AI Psychosis - How to Avoid Losing Your Mind in the Age of AI

AI is wonderful. AI is stressful. AI is powerful. AI is exhausting.

This is a journal entry about the last couple of years of my life as a developer, written after a forced two-week break from every AI tool and every line of code. I use the word "psychosis" as a metaphor, nothing clinical. But the feeling of losing my grip was real, and I think a lot of us are going through it quietly.

Finding Love for Programming

I've been a software engineer for over 10 years now. I started programming when I was 18 years old, and I was immediately absorbed by the ability to solve problems with code.

I still remember hitting the flow state for every single project I worked on during college. I would spend hours in front of my computer, completely immersed in the code. I'm sure you can relate, we've all looked at the clock and realized it's 3:00 AM and we haven't eaten dinner yet after spending countless hours mining in Minecraft.

Well programming is the same, countless hours, missed dinners and sleepless nights. Except at the end you have something that works, and you understand it down to the last line because you typed every one of them.

I love programming.

Keep that sentence in mind, because everything that follows is about what happens when the thing you love starts writing itself.

Welcome ChatGPT

That was the whole announcement. One tweet. I remember trying it like everyone else, and I remember not using it for real work. It was a toy, a very impressive toy, but my code did not live there.

At the time I was finishing college and using GitHub Copilot as an autocomplete. It was good at guessing the line I was about to write, and that was it. I was still the one thinking, and I still knew everything that happened in my codebase.

Looking back, that was the last period where AI was a tool in my hand instead of a hand on the wheel.

Cursor, Then Claude Code

My real introduction to modern AI development was Cursor. This was before agents, when the AI lived in a sidebar and edited the file you were looking at. I genuinely enjoyed it. It felt like pair programming with someone who had read every documentation page on the internet, and I still read every diff.

Then I switched to Claude Code, and that was a different animal. It was amazing. The AI did not just suggest code anymore. It ran commands in my terminal and came back with a summary of what it had done. The better the models got, and the better the tools got, the faster we could build.

I looked at code less and less. I stopped being the person typing and became the architect. I would describe the system and let Claude write the implementation. Whole features would land while I was thinking about the next one.

I want to be honest: I loved that period. I could build faster than ever, and I could build more complex systems than I would have dared to start alone. Things that used to be months of evenings became weekends.

This is where the story is supposed to end, with the happy developer shipping ten times faster. It is not where it ended.

Too Many Possibilities

Here is what I think the psychosis actually is, at least for me.

When you are a software engineer with an agentic AI at your fingertips, you have no constraints anymore. Every idea is buildable, and every "what if I also added..." costs one prompt. The friction that used to stop you from starting a new project, the sheer amount of typing it would take, is gone.

And it turns out that friction was doing a job. It was forcing me to choose.

Without it, I jumped. I would get 70% of the way into something, then have a new idea and start that instead. The AI never says "are you sure?". It says "great idea, let me scaffold that for you". So I ended up with a graveyard of half-finished projects, each one technically impressive and none of them published.

Infinite possibility is noise. Structure has to be put in place to actually finish anything, and I had removed all of mine.

Waiting for the Token Reset

This is the part I am least proud of.

I was paying $200 a month for my subscription, and that subscription comes with usage limits that reset on a schedule. At some point my brain reframed the whole thing as a resource optimization problem: every token I did not use before the reset was a token I had paid for and lost.

So I stayed up. I would be awake when the limit reset so I could immediately queue up work. I would plan my sessions around the window instead of around my life. I was building at 2 AM because the meter had just refilled, with nothing urgent to build.

The sleepless nights I used to spend in flow, absorbed in a problem, were now sleepless nights spent babysitting a quota. That is not the same thing at all, even if it looks the same from the outside.

My sleep schedule collapsed. My mental health followed it, because it always does. I was tired during the day, and I was going to bed late enough that it started to weigh on my relationship. And on top of that, I would get frustrated every time a model got degraded (hello, Opus 5).

Bored While the Machine Works

There is a strange side effect of agentic coding that nobody warned me about: it is boring.

When the AI is writing code and running tools, you are waiting. For minutes at a time. And what do you do when you wait? You open another tab and put on a YouTube video "while it runs". Then it finishes and you type the next prompt without closing the video.

My screen time grew a lot during this period. I was consuming more, in the gaps.

The flow state I described at the beginning, the one where nothing else exists, not even you, never came back in that form. What I got instead was a lighter version, an attentive-manager kind of focus, constantly interrupted by the little dopamine snacks I was using to fill the waiting. I was more productive on paper and less present in my own work.

The Involuntary Break

I did not decide to stop. I had to.

My mental health had been bad for a few months, and not only because of AI. Life was heavy in other ways, and the sleep debt and the constant low-grade stimulation made everything worse. At some point I could not keep going, and for about two weeks I did not touch an AI tool or write a single line of code.

The surprising part is how quickly it felt relaxing. After a few days the noise went down. No token meter, no half-started projects calling me. Just time.

I do not think you need two weeks. I think one week is enough to remember what a normal relationship with your computer feels like. But I needed to have the break taken from me before I understood I needed it.

What I'm Doing Differently

I am not quitting AI development. I will get to why in a second. But I came back with rules, because I now know what happens without them.

One project until it ships. I pick one thing and I stay on it until it is published, and "mostly done" does not count. New ideas go in a note. This is the single most effective constraint I have found, because it puts the friction back where the AI removed it.

Take real breaks. A week away from the tools, on purpose, before the break is forced on me. I now know what the warning signs look like, starting with planning my evening around a reset.

No AI Fridays. I heard about No AI Fridays from the CEO of htmx, and I am going to try it. One day a week, you turn off your AI assistants and write the code with your own hands. The site puts it better than I can:

If the productivity gains from AI are so big, spending one day a week to minimize its downsides shouldn't be a difficult trade-off.

It also frames that day as a way to review the choices the AI made for you during the week, which is exactly what I stopped doing.

Stop optimizing the meter. I am done being awake for a quota.

I Still Love AI Dev

This article is not a warning against AI, and I would be a hypocrite if it were. I build with Claude Code every week, and I teach people how to build MVPs this way.

I really like AI development because it gives me speed, and I still control the architecture of the software I build. I design the systems. The AI fills them in. It lets me push harder into topics where I am not an expert, and ship things that would have stayed in my notes forever.

And I know I did not lose my skills, because in my day job I still code by hand every day. The muscle is still there. I just needed to stop letting the tool decide when I sleep.

AI is wonderful. AI is stressful. AI is powerful. AI is exhausting. All four are true at once, and the job now is to keep the first and the third without paying for them with the second and the fourth.

If you recognized yourself somewhere in this article, take the week. You will not fall behind.

---

Join the Community

If you're building a SaaS or just getting started with AI-powered development, come hang out in our Discord community. We share projects, discuss tools, and help each other ship faster.

Follow me on Instagram for behind-the-scenes content and build-in-public updates.