applies to writing, analysis, and reasoning. If AI drafts your arguments, structures your thinking, and generates your first response to every hard question, you practice those things less. Less practice means less fluency. That is the actual risk.
What this means for knowledge workers
Most knowledge work involves a small set of high-value cognitive tasks: forming judgements under ambiguity, constructing arguments, spotting what is wrong with a plan, synthesising conflicting information. These are the tasks that get you hired, promoted, and trusted.
If AI handles the first draft of every memo, every analysis, every proposal, you are watching those specific capacities get less exercise. You may still hold the same job title. Your output may even look better. But your ability to produce that output independently, under pressure, without the tool, quietly deteriorates.
This matters most in the moments when AI is not available or not appropriate: a boardroom conversation, a difficult client call, a decision that cannot wait for a prompt. Those moments reveal what you can actually do.
Three concrete things to do
First, identify the two or three cognitive tasks that are most central to your work. These are the ones where your judgement is specifically what your organization pays for. Protect practice time for those tasks. Do them yourself before you ask AI to help.
Second, use AI after you think, not instead of thinking. Write your own rough argument first. Form your own view on the data. Then use AI to test, extend, or improve what you have already built. This keeps the cognitive work with you.
Third, review AI output critically and specifically. Do not ask whether it sounds right. Ask where the reasoning is weak, what it has missed, and whether you would have reached the same conclusion. That act of critical review is itself a form of practice.
Steve Raju is the author of Cognitive Sovereignty: How To Think For Yourself When AI Thinks For You, published April 14, 2026.