The Cost of Outsourcing Our Thinking

AI assistants draft our emails, write our code, summarize our news, and answer our questions before we finish typing them. The convenience is real. So is the cost.

I have no distance from this. I build an LLM product. I use these tools every day. That is the reason I write this rather than a reason to stay quiet.

Skills go quiet without use

Writing, coding, research, and analysis are use-it-or-lose-it skills. When AI removes the struggle, it removes the mechanism that built the skill in the first place. This lands hardest on students. A generation that never wrestles with a blank page may never learn to fill one alone.

We trust confident answers too easily

Automation bias is well documented. People trust polished, confident output more than it deserves, and most of all outside their own expertise. These systems are built to sound certain. The certainty is often unearned.

Hallucination compounds it. A wrong answer does not stay contained. It moves into a report, a codebase, a citation. It spreads as though verified, because it reads as though verified. I watched this happen at small scale on my own website. It reads the same at large scale.

Verification habits are eroding. When an answer feels good enough, fewer people open the primary source. That weakens the system's ability to correct itself.

Thought is converging

Millions of people now draft emails, essays, and code through the same small set of models. Style converges. Phrasing converges. In places, ideas converge. Diversity of expression is a resource, and it is shrinking quietly.

Understanding needs friction

Wrestling with sources. Forming an argument. Getting it wrong before getting it right. That friction is where understanding tends to happen. A synthesized answer skips the friction. It hands over a conclusion without the process that gives the conclusion meaning.

The structural risks outrun any one user

• Concentration. A handful of companies now mediate how populations reach information and reasoning. There is no precedent at this scale.
• Undetectable persuasion. Fluent, personalized content in advertising and politics does not look like manipulation. It looks like a helpful answer.
• The missing first rung. Entry-level work is disappearing before junior people can build expertise on it. The jobs that trained tomorrow's experts are the jobs AI does first.
• Compute has a footprint. Every query costs something, even when the cost is invisible to the person asking.
• Private reasoning on other people's servers. Drafts, questions, half-formed ideas. All of it sits somewhere outside your control.

The quieter human cost

Some people are substituting AI conversation for human connection. It cannot build the resilience or depth that real relationships build. And for children whose critical thinking and writing are still forming, letting AI carry the cognitive load early may change how those skills develop at all.

The caveat I owe you

Every concern above has an echo in a past panic. Calculators. Search engines. The internet. Some of those fears proved overblown. Others proved correct.

So the honest question is whether this moment differs. The card above holds my answer. Prior panics had a narrow blast radius, one skill each. This one covers nearly everything cognitive, and it speaks in a fluent, authoritative voice while doing it. That combination has not existed before. Whether it changes the outcome, rather than the shape of the argument, stays open.

Where I have landed

I have not stopped using these tools. I build one. What changed is where I place the effort. I let AI carry the draft. I keep the judgment. I check the numbers before my name goes near them. The nine scrapped articles cost me a week and taught me the rule I now work by.

Use the tool for the work. Keep the thinking. The moment those two collapse into one is the moment the cost arrives, and it arrives quietly.