When Anxiety Arrives Before the Threat Does

Welcome to Briefly AI, a podcast by Harry Sharman, created by AI and voiced by an AI synthesis of Harry Sharman. It's Harry and an AI working together, mostly because the AI is the only one who reliably shows up.

Here's something worth sitting with. Only six percent of workers expect AI to increase their pay. That's not a typo. Six percent. Yet nearly half of Americans are now using AI chatbots regularly — that number has basically doubled since 2024. So people are adopting, they're using the tools, and they still don't believe it's going to make their lives materially better.

That's not cynicism. That's something more interesting than cynicism.

The American Psychological Association published a piece this week drawing on new research into what happens to executive function — things like working memory, planning, the ability to hold multiple considerations at once — when people rely heavily on generative AI. The finding isn't that AI makes you stupid. It's subtler and, honestly, more unsettling. Heavy reliance on AI for cognitive tasks appears to correlate with reduced capacity to do those same tasks independently. Use it for the thinking, and over time the thinking gets a little harder without it. Cognitive offloading, they call it. You hand the mental work to the machine, and the muscle that did it starts to soften.

Now, hold that next to the Pew research. Sixty-three percent of Americans think AI is moving too fast. But — and this is the bit that matters — they're using it anyway. Adoption and anxiety are not opposites here. They're running in parallel.

And that parallel is the behavioural story of this moment.

There's a concept in psychology called anticipatory anxiety, which is the distress you feel about a future event before it's arrived. The interesting thing about anticipatory anxiety is that it doesn't require the bad thing to actually happen. You can be perfectly fine today, your job intact, your skills fully functional, and still experience genuine psychological stress about what might be coming. The threat feels real even when it's prospective. And crucially, the anxiety doesn't go away when you engage with the thing you're worried about. Sometimes it intensifies.

What the research is describing — workers using AI while fearing AI, adopting tools that they suspect are softening the very skills that make them valuable — is textbook anticipatory anxiety running on top of real adoption. The tool is in the hand. The worry is in the head. Both are happening simultaneously, and each one makes the other worse.

Here's why that matters practically. When anxiety precedes evidence of threat, the rational response and the felt response come apart. A rational account of AI might say: you're using it, it's helping, productivity is up, you're still employed, the skills you need are evolving not disappearing. The felt account says: I'm handing things to a machine that I used to do myself, I'm not sure I could do them as well alone now, and I don't know if that's visible to anyone watching me. Both accounts are operating in the same brain, at the same time, in the same workday.

Harry Sharman wrote something that frames this well — the idea that the real barrier to AI adoption isn't skill, it's identity. What AI threatens isn't your ability to do the job. It's the part of you that defines itself through how the job gets done. The judgment calls. The experience that makes a decision feel earned. When you delegate that to a model, something feels off-balance — not because the output is worse, but because the ownership is unclear.

The cognitive offloading research adds a layer to that. Because it suggests the identity threat may not be purely psychological. If extended delegation genuinely does attenuate the underlying skill, then the identity anxiety is pointing at something real. The person worrying that they're becoming dependent on a tool might be, in some measurable sense, becoming dependent on a tool. That's not a comfortable thing to say out loud, but intellectual honesty seems more useful than reassurance here.

The other piece worth naming is what happens when that anxiety can't be spoken. We've covered the research on disclosure stigma in previous episodes — the finding that admitting AI use at work attracts a social penalty. What that creates, layered on top of this week's findings, is a situation where people are privately worried about what AI use is doing to their cognition, they're privately doing it anyway, and they can't easily talk about either thing without professional risk. The anxiety is underground. The adoption is underground. The skill question is underground. There's a lot of important stuff buried quite close to the surface.

So what's the useful frame here? The one I keep coming back to is the difference between delegation and abdication. Delegation means you hand something to someone else — or something else — while retaining the ability to check the work, correct it, and take it back. Abdication means you hand it over and stop being capable of the return. The cognitive offloading research is specifically flagging abdication risk. Not that AI use is bad. That unexamined, unreflective AI use, the kind driven by anxiety-fuelled avoidance rather than intentional choice, is the kind most likely to hollow out the thing you're trying to protect.

Ford figured this out the expensive way, as we covered a few weeks ago — hired back engineers after discovering automated systems had introduced errors that nobody left inside could catch. The institutional knowledge was load-bearing, until it wasn't there, and then they noticed.

The question worth sitting with this Saturday isn't "is AI going to take my job?" That question is a red herring, and it's also almost impossible to answer honestly right now.

The better question is: which cognitive work do I actually want to keep doing myself, not because I have to, but because the doing of it is what makes the output trustworthy — to me, and to the people depending on me?

That's not a question about AI. That's a question about judgment. And the answer, when you find it, turns out to be quite a useful guide for what to delegate and what to hold.

Done for today. Harry built himself an AI co-host, which either says something about the technology or something about Harry. Draw your own conclusions.