dl-4-on-skating

the largest open-source community

What is the largest, most influential open source community? No, it's not rust you silly sausage. I think (and I have done the maths!) it's skateboarding. There are way more skateboarders than people who program, and orders of magnitude more than people who are actually any good (author excluded from these statistics). They have way more tricks than there are design patterns. An ollie engages way more complex a biochemical machinery than node.js allows actual machinery without instantly crashing or failing to build the next day.

Skateboarders share their "code" (tricks) openly through videos, others fork and modify those tricks adding their own variations, there's constant iteration and improvement on existing techniques, knowledge spreads freely without proprietary restrictions, people build directly on top of what others have done, and the community collectively advances the "platform" (culture as a whole and skateboarding in particular).

Rodney Mullen says, skateboarders "form an open-source community in the sense that the greater contribution we make through the tricks that we create, the more we embed what we do into the language, the ways other people project themselves, that weaves us into the fabric as a whole." The whole community shares these common groups of tricks, but each skater tries to distinguish themselves in the individual way they do them. It's a culture where "together we rise" - the way skaters make a name is far from a zero-sum game.

digital depression

Skateboarding has had massive cultural influence - it basically shaped entire genres of music (punk, hip-hop connections), fashion, art, filmmaking, and general attitudes about creativity and DIY culture. Think about the work of Glen E. Friedman, who shot both the best skaters and the most iconic album covers in punk rock and hip hop - he was the first to really clock the connection between these seemingly different cultures and their mutual respect for each other.

Skateboarding consistently produces something that's genuinely positive - it gets people active, creative, builds community, encourages persistence and problem-solving. When someone lands a trick they've been working on for months, that's pure joy. When kids see a skate video and get inspired to go outside and try something new, that's culturally valuable in a really direct way.

Compare that to software: node may run javascript on the server but, like, why? By the time I finish typing this sentence your node codebase has become unbuildable because someone upgraded a library in your node_modules. Besides that, things we put on the Internet tend to make you addicted or depressed or both at least (and also at most) as often as they turn out useful.

Thousands of people work 9-to-5 building social media platforms optimised for constant engagement, apps designed to be addictive, systems that increase anxiety or isolation. The cloud is becoming a goon cave because that is what keeps the money printer going brrr. The "open source" aspect amplifies the good stuff in skateboarding - shared knowledge leads to more creativity, more people having fun, more cultural innovation. With software, the work is at a constant tension with the need of maintainers to put food on the table.

age of abundance and the race to the bottom

Our hardware is absolutely insane now. An M4 or a Zen 5 will run basically any reasonable code instantly. Leave it to the programmers of the present to make software slower than it was 20 years ago. We're choking these beastly machines with layers upon layers of crap - interpreters sitting on libraries sitting on containers sitting on frameworks sitting in cafes drinking overpriced flat whites. It's like you need to fry an egg so you set your whole house on fire.

And this is where the skateboarding analogy breaks down. Adam Wathan (founder of Tailwind CSS) talked about their struggle to stay profitable because AI labs happened to see every time tailwind was used anywhere ever and have come to enjoy spitting out Tailwind's intellectual property at 3 millionths of a dollar per token. Something like Vercel's v0 spits out Tailwind components that are noticeably worse than what the paid Tailwind resources provide. Worse or not, it is enough to pass the "eh, ok" bar and be adopted into apps driving a regression to the same shit mean via the next training cycle.

now make me a notebook

Anyway, this is why I'm so (sooooo) happy marimo is open source. At Querio we bet that a BI tool needs to stand in between the data team and the business users - they need a common environment to collaborate that adapts itself to each user-base's needs and capabilities.

A Jupyter notebook is an expressive programming environment and it is pretty iconic, but it has issues. Marimo added reactivity to the execution model and interactivity to the interface through a widget system. A notebook that is reactive and has widgets is pretty much everything you need to build data-apps that are ultimately the common unit of output for any respectful BI software.

So 2 weeks ago we set out to build a reactive kernel with widgets. Hands shaking, I git clone the Marimo repository, teeth chattering: "Will I understand anything?". I stepped through the code for a few hours, maxed out my Claude and Gemini limits to help me map things out. In the end it wasn't that hard (the code is good). I opted out of AI'ing even parts of the thing for the same reason v0 is bad for Tailwind - the code is just not that good man! Marimo's system and (I suspect) any useful and novel software just takes a special instinct to produce. The kind of instinct that has not yet emerged during sampling a probability distribution.

Either way, for the first time in a while I felt like I was learning new things! I learned about Protocols, compile, a lot about the import system (running code is not actually that hard...), WebComponents, and just generally how to lay out a big project in Python.

So thanks to Marimo, we managed to ship a reactive kernel in 2 weeks! I can't believe I am typing this. I won't let anyone actually use it because it swallows its own errors, handles 0 edge cases, does not support SQL, no syntax highlighting, no console output... you get it: you can't use it. Soon though...

good artists borrow, great artists steal

Emboldened by my positive experience studying Marimo I felt confident to git clone the Telegram repo, that I use every day to exchange picutres of my cat with my friends and family, to find ways to improve Querio's AI interface aka textarea. I will freely admit that I built the first version with the help of none other than Claude who first correctly suggested I use a contenteditable div, then failed to render pills (or badges or lazagnes or tags) inside it, so instead we are currently using a hidden input field that mirrors its contents into the div, where it conditionally styles thing. Yes I know. I am ashamed.

Pretty soon after cloning the repo I discovered that you can render HTML inside contenteditable - any CSS I want (pill included) becomes trivial. The repo used signals to avoid re-renders, understood a pill to be a distinct object, used a reducer pattern instead of context to manage data, a lot of useRef and useEffect mods to keep stable references. It took me 3 days to port what I have learned. The result was more performant (i am sure I counted the re-render flashes right), and most importantly it sparked joy which Claude, for all of its usefulness, somewhat diminished after coming on the scene.

conclusion

Skating through 2 large codebases in completely different domains made me realize how much knowledge is sitting in there, waiting for me. I thought really really great Engineers were made one of two ways. They either were already programming game engines while I, still wearing braces, tried to get served at the pub. Or, they joined Meta where a 10000x ex-Netflix ex-Twitter ex-Facebook Mega-Engineer called Googlealf The White mentored them to be a code wizard. But I was wrong! All the pieces are out there, just gotta assemble my puzzle. I guess being really really great isn't something that happens TO you. I can imagine this reads very dumb to someone who doesen't posess such dramatically mismatched quantities of skill and ambition but hey! no! you're dumb! jk, ily.

Anyway. Next time you are reaching for your favourite VSCode fork, maybe...don't. Maybe write it yourself, or study the old masters. Read the docs. Remember why you're into this programming thing in the first place. Look around, the foundation is wobbling. The boat may be sinking. Get lost in a big codebase. Find your way. Read a paper, then read another. Write a blog post. Buy Querio for your company. There are no adults in the room. Most code is terrible. Complain. Stop gooning. Build your own house. Open source it. Dig deeper and don't settle for shadow divs for your textareas.

By the way, I don't skate and never have. But one night at university, before I held programming opinions, I was out with this girl and I wanted to impress her by hopping over a bollard which happened to be less structurally sound than I expected it to be. This misapprehension caused me to eat asphalt and lose half of my front tooth. Look at my LinkedIn picture and guess which half is dentist glue shaped like a front tooth. First person to guess correctly gets a special Querio surprise! I'll be waiting.