How to Stay Sharp in the AI Era

How do I stay sharp in the AI era?

Keep the part of the work that creates the skill. Use AI for retrieval, drafting, and checking, but do the attempt, the judgment, and the explanation yourself. The goal is not less AI. It is making sure the thinking that builds judgment stays in your hands while the routine parts move out.

Most people describe the same experience: a few months of leaning on an assistant for everything, then a quiet feeling that they could not produce the work alone anymore. That feeling is worth taking seriously, because it is pointing at something specific that happened to your habits rather than something that happened to your brain.

The distinction that matters is between offloading storage and offloading transformation. Looking up a date, a formula, or a definition is storage, and no one gets sharper by memorizing phone numbers. Forming the argument, weighing the options, and retrieving the conclusion are transformation, and those are the moves that build the mental model you keep.

Is AI actually weakening how I think?

Not automatically, and the research picture is more careful than the headlines suggest. What surveys and studies report is a pattern of reduced effort and reduced confidence when people lean on AI for tasks they would otherwise reason through. That describes a habit, not a verdict about your brain. The mechanism is practice you stopped doing, not a new injury.

A Microsoft Research survey of knowledge workers found that people reported putting in less cognitive effort and feeling less confident in their own answers when they used generative AI for their work. Those are self-reports, and they are a signal rather than proof. MIT Media Lab researchers studying people writing with ChatGPT reported weaker engagement and less sense of ownership over the writing compared with people who wrote unaided.

The wider culture has absorbed the worry too. The Harvard Gazette asked in late 2025 whether AI is dulling our minds, and Oxford made "brain rot" its word of the year, which tells you how many people recognize the feeling.

What is cognitive offloading?

Cognitive offloading is using a tool to carry mental work you would otherwise do yourself. The term is neutral. Some offloading frees attention for harder things, and some of it removes the very effort that produces learning, which is why the useful question is never "is offloading bad" but "what exactly did I hand over."

The rule of thumb: offload the lookup, keep the reasoning. If the assistant does the interpreting, comparing, and concluding while you nod along, you have handed over the transformation step.

Which parts of a task should I do before asking AI?

Do the attempt, the skim of the primary source, and a first pass at your own judgment before you ask. Even a rough or wrong attempt gives you something to compare the answer against, which is what converts the AI output into a check on your thinking rather than a replacement for it. Five minutes is enough to start.

Write your guess down before you prompt. One sentence on what you think the answer is and why. Then when the assistant responds, you are reading it as a second opinion instead of as the source of truth, and you will notice immediately when it agrees with you for the wrong reasons.

How do I check an AI answer instead of absorbing it?

Inspect the chain, not just the conclusion. Ask where each claim comes from, open the sources it names, and see whether they actually say what the answer claims. Then compare it against your own attempt and decide which parts you accept, which you reject, and which you still need to verify. Confident tone is not evidence.

The tell is a fluent answer with no traceable sources. Fluency is cheap to produce, and it is exactly what makes an unverified answer feel finished. Treat a missing citation as an open question rather than a small omission.

What daily practice keeps comprehension active?

Read or listen to one real thing, close it, and rebuild the main claim and one supporting detail from memory before you check anything. Then, if you use AI, ask it to challenge your reconstruction instead of producing one. Ten minutes of that targets the comprehension, recall, and analysis that AI cannot exercise on your behalf.

A tool helps when it removes friction from the recall and check step. Absorb is built around short comprehension sessions with scored feedback, so you can see what you actually absorbed instead of trusting the feeling that you did.

Keep the material real. Use the document, briefing, or meeting you actually have to deal with, not a passage chosen to be pleasant. Practicing on your own material is what makes this feel like a productive alternative to doomscrolling rather than one more task to maintain.

How can I measure whether I am still thinking for myself?

Test the far thing, not the assisted one. Once a week, take a real task you used AI for and do a version of it unaided, then compare the two. If you can explain the reasoning from memory, and defend it when someone pushes back, the thinking stayed yours.

Expect the unaided version to feel worse: slower and less polished, because you are doing the parts the tool was doing. That discomfort is the practice working, not evidence that you have lost the ability.

None of this is a promise about memory, intelligence, or long-term brain health. It is a habit: attempt, inspect, compare, decide, and explain from memory. Keep those five moves and the assistant stays a tool you direct rather than a substitute for your judgment.