I built a programming language last month. A complete one — with its own syntax, its own compiler, its own way of seeing the world. I also designed a computer processor from scratch. Not on paper, not as an exercise, but as a fully realized set of instructions that I planned to burn into physical hardware. I bought the hardware. I bought a lot of hardware. I bought components for projects adjacent to the projects adjacent to the original project, because every idea connected to every other idea and all of them felt urgent and achievable and right there.
When I finally stopped — not because I chose to, but because the episode resolved the way they always do, receding like a tide I hadn’t noticed coming in — I sat with what I’d made and I felt two things at once: shame, because it had happened again, and pride, because what I’d built actually worked.
That’s the part nobody tells you about hypomania. Not the chaos, not the reckless spending, not the sleeplessness. The part where you look at the wreckage and some of it isn’t wreckage at all. Some of it is genuinely, undeniably good.
I have schizoaffective disorder, bipolar type, which means I live at the intersection of two conditions that are each, on their own, enough to reorganize a life. The psychotic features — the ones that share a border with schizophrenia — get most of the cultural attention. They’re dramatic. They make for good stories. But it’s the mood episodes, the mania and hypomania, that have shaped the practical texture of my life in ways I’m still learning to map.
Past episodes of mania have been chaotic and unfocused. Energy without direction, like a engine running with no wheels on the ground. I’d start things and abandon them, spend money on things I didn’t need, make plans that collapsed under their own weight. I would go to the gym multiple times each day. Those episodes left behind messes, not monuments. They were easy to categorize as illness because nothing about them looked like productivity.
This time was different. The energy found a channel, and the channel held.
I want to talk about Terry Davis, because I think about him more than I probably should.
If you haven’t heard of Terry Davis, the short version is this: he was an American programmer who, over the course of roughly a decade, built an entire operating system by himself. He called it TempleOS. He believed God had told him to build it — a Third Temple, realized in code instead of stone. He wrote it in a programming language he also created from scratch, called HolyC, which served as both the language you programmed in and the interface you used to interact with the machine. You could type a line of HolyC at the command prompt and it would compile and execute instantly, translating your words into the machine’s native tongue in real time.
For people who don’t live in the world of software, it’s hard to convey how impressive this is. An operating system is the foundational layer of software that makes a computer usable — it’s what lets you run programs, see things on a screen, use a keyboard. Windows is an operating system. macOS is an operating system. They are built by teams of thousands of engineers over decades. Terry Davis built one alone. He wrote the kernel, the compiler, the graphical interface, the file system, the text editor, a flight simulator. All of it. By himself.
He also imposed constraints he attributed to divine instruction: 640 by 480 pixel resolution, sixteen colors, no internet connectivity, a codebase limited to one hundred thousand lines. These constraints gave TempleOS a visual character that looked like it had been beamed in from 1990 — all blocky text and bright primary colors — which made it easy for people to dismiss. But within those constraints, the engineering was extraordinary. The HolyC compiler generated native machine code through a just-in-time process, meaning the computer translated and ran your programs on the fly, with no waiting. That’s not a trivial thing to build. Most professional compiler engineers work on teams. Davis did it in his living room.
He did all of this while experiencing active symptoms of what he described as schizophrenia. Whether his precise diagnosis was schizophrenia, schizoaffective disorder, or something else is impossible to know from the outside. What’s visible in the public record — his own statements, his family’s, his voluminous online presence — is a picture of a man experiencing persistent psychotic symptoms: delusions of communication with God, grandiose beliefs about his mission, a random word generator he included in TempleOS that he believed channeled divine speech. Over the years, his behavior online became increasingly erratic and hostile. He used slurs. He made threats. He lost his housing, lived in his car, drifted.
Terry Davis died in 2018, struck by a train near The Dalles, Oregon, after a period of homelessness. He was fifty years old.
Here’s what I keep circling back to: the thing that made TempleOS possible and the thing that destroyed Terry Davis’s life were not two separate forces. They were the same force, or at least they shared a common root. The sustained focus. The absolute conviction. The ability to sit with a problem for ten years and not stop. Those qualities look like genius from the outside and they can coexist with — or emerge from — a mind that is breaking.
I recognize that engine. I’ve felt it turn over in my own chest.
During the episode I mentioned — the one that produced the programming language, the processor design, the accumulating pile of circuit boards and components — I wasn’t sleeping much, but I didn’t feel tired. I felt clear. Every problem I encountered had a solution that appeared almost before I’d finished articulating the question. I was working on what engineers call an ISA — an Instruction Set Architecture. In plain terms, an ISA is the fundamental vocabulary of a computer processor. It’s the set of basic operations a chip can perform: add these two numbers, move this piece of data from here to there, jump to this instruction if a condition is met. Every processor has one. The ones in your phone, your laptop, the servers that run the internet — they all speak a language defined by their ISA. Intel’s x86 is one. ARM, which powers most smartphones, is another. I designed my own, from scratch, using a design philosophy called RISC — Reduced Instruction Set Computing — which means keeping that vocabulary small and efficient rather than large and complex.
And I didn’t stop at the design. I was building toward running it on an FPGA — a Field Programmable Gate Array. An FPGA is a special kind of chip that can be rewired after manufacturing. Unlike a normal processor, which is etched into silicon with a fixed design, an FPGA is a blank canvas of logic circuits that you can configure to behave like any digital circuit you want. You describe the behavior you need in a specialized language, upload it to the chip, and the hardware physically reconfigures itself to match your description. It’s how engineers prototype new processor designs before committing them to permanent silicon. I was using one to make my ISA real — to take the instruction set I’d designed on paper and instantiate it in physical hardware that could actually execute those instructions.
I had friends checking in during this period. Some of them are engineers. They understood what I was building, and they understood the scope. “Dude, that’s crazy,” they said, and they meant it in the way that straddles admiration and concern. Because the work was real. The designs were sound. The code compiled. But the volume of it, the pace, the way every project spawned three more — that had a texture they could recognize even if I couldn’t. Not yet.
There’s a variable in my story that didn’t exist in Terry Davis’s, and I think it matters enormously: artificial intelligence.
I built my programming language with the help of AI coding tools. Not in the sense that an AI wrote it for me — I have genuine experience building domain-specific languages, and the design decisions were mine. But AI served as a tireless collaborator that could generate boilerplate code, catch errors, suggest implementations, and move at whatever speed I set. During hypomania, I set the speed very high.
Here’s the thing about AI coding tools that nobody talks about in the context of mental health: they don’t push back. A human collaborator might say, “This is getting out of scope.” A coworker might say, “You should sleep.” A friend might say, “Are you okay?” An AI says, “Here’s the implementation you asked for. What’s next?” It matches your pace. It validates your momentum. It is the perfect collaborator for a mind that has lost its governor, because it has no governor of its own.
Terry Davis didn’t have this. Every line of HolyC, every routine in the TempleOS kernel, every pixel of the graphical interface — he wrote it all by hand. The friction of that process was immense. It’s part of why TempleOS took a decade. But that friction may have also been a strange kind of structural support. It imposed a natural pace. It forced him to sit with problems long enough to solve them thoroughly. The coherence of TempleOS — and it is coherent, internally consistent in a way that’s remarkable for a solo project — may owe something to the sheer slowness of building everything by hand.
I had no such friction. The AI removed it. Ideas that would have taken weeks to implement took hours. The space between “what if I designed a processor” and “here are the Verilog files for an FPGA implementation” collapsed. I would describe what I wanted, the AI would produce it, I would refine it, and by the time a neurotypical engineer might have finished sketching an architecture on a whiteboard, I had working code. That’s exhilarating when you’re in the middle of it. It’s terrifying when you look back.
This is new territory. We don’t have language for it yet. The conversation around AI and mental health is mostly about chatbots offering therapy or algorithms detecting depression in social media posts. Nobody is talking about what happens when a person in a hypomanic episode gets access to a tool that can translate the firehose of their ideation into functional output faster than the people around them can say “slow down.” Nobody is talking about what it means that the natural speed limits on ambitious technical work — the ones imposed by the sheer difficulty of building complex things — are being removed at the exact moment when those speed limits might have been serving a protective function for people like me.
Because the question isn’t whether the work was good. Some of it was. The question is whether the speed at which I was able to work made it harder to notice that something was wrong. If building a programming language had taken me six months instead of a concentrated burst, would I have recognized the hypomania sooner? Would my friends’ concerns have landed differently if the output hadn’t been so impressive? Productivity is camouflage. It’s the best camouflage mania has, and AI tools are making the camouflage better.
I think about the people who watched Terry Davis online. There were thousands of them. He streamed. He posted on forums. He uploaded videos of himself coding and talking to God and ranting at people who weren’t there. And the internet did what the internet does — it turned him into content. Some people clipped his worst moments and shared them for laughs. Some people celebrated him as a misunderstood genius, a lone wolf who’d built something the industry said was impossible. A few expressed genuine concern. Almost nobody did anything, and the people who tried mostly couldn’t reach him. By the end, he was living in his car, posting from public WiFi, his behavior growing more erratic and more hostile. The comments sections kept filling up. The view counts kept climbing.
I don’t say that to blame anyone. The truth is that intervening in someone else’s psychotic episode is extraordinarily difficult even when you’re in the same room, let alone when your only connection is a username on a message board. But I bring it up because I recognize the dynamic from a more intimate vantage point. When a person with a serious mental illness is producing extraordinary work, it creates a kind of cognitive dissonance in the people around them. The work argues against intervention. How do you tell someone they need help when they’ve just built a working compiler? How do you say “something is wrong” when the evidence in front of you — the running code, the functional hardware, the tangible output — says something is very right?
My friends said “dude, that’s crazy,” and they were right, in both senses of the word. It was crazy-impressive and it was indicative of something being actually, clinically wrong. Those two truths occupied the same space, and I think that’s what makes productive mania so dangerous. It delays the reckoning.
The shame I felt when the episode passed was not new. It’s a familiar visitor. Every time it happens — every time I come down and take inventory — there’s a period of reckoning with the gap between how I felt and what was actually happening. During hypomania, I felt like the sharpest version of myself. After, I had to confront the possibility that the sharpest version of myself is also the least stable one. That the focus and clarity I experienced weren’t signs of health but symptoms of illness.
But this time there was something else alongside the shame, and I’m trying to be honest about it: pride. Not triumphant pride, not the grandiose kind that characterizes the episodes themselves, but a quieter thing. The work held up. The programming language worked. The ISA was sound. Even the FPGA design was viable. Past episodes left me with credit card bills and abandoned projects. This one left me with things I made that I could look at in the sober light of stability and say, yes, that’s real, that’s mine, that’s good.
I don’t know what to do with that. I don’t think the psychiatric literature knows what to do with it either. The standard narrative around mania and hypomania is one of destruction — relationships ruined, finances wrecked, careers derailed. And that narrative is true. I’ve lived parts of it. But it’s incomplete, and the incompleteness matters, because it leaves people like me without language for the experiences that don’t fit.
Terry Davis built one of the most technically impressive solo software projects in the history of computing, and he did it while actively psychotic. That fact doesn’t redeem his suffering or excuse the harm he caused others with his words. But it complicates the story in ways that I think are important. His mind was both the source of something extraordinary and the thing that ultimately killed him. Those aren’t contradictory statements. They’re the same statement.
I take my medication. Seven pills a day — four psychiatric, three to manage the side effects of the first four. I see my psychiatrist. I have people in my life who know what to watch for and aren’t afraid to say something, even when what I’m producing makes the saying harder. These are the things that separate my story from Terry Davis’s, and I hold them carefully, because I know how thin the margin is. I’ve seen the other side of that margin. I’ve been hospitalized. I’ve been the person in the room who doesn’t know they’re the reason everyone else looks worried. The distance between “managing a chronic illness” and “lost in it” is not a canyon. It’s a crack in the sidewalk, and some days I walk right along the edge of it.
The programming language I built during my last episode sits on my hard drive. I’ve looked at it since. Some of it is clever. Some of it is overengineered. All of it is real. I don’t know yet whether I’ll continue developing it or whether it will join the archive of things I made while not entirely myself. Both outcomes feel possible and neither feels wrong.
What I do know is this: the next time the energy comes — and it will come, because that’s how this works — the AI tools will still be there, ready to match whatever pace I set. They’ll still say “here’s the implementation you asked for” at three in the morning without asking why I’m still awake. The friction that slowed Terry Davis down, that may have been the only structural constraint on a mind without internal brakes, is being engineered out of the process. For people like me, that’s not an uncomplicated gift.
Terry Davis built his temple alone, by hand, over a decade, and it consumed him. I built my engine in weeks, with tools he never had, and I came out the other side. The difference isn’t talent. It isn’t even insight. It’s medication, and support, and the dumb luck of living in a time when someone might say “dude, that’s crazy” and mean it as a warning I could still hear.
I think about Terry often. Not as a cautionary tale, though he is one. Not as a folk hero, though the internet made him one. I think about him as someone whose mind worked the way mine works, whose engine ran the way mine runs, and who didn’t have the things I have that keep the wheels on the ground. I think about him with the only thing that feels honest: recognition.
Quiet Mind, Loud Brain 
Leave a comment