Keeping Your Judgment When the Tool Can Do the Thinking

AI can take real work off your plate. It can also quietly take your judgment — not by force, but because you stop exercising the capacity that keeps the judgment yours. The line between the two is worth knowing precisely.

Every powerful new tool arrives with the same question attached: does it amplify the person using it, or hollow them out? Calculators, spreadsheets, GPS, and now AI — each one lets you hand off something you used to do yourself, and each one raises the quiet worry that in handing it off, you lose something.

The usual framing of this debate is unhelpful, because it sorts people into camps: the disciplined few who'll use AI well, and the passive many who'll be diminished by it. That framing is both flattering and wrong. The more useful question isn't what kind of person are you — it's what specifically can you hand to the tool, and what does handing it over cost you when you do? Because the answer isn't the same for every task, and the cost isn't always visible while you're paying it.

Agency isn't a trait. It's a capacity that gets used or lost.

Start with what's actually at stake, because "agency" gets thrown around loosely. Agency isn't a personality type some people have and others don't. It's closer to the felt sense that your own actions matter — that you are the one deciding, not merely watching a decision happen. And the research on how that sense forms and fades is clear about one thing: it's built and maintained through doing, and it erodes through not doing.

This is the finding underneath learned helplessness. Passivity turns out to be the default response when action stops feeling connected to outcomes; the sense of agency is the thing that has to be actively maintained, not the thing that erodes only under damage (Maier & Seligman, 2016). Which means agency is less like a fixed trait and more like a muscle: exercised, it holds; unused, it quietly weakens. That reframes the AI question entirely. The risk was never that AI makes some people passive and not others. It's that handing over the doing — any doing — gradually weakens the capacity that doing was maintaining, in anyone, regardless of how sharp they started.

What the research on automation already told us

We don't have to guess about this, because AI is only the newest instance of a pattern studied for decades under the heading of automation. When people work alongside a capable automated system, two specific things reliably happen to their judgment.

The first is complacency: as the system proves reliable, people monitor it less. They stop actively checking, on the reasonable-feeling assumption that the system has it handled. The second is automation bias: people come to accept the system's output without independent verification, and — this is the sharp part — they do so even when the output conflicts with information they themselves hold (Parasuraman & Manzey, 2010). The tool's answer starts to feel more authoritative than their own read, so their own read goes quiet.

And underneath both is the slower cost: skill degradation. A capacity you stop exercising atrophies. Operators who lean on automation for a task gradually lose the ability to do that task manually — which matters most at exactly the wrong moment, when the automation fails or hits a case it can't handle and a human has to step in. The person is now out of the loop precisely when being in it counts. The de-skilling is invisible right up until the tool reaches its edge, and then it's the whole problem.

None of this is an argument against the tools. It's a precise account of what over-relying on them does, so you can tell the difference between using AI and being quietly de-skilled by it.

The line: what you can hand over, and what you can't

This gives a usable rule, and it's not "use AI less." It's about which layer you delegate.

You can safely hand over the doing — the drafting, the summarizing, the first-pass analysis, the mechanical and the well-structured. That's what the tool is genuinely good at, and offloading it frees you for higher-value work, exactly as the optimists say.

What you can't hand over — or rather, can't hand over without paying for it — is the judgment about whether the output is right. And here's the trap that makes this hard: judging whether an AI's answer is correct requires the very expertise that doing the work yourself would have built and maintained. If you've outsourced not just the task but the underlying competence, you're no longer able to evaluate what comes back. You can accept it or reject it, but you can't actually judge it — which means you're not overseeing the tool, you're deferring to it. The oversight becomes theater.

There's a second thing that doesn't delegate: knowing whether you've been captured. The uncomfortable finding from the psychology of bias is that people can't reliably introspect their way to "am I over-trusting this?" — the sense of having evaluated something and the fact of having evaluated it feel identical from the inside (Pronin, Lin, & Ross, 2002). So "I always check the AI's work" is not something you can verify by consulting your own confidence. It has to be built into how you work, not left to a feeling of diligence that will report back "all good" whether or not it's true.

Margaret Hamilton, who led the Apollo flight software, built systems designed to keep humans able to intervene — error-handling that surfaced problems rather than silently absorbing them, so a person stayed in the loop and in a position to act. That's the posture worth borrowing. The goal isn't to do less with the tool. It's to stay in a position to catch it when it's wrong — which requires keeping the judgment, and the competence the judgment rests on, in your own hands.

In practice

Delegate the task, not the understanding. Use the tool to do the work faster, but make sure you could still evaluate the work without it. The moment you can't tell whether the output is good, you've crossed from using it to depending on it.

Build the check into the process, not the intention. Since your own sense of "I checked" isn't reliable evidence that you did, install actual verification — a step where you test the output against something independent, not a vibe that it looks right. Especially under time pressure, which is exactly when the skipping happens.

Notice what you're losing the ability to do. If there's a skill you now only exercise through the tool, that skill is quietly degrading. That's fine for things you're happy to fully outsource forever. It's a real cost for anything where you'll eventually need to step in — and you don't want to discover the gap at the edge case.

Keep the decisions that are actually yours, yours. The tool can inform a judgment. It shouldn't become the place the judgment happens, because a judgment you've fully handed over is one you can no longer be sure is right — and, over time, one you're less and less able to make at all.

The reframe

The question was never whether AI makes people passive or powerful, as though that were a fact about their character. It's that any tool you hand your doing to will, over time, weaken the capacity that doing maintained — and the capacity most worth protecting is judgment: the ability to tell whether the answer in front of you is right, which rests on competence that only stays sharp through use.

AI is an extraordinary amplifier for the work you can verify and a quiet solvent for the judgment you stop exercising. Keeping agency in an AI world isn't about being one of the disciplined ones. It's about knowing exactly where the line is — hand over the doing, keep the judging — and building your work so that the judgment, and the skill it depends on, stay yours.

References

Maier, S. F., & Seligman, M. E. P. (2016). Learned helplessness at fifty: Insights from neuroscience. Psychological Review, 123(4), 349–367.

Parasuraman, R., & Manzey, D. H. (2010). Complacency and bias in human use of automation: An attentional integration. Human Factors, 52(3), 381–410.

Pronin, E., Lin, D. Y., & Ross, L. (2002). The bias blind spot: Perceptions of bias in self versus others. Personality and Social Psychology Bulletin, 28(3), 369–381.