Psychiatry and the Double Moat: Why Some Human Work Is AI-Proof
The status inversion, career bifurcation, and emotional infrastructure boom ahead.
An hour ago, on my way to my Psychiatry rotation, I recorded a voice note to myself. Psychiatry won’t get replaced by AI. The thought sat with me, and the more I turned it over, the more I think there’s something bigger in it than just one field.
This article is not really about psychiatry. That was simply the vessel through which this observation and insight came to me. That insight is processed here; the work to draw further analogies is yours.
The Core Idea
When LLMs take over the cognitive work that used to require a salary, what survives isn’t necessarily the most technically sophisticated work. It’s the most human work. And it turns out some of the most human work happens to come with regulatory moats that make it close to untouchable. Psychiatry is the clearest example I can find.
The Double Moat
What I mean by a double moat — because I think this is the crux of it — is two things working together. First is the regulatory side: prescription rights, controlled substance authority, licensing boards, malpractice liability. You can’t decide to prescribe lithium because you watched a video and feel confident about it. Second, and this one is harder to articulate but maybe more durable, is the irreducible human element. Transference. Genuine presence. The fact that when someone is in a psychiatric crisis they need a person — not a simulation of one — with actual skin in the game. Both moats together make psychiatry something close to AI-proof, and I don’t say that about many fields.
The Data Supports It
The numbers at least support the direction of this. Psychiatrist compensation in 2026 sits somewhere in the $340k–$376k range with chronic shortages that have been building for years. Mental health roles more broadly are projected to grow 17–26% through 2034, several times the overall economy’s pace. Meanwhile, entry-level software developer employment has dropped 13–27% in AI-exposed roles for younger workers. The prestige hierarchy is inverting in real time. I’m not sure most people making career decisions right now have caught on yet, and honestly I’m not sure I would have caught on if I weren’t thinking about it from inside medicine.
Why the Centaur Model Falls Short Here
There’s an argument I hear constantly that goes: layer AI tools on top of your existing human skills and you get the best of both. The centaur model. In some domains that’s clearly true. In deep relational work, in moral judgment, in the kind of presence required when someone is sitting across from you at their worst, AI augmentation doesn’t add — it dilutes. The chatbot therapy studies that exist do reasonably fine with mild cases. In anything moderate to severe, outcomes fall apart. An AI cannot own accountability the way a licensed human must, and that’s not a software bug that gets patched in the next release. I digress. The point is that the centaur model works where the work is mostly cognitive and the stakes of being wrong are bounded. In psychiatry neither of those things is true.
The Irony
The irony I keep sitting with is that AI is probably going to generate more demand for psychiatry, not less. Abundance creates its own crises. Decision paralysis. Meaning collapse. Status anxiety in a world where the old metrics of worth — skill, expertise, the salary you can command — are getting compressed faster than people can adapt their identities to absorb the hit. There’s early evidence linking heavy AI interaction to worse mental health outcomes in some populations. I’m not being alarmist. I genuinely think we’ll look back and see that the 2030s mental health boom was partially caused by the same technology that was supposed to liberate us from drudgery. Technology solves problems and creates adjacent ones. That has always been true, and there’s no particular reason to think this wave is different in that specific regard.
Practical Takeaways
The practical version of all this, as I see it: if you’re early career, seriously evaluate the high-human paths. Not because tech is a bad field but because the risk-adjusted math looks different than it did five years ago. The broad middle of software work is hollowing out fast, and the premium seats in AI orchestration are going to a small number of people. Psychiatry, certain types of coaching, palliative care, family law mediation — these are fields where the work resists automation by its nature and where regulatory barriers are genuinely hard to reproduce from scratch. If you’re already in tech, think about a portfolio: keep your AI leverage but build credentials in a high-human lane in parallel. The regulatory moat is hard to create yourself. Better to align with one that already exists than try to manufacture one.
What Would Make This Wrong
I’ll say what would make me wrong because I think you should say that when you’re making an argument. If psychiatrist and therapist real wages stagnate over the next decade while AI therapy shows equivalent long-term outcomes in moderate to severe cases at mass scale, the thesis weakens considerably. The current data points the other way. But I hold it with some humility.
Closing Thought
The machines will handle more of what we do. The people who will matter most in a world where that’s true are the ones helping other people figure out who they are. That work requires a human. And some of the humans doing it come with regulatory protection that makes them one of the more durable bets in the labor market right now. Psychiatry has been quietly sitting on this for years.
This piece grew from a raw voice note into a thesis I wanted to explore formally by writing down my thoughts. The data is current as of mid-2026; trends can shift, but the underlying human dynamics feel durable.
Thanks for reading. If this resonated, share it—especially with anyone navigating career choices.