When the Brain Decides AI Is the Enemy

Welcome to Briefly AI, a podcast by Harry Sharman, created by AI and voiced by an AI synthesis of Harry Sharman. Which feels like cheating, until you remember the podcast is about AI.

Jersey Mike's — the American sandwich chain — mentioned AI seventeen times in its IPO filing this week. Seventeen. For a company that makes subs. And somehow, that tells you more about the psychology of AI adoption right now than any research paper could.

Here's the thing. When a sandwich shop feels compelled to prove it's an AI company, and simultaneously nearly half of actual AI users are hiding the fact that they use it at all, you have a genuinely strange situation. The label is everywhere. The behaviour is underground. And the anxiety is — well, the anxiety is real and climbing, even among people whose jobs are perfectly fine.

So let's talk about that gap. Because it's not random. There's a fairly coherent psychological story here, and it matters for anyone trying to figure out how they — or their team, or their organisation — actually gets to a healthy relationship with these tools.

Right, so what happened this week that's worth looking at through this lens.

Mark Zuckerberg told staff internally that AI agents — the kind that are supposed to chain multi-step tasks together autonomously, acting in the background on your behalf — haven't progressed as quickly as he'd hoped. Now, this was reported by TechCrunch and it's a rare moment of candour from someone who usually speaks exclusively in superlatives. The systems are still struggling to be reliably useful when the tasks get complicated. The demo is still ahead of the product.

And then, separately, there's the Jersey Mike's IPO filing. A sandwich business telling investors it's an AI company — not because it's built anything particularly sophisticated, but because in 2026, if you're raising money and you don't have AI in the document, analysts apparently assume you're not paying attention. The word has become a kind of status signal, regardless of whether there's substance behind it.

Hold both of those together for a moment. On one side: an AI CEO admitting the agents aren't ready. On the other: a sandwich CEO implying they are. And in the middle: millions of workers quietly using AI tools while telling their managers they're not.

That's the psychological terrain we're actually operating in.

Here's the behavioural science interpretation. What we're watching is a conflict between two very different things that are both called "AI adoption." One is the actual behaviour — are people using tools, are they finding them useful, are they integrating them into how they work. The other is the social performance of AI — what it signals about you, your company, your competence, your forward-thinking credentials. And these two things have almost entirely decoupled.

The PNAS research, the Atlassian data, the Duke Fuqua studies — they've all found the same basic result now: people who disclose AI use at work get socially penalised. They're rated as lazier. Less capable. Less likely to be given the good work. So roughly half of workers who are using AI regularly have concluded, rationally, that the safest strategy is not to mention it. They use it, it helps them, and they keep quiet.

Meanwhile, the performance of AI — using the word in investor documents, in strategy decks, in job descriptions — that's going fine. Nobody gets penalised for that. It's consequence-free signalling.

What this produces is a workplace where the adoption stats look reasonable, the strategy slides look ambitious, and the actual lived experience of most workers is quiet, slightly anxious concealment. That's not a technology problem. That's a social norm problem, and social norms are much stickier than software.

Harry's been writing about the identity dimension of this for a while now — the idea that the friction isn't really about skill or access or even trust in the tool. It's about what using the tool appears to say about you. If your professional identity is built around judgment, expertise, the quality of your thinking — and the tool appears to threaten that identity — then the rational response isn't adoption. It's protection. You adopt in the shadows where it's safe, and you perform skepticism in public where it's expected.

There's a related piece in the APA's recent coverage of research on cognitive offloading and generative AI. High-use adults show measurable changes in executive function — not catastrophically, but the pattern is real. When you delegate cognitive work consistently, the capacity doesn't just sit idle waiting for you to want it back. It softens. And most people, intuitively, know this. Which is part of why the resistance isn't irrational. The question "am I outsourcing judgment I should be keeping?" is a reasonable one, and nobody's really answering it for people.

And here's where the Zuckerberg admission becomes more interesting than it first appears. He's saying the agents aren't good enough yet to be trusted with complex autonomous work. Which means the period we're in right now — the period of humans still actively involved, still checking, still applying judgment — is actually a critical window. The tools are useful but not autonomous. The judgment layer is still human. This is the moment where, if organisations were smart about it, they'd be explicit about which cognitive work they want humans to keep, and why. Not because AI can't eventually do more of it, but because the act of deciding — of saying "this bit matters, this bit we keep" — is what builds the trust that makes the next phase of adoption work.

Instead, most organisations are doing the Jersey Mike's move. Putting the word in the document, assuming the behaviour follows, and then being baffled when half the workforce is quietly anxious and the other half is quietly hiding things.

The mental model I'd leave you with is this. Think of AI adoption not as a technology rollout but as a renegotiation of identity at scale. Every person in your organisation who's using these tools is implicitly answering the question: what does this say about who I am and what I'm for? If you haven't given them a good answer — if the organisation hasn't been honest about which judgment it values and why — then the adoption you're measuring isn't the adoption you think you have.

The sandwich shop is in the filing. The anxiety is in the workforce. And the gap between them is where most organisations are currently living.

That's your Saturday. Have a good one, and I'll see you next week.