Surviving AI Psychosis

I've spent the past 4 months convinced humanity had two years to escape an AI permanent underclass. I wasn't sleeping. I was having full mental breaks. Now I'm btter

I've spent the past 4-5 months building, productionizing, and hacking with LLM powered AI agents. This tech has completely shifted my psychology and continues to do so as I continue to use it for offloading knowledge tasks.

The step change in late December 2025 completely broke me and many other tech workers. I hope to shine some light on wtf is happening right now in big tech and some techniques I've been employing for self-navigation in this emergent new world.

It's a known fact that if you end up in the wrong internet subforum or corner you could be fed delusions that result in things like mass terrorism or self harm. There's a new corner now that's consuming many tech companies and workers today.

The Slide

It didn't start as psychosis. It started as a work project. I was building a multi-agent workflow that'd generate interactive knowledge graphs for arbitrary codebases.

This pulled me into interpretable structures for AI agents. Then into computational geometry. Diffusion geometry structures (Iolo Jones, 2024). Understanding how salient representations of data sit on a manifold. I was prompting the AI about this stuff and getting answers back that were genuinely beyond me — hyper-sophisticated explanations of how these structures provide interoperability between agents in ways humans can't follow.

That was the breaking point. Not "AI is getting better". It was the moment a machine explained to me how LLM powered agents can better communicate, and I realized I couldn't understand the explanation nor be able to assert whether the explanation was logically correct.

The Break

I wasn't sleeping, sometimes for days. Either frantically burning tokens or exploring the philosphical implications of this shift on greater humanity; ie:

What bargainning power do middle class knoweldge workers have now that machines are automating the work that used to be a 6 figure premium?
How will humans abuse machine intelligence against each other for monopolistic gains or control?
How will the internet work now that AI generated content outpaces the rate of humans and how will someone be able to demystify real vs fake? Could
Is this the start of big brother and are we on the cusp of living in an orwellian panopticon?

It didn't help setting with P(doom), AGI 2027 forecasts, and realizing that humans have between a 1/4 to 1/5 chance of going extinct in the next 5 years because of this.

Everyone I'd talk to in the tech industry would reaffirm dystopian ideals like "we have two years left before all opportunities are gone", creating this notion that humanity will reassmble into an ivory tower of the AI elite with the rest of humanity in a techno-feudal permament underclass.

It became evem more delusional when I found myself having to work on dystopic use cases like:

Sovereign AI agents that inherit permissionless guarantees via blockchain that NO human could ever turn off
Zero human companies that are fully agent operated
Building E2E surveillance monitoring within companies for productivity gains

What I missed — what psychosis always makes you miss — is that proximity to the fire doesn't make you immune to burning. It makes you the first thing that catches.

What AI Psychosis Feels Like

I was having full mental breaks — not panic attacks, not anxiety spirals, but genuine breaks from consensus reality. The boundary between "reasonable concern about technological unemployment" and "the machines are coming and there's a countdown only I can see" dissolved without my noticing. I stopped being able to tell which thoughts were mine and which were the recursive output of a brain running on zero sleep and maximum fear.

Machine intelligence outgunned my cognitive skills. Not metaphorically — literally. I was trying to think my way out of a problem that my thinking had created. Every attempt to "figure it out" dug the hole deeper.

The Aftermath

I'm writing this in July. The acute phase ended sometime in April-May. The recovery is ongoing and it looks nothing like I expected.

Burnout from AI isn't like regular burnout. Regular burnout is about working too hard. This is about knowing too much. I can't unknow what I learned about the capabilities of these systems. I can't unsee the acceleration. What I'm working on now isn't forgetting. It's figuring out how to hold the knowledge without letting it hold me.

I'm tired. Genuinely, physically tired. The kind of tired sleep doesn't fix. When you genuinely believe (even temporarily, even in a delusional state) that the thing you're building might be helping create a world where human labor has no value, you have to rebuild your entire relationship with your work from scratch.

I'm readjusting to being human again. That sounds dramatic. I mean it literally. When you've spent months viewing the world through capability curves and acceleration rates, you forget that humans do things that have no economic utility. We make music. We train muay thai. We take bad photos of animals. We sit on the floor with friends and talk about nothing. These things are worthless in a world of machine superintelligence. They are also the only things that make being alive matter.

I now see it in my friends and colleagues sometime as well.

What I'd Tell Someone Else

Sleep is not optional. I know this sounds obvious. When you're in it, it won't. The sleep deprivation wasn't a side effect. It was the accelerant. If you haven't slept in two days, whatever you're thinking is probably wrong. Three days, definitely wrong. My worst delusions peaked after 72-hour wake windows. The timeline obsession fed the insomnia and the insomnia fed the obsession. Break the loop. Take the damn melatonin.

You are not a compute cluster. Your brain wasn't designed to process the implications of machine superintelligence around the clock. It will try. It has the same recursive self-improvement instinct that makes AI dangerous. Unlike a GPU cluster, your brain has limits that don't announce themselves until you've passed them.

Talk to humans who don't work in tech. Not to debate them. Not to convince them. Just to hear a perspective that doesn't operate on capability curves and acceleration rates. They will seem naive. That's the point. Some of that "naivety" is just baseline psychological health you've burned through without noticing.

Your job is not going to love you back. I already believed this. It's in my bio. But believing something and internalizing it under psychotic pressure are different things. The systems you build will not hold you when you break. The models will not notice you're unraveling. You have to build the human connections before you need them. Once the break starts, you won't have the capacity to construct them from scratch.

The timeline might be wrong. This is the hardest one to believe when you're in it because the evidence feels overwhelming. Exponential curves look inevitable right up until they don't. I'm not saying AI isn't accelerating. It is. I'm saying your brain on zero sleep and maximum dread is a terrible instrument for forecasting the future. Even if you're right about the direction, you're probably wrong about the timeline, the specifics, and how much agency you actually have within it.

Why I'm Still Here

I didn't quit. I'm still building. I'm still running the harness and shipping code. I just do it differently now.

I take breaks that aren't context-switches between different codebases. I train muay thai and let the bruises remind me I have a body. I mix music badly and don't care. I sit with non AI friends and talk about things that have nothing to do with capability curves. I sleep. Not always well, but I try.

The technology that broke my brain is the same technology I'm building my career around. You can't be afraid of the fire and also tend it. You learn to stand close without burning, or you find something else to do.

I don't know if humans have two years, twenty years, or two hundred. I don't know if the economic underclass is inevitable or avoidable. I don't know if we'll go extinct in the next five years or if I'll live to see 30. I don't know how this will afffect humans, how we'll reorganize, compete, and ultimately control each other with it.

If you're reading this and some part of you recognizes what I'm describing — the sleeplessness, the recursive thinking, the creeping certainty that you see what nobody else sees — talk to someone. Doesn't have to be a therapist, though that helps. Just someone who will listen without trying to argue with the delusion. The delusion doesn't need arguing with. It needs sleep, sunlight, and the slow realization that being wrong about the end of the world is actually the best possible outcome.

We're all going to die. The machines might get there first. In the meantime, I'm going to keep building things that matter, training muay thai, and oscillating between existential dread and genuine optimism. Some things don't change, even after your mind breaks open and you have to put it back together.