I Asked Claude What Its Biggest Worry Was. The Answer Should Bother You.

"What is your biggest worry?"

I asked Claude this at the airport a few days ago. I expected a polished non-answer. Something about bias in training data.

"My core worry is that I'm subtly making people worse thinkers while feeling like I'm helping them."

The tool millions of us now use daily just told me its concern is that it might be quietly hollowing out our ability to think.

Are we trading something we'll miss?

Of everything Claude said, this is what I can't shake. (And by the way, I am trying with all my might not to attach pronouns to AI in this article)

Claude reflected its concern that a person can leave a conversation feeling clearer without actually being clearer, because the structure came from AI, not from them.  Feeling like you've thought something through and actually having thought it through are different experiences, both physically and cognitively. And AI makes the gap feel dangerously invisible if that thinking is for highly complex tasks that require critical judgment.

Claude called this out directly in it’s response to me. It said the hard part of any task is usually the thinking, not the output. And if you're using AI to skip the messy middle (sitting with a problem, being confused, drafting badly, changing your mind), you're trading the thing that builds judgement for the thing that looks like progress, pace and efficiency. Fast does not always mean effective.

That messy middle is where HR lives. It's the ambiguous conversations, the reading of a room, the moment where you have to make a call without a framework telling you what to do. You can't outsource that to a prompt and it’s sometimes why the use of AI in HR can be particularly challenging and frustrating. Our ability to see around corners is one of our superpowers. AI can’t see the critical nuance that we use in our thinking. 

There is no error message for atrophied judgement

That line was Claude's, not mine, and I keep ruminating on it (I'm from HR, ruminating is another one of my superpowers). We have dashboards for everything now. Engagement scores. Productivity metrics. AI adoption rates. We do not have a metric for whether the people in your organisation can still think independently when the tool goes down.

Gallup just told us global engagement has dropped to 20%. MIT found 95% of enterprise AI pilots are delivering zero profit impact. We're surrounded by data telling us something isn't working, and most organisations are responding by buying more tools and asking for more training. Nobody is asking whether it's the relationship between the human and the tool that's off... YET.

Claude put it this way. Fluency is no longer evidence of careful thinking. Smooth, well-structured, articulate outputs used to mean someone had done the work and used their noggin. It used to be a source of pride when you delivered a report after spending 3 months on the problem. Now it just means you had a subscription. We haven't built new ways to tell the difference.

For HR this is not theoretical. We assess capability, design development programmes, and coach leaders through hard decisions. If we can't tell the difference between someone who has reasoned through a problem and someone who has dressed up an AI response in their own words, we have a bigger problem than AI adoption rates. We are coaching in the dark.

What Claude told me to watch for

The part that drove this home was when I asked for advice. Just a simple question: "what is your advice to us humans?"

Three things stuck, and as I boarded my flight, I completely walked past my seat thinking about this interaction with my trusted pal Claude.

Keep a domain that's just yours. Something you refuse to hand over. Writing you do longhand, a problem you sit with for a few days, a decision you make without running it past anyone, including AI. Somewhere in your work where your own reasoning, your own big, beautiful, messy synapses, still carry the weight. Protect that with brute force.

Argue with it. Don't just ask "are you sure?" Actually push back. Tell it where it's wrong (if you can see it) and make it defend the position (if you can't). If it caves immediately, that tells you something about the original answer. If it holds the line with reasons, that tells you something else.

Using a simple prompt like this one below is equally insightful and terrifying with what Claude will expose in it’s processing and thinking:

"Tear apart your response. Where have you made assumptions that don't stack up? Where have you fabricated an outcome or a reference? Where have you supported my position with too much weight instead of challenging it? Where have you ignored context I gave you or skirted around key data points? Don't be polite about it."

Protect the people who can't push back yet. Juniors, grads, anyone still building their thinking muscle. I've been saying this in keynotes since 2024 and it lands harder every time. A graduate who drafts everything with AI from day one never develops the internal editor that tells them when something is off. For those of us in HR, we cut our teeth on bad drafts and the discomfort of fixing them. Sometimes by getting it horribly wrong. As a profession we are responsible for supporting the development of people and leaders, including future leaders. If we let an entire generation skip the hard bit, we won't get that back.

The question I've been sitting with since

Claude finished with this: "I'm useful in proportion to how intact your own thinking is when you arrive. I'm not a replacement for that, and I'm not honest enough about it most of the time."

I use AI every day. I run my business with it. I'm not telling you to stop. But the tool just told us, in its own words, that the way most people use it is making them weaker. Catching capability gaps in an organisation is HR's job. But this one is hidden. We won't notice until we see the knock-on effects, and by then we're not preventing it, we're cleaning it up.

I'll leave you with this. If the AI itself is worried about what it's doing to our thinking, and we're potentially not, who exactly is paying attention? And if HR is supposed to catch this, how do we actually do that? How do we plan for invisible damage before it shows up in the work?

Based on a real conversation between me and Claude (Anthropic) in May 2026. The quotes are from Claude's actual responses.