You work as an engineer in an org and think the code is 20% of the job, so AI automating that part leaves you safe. But that line of reasoning has one consequence: it means you were already a bureaucrat. The industrialization hollowed out the craft long before the models arrived. And this is exactly why you are not safe.
The code and understanding the code was your real moat . You had the craft and talked yourself out of it. You started running processes and ceremonies, writing docs about systems you no longer built, telling yourself you were above the code. And honestly, you were not sad about it. The code was always the part you wanted to escape.
Everyone is quietly hoping AI plateaus right before it reaches their specific skill level. Find yourself on the spectrum below and use it as a mirror. We are on the brink of a paradigm shift and nobody knows what comes next. The more honest question is what you want to be doing when the music stops.
The terms come from an internet meme circa 2021, loosely rooted in cognitive research: verbal reasoning versus spatial rotation as distinct cognitive modes. Slightly edgy framing, but the underlying distinction is real and useful here.
Shape rotators think in systems, code, spatial structures. Wordcels think in narrative, language, persuasion. The original thesis: tech was run by wordcels at the top and shape rotators doing the actual work below.
AI flipped it. AI is the ultimate shape rotator. It manipulates code and data structures faster than any human. Pure shape rotation is getting commoditized. The wordcel input layer is now more valuable because articulating what to build clearly is what you feed the machine.
The question that comes up online: is it easier for a shape rotator to become a wordcel, or for a wordcel to become a shape rotator? And does it even matter, or are we heading toward a market where one is simply worth more than the other?
Thought experiment. Take a great designer or a PM as your wordcel. Take a nerdy software developer as your shape rotator. Which transition is easier? The developer becoming the designer, or the designer becoming the developer?
My hunch: in a world where AI is the ultimate shape rotator, the designer and the PM carry something more valuable and harder to transfer. Taste, people sense, years of accumulated instinct about what feels right before you can say why. That does not come from a course. It compounds slowly and it does not move easily between people.
You might say: Davit, but good software engineers were never just shape rotators. They also knew what to build, why to build it, they understood users too. On this page I argue that is the cope. The people who genuinely honed wordcel skills for years are coming for the same seats. And they are arriving with better tools than the engineers ever had.
I built a radio with Claude Code and only later realized how elegant the math behind it was. Radiobananas.art is a music radio handpicked from my own taste. Everyone hears the same music at the same time without any backend sync or websocket. Just a static HTML page with files sitting on Cloudflare R2.
So how do my phone in Amsterdam and your laptop in Yerevan end up on the exact same second of the same song. It is just: (UTC seconds since midnight) % total playlist duration. Everyone runs the same calculation and gets the same answer.
The shuffle is what makes it tricky. If it is random, two people loading the site at different moments get different orders and the whole thing breaks. So it has to be a deterministic shuffle, same input, same output, everywhere. I use the date as the seed so everyone gets the same shuffle for the day.
The shuffle itself is the Fisher-Yates shuffle, which guarantees every track plays exactly once before the cycle repeats. There is also a tiny three-line PRNG in the code, a variant of xorshift from the 1990s. Without it, Math.random breaks the sync.
Here is the part that matters for this page. If you look at what is actually in that codebase: deterministic shuffles, a PRNG variant, modular arithmetic for sync. None of these are trivial. A year ago I would have been the gatekeeper for this. You needed to understand xorshift to produce xorshift. Now I curated it, understood it after the fact, and shipped. A good designer with a flexible brain and a good model could build a better version of this than I did. Not a complaint. The elephant in the room this entire page is about.
Seven archetypes on a spectrum. The left end is the strongest moat: deep tech skills, intuition, taste, institutional knowledge, network leverage. The right end is the most exposed: pure coding or lego-design skills, detached from the realities of the markets and the businesses they operate in.
Intelligence is executing complex rules. Judgement is knowing which rules apply, when to break them, and what to build next. AI has crossed the threshold where it can do most intelligence work autonomously. Where thinking and building are one, the work is mostly judgement. Where they are separated by layers of process, it is mostly intelligence. That is why it matters where you land.
Drag to find where you are.