AI Psychosis
A specific kind of mania sets in when you first experience what agentic AI can ship in production. The clinical phrase — AI psychosis — was coined by Danish psychiatrist Søren Dinesen Østergaard in a 2023 Schizophrenia Bulletin editorial about generative models triggering clinical delusions. Builder AI psychosis is different, and almost everyone shipping with agents right now has a case of it.
It starts innocently. You delegate a task that used to eat an afternoon — drafting an SEO brief, cutting a vertical video, scraping a dataset — and an agent finishes it in four minutes. You feel a jolt of leverage: you just bought back three hours of your life. Then, instead of logging off, you immediately fill those three hours with three more ambitious pipelines. That's the first symptom.
AI tools don't buy you leisure time — they blow open the ceiling on solo throughput, replacing low-grade manual work with 16-hour days of high-frequency decision fatigue and arbitrage anxiety.
- The Jevons paradox of personal output — saving 3 hours on execution doesn't shorten your workday; it tempts you to spin up 3 new automated workflows.
- Pure strategy is exhausting — stripping away mechanical grunt work removes the low-cognitive-load "flow state" buffers from your day, leaving only nonstop taste and routing calls.
- Arbitrage anxiety — when content costs approach zero and API gates are closing, every hour away from the terminal feels like watching a temporary market inefficiency expire.
The High
Before AI, my days revolved around managing freelance writers, editing Google Docs, and herding publishing calendars. I had never touched video production. Today, I run programmatic SEO pipelines, render multi-scene video reels, cross-post across social networks, and pressure-test product strategy — entirely solo.
The inflection point wasn't a single chat model; it was realizing that any repeatable creative recipe can be codified into a skill and put on a cron schedule. Once you stop doing the task and start engineering the loop that does the task, one operator genuinely outputs like a 20-person agency. Looking at your Friday commit and publish logs feels intoxicating.
The Paradox
Here is what nobody warns you about: you end up more exhausted, not less.
On paper, that makes no sense. I'm no longer scrubbing timelines in Premiere, uploading MP4s to six dashboards, or chasing late freelance drafts. Yet my brain is more fried at 8 p.m. than when I did everything by hand.
Why? Because delegating execution leaves behind 100% high-stakes judgment. All day, every day, you are context-switching across strategic bottlenecks: Is this content angle working? Did this prompt drift? Should we kill this format or double the cron frequency? Manual execution used to act as a cognitive buffer — you entered a mechanical flow state for two hours and gave your prefrontal cortex a rest. Pure direction has no autopilot. You trade grunt work for relentless decision fatigue.
Arbitrage Anxiety
Worse than the decision fatigue is the ticking clock.
Right now, there is a glaring market inefficiency for technical operators. Written and video production costs have collapsed toward zero. Short-form algorithms on TikTok, Instagram Reels, and YouTube Shorts still reward high-velocity programmatic media. Meanwhile, legacy platforms are locking down their APIs, and old distribution moats are eroding faster than new gatekeepers can form.
This isn't FOMO; it's arbitrage anxiety. When you spot a mispriced asset in a market, you know the spread won't last. Within six to twelve months, the loopholes tighten, platforms throttle synthetic volume, and SERPs saturate. Every Saturday afternoon you spend away from the keyboard feels like watching a temporary alpha window close.
I told my fiancée recently that we have to lock in while the window is open. Going outside or taking a slow weekend feels irrationally expensive when an afternoon of wiring up a new skill can compound 24/7 for the next six months.
Always On
In practice, builder AI psychosis looks like 6 a.m. to 10 p.m., seven days a week — not because a boss demands it, but because the possibility space expanded so violently that logging off feels like leaving money on the table.
Before LLMs, output scaled linearly with human hours. Everyone was throttled by the same biological bottleneck. Now the game has shifted from who grinds the most hours to who compounds automation the fastest. A solo builder with sharp orchestration can out-ship a ten-person team stuck in manual workflows — so the return on obsession has never been higher, and the perceived opportunity cost of rest has never been steeper.
The Bet
The darkest joke of all is that none of this plumbing may matter in three years.
If frontier labs actually deliver autonomous systems that one-shot entire businesses end-to-end, today's clever prompt harnesses and multi-agent crons will get paved over overnight. When everyone has infinite one-click leverage, nobody has an edge.
I know that — and I still wake up at 6 a.m. Because it's an asymmetric bet: if frontier models flatten the board tomorrow, I lose nothing I wouldn't have lost sitting on the sidelines. If capability plateaus or diffuses slowly over five years, the operators who built distribution and cash flow during the arbitrage window win the decade.
I'm trying to catch lightning in a bottle one more time so I never have to work for money again. That's AI psychosis: rationally defensible, completely consuming, and immediately recognizable to anyone else staring at three terminal windows at midnight.
Newsletter
Get the next post by email.
One email when I publish something new. No spam, no fixed schedule, unsubscribe anytime.