Things are made by teams

“How do we get people to meaningfully coordinate in service of doing stuff that matters?” This is one of those big, civilization-spanning sort of questions. There are many answers, there’s no one right way (but there are many wrong ones, as discovered thru evidence of trying them) - and while there’s generic motivators for human behavior, there’s a sort of loss and amplification factor that happens when you mismatch the reward function. To be more blunt - hunger is a powerful motivator, but hungry people tend to not perform at their peak.

Enter: AI. You can now summon minds via API, channelling tens of thousands of digital workers to build a glorious edifice to whatever you’d like. Every time there’s a new model release, Twitter and Bluesky are filled with “look at this!” style posts that show off some technically impressive output - working circuits, CAD designs, music videos, or fully formed games, all from a prompt. The more dedicated amongst us create edifices that continue to build themselves, creating an ouroboros of work - a factory that grows to meet the needs of the expanding factory. In this way, AI acts as both a force multiplier of an individual's will as well as a friction reducer on the collaboration that would normally be needed to build large scale projects. It is the it can only good happen machine, because it does not know any other way of being.

Cynics will point out, correctly, that many of these grand projects have not had much impact. Gas Town is an instructive example — it consumed itself, creating structure and scripture for an ever growing chorus of characters. It is not the only example of this; the internet is now littered with such artifacts. A single idea is echoed across the network, remixed and copied. Remember the fun “pick a player to make a NBA team from” game? What about the NFL version? The college football version? The pro wrestling version? The F1 version? Or perhaps you don’t remember any of them, because it came and went in a few weeks and now we’re all onto the next thing? This isn't really an indictment against making things, people should make all sorts of things! The cost of creation is effectively zero.

I think we've hit this weird maladaptive loop when it comes to these sort of edifice projects. There's a reaction based on style more than substance, because we're kinda wired for effort being a proxy for 'worthy of attention'. Someone hands you a wooden sculpture and says "I bought this from Amazon" vs. "I carved this myself", even at the same quality, you're probably gonna think the hand-made one is more valuable. Unless you're a supply chain sicko. Then you're gonna think about how much it costs to ship a widget around the world. This, too, is the power of people meaningfully coordinating to do big things. It's easy to discount agentic work because it's synthetic, because it's inefficient, because of a million other reasons -- but it's interesting how quickly we do so. There are people building their dream apps every day, and you've probably never heard of most of them. The amount of software being created is going through the roof. Like, this is the first time in my adult life that I can recall a generalized compute shortage.

I’m not going to write a paean to “taste”, but I think it’s worth pointing out that a shortage exists because we don’t have a great way to curate ideas and effort when both of those are available via an API. Shortages cut both ways - they drive up the cost of execution regardless of whether you’re building something like polytoken.dev or Gas Town, and that’s [good/bad]. I’m not using this blog to judge. I think wealthy people have always been able to create monuments to hubris, AI has simply made that more accessible. The difference between making a good thing and making a bad thing, imo, is the why and not the how. Historically, open source and free software have held the vanguard of building things with the intent to create a community of shared benefit (in the sense that code is knowledge, and knowledge should be shared freely). I think AI should force us to grapple with what this actually means, though.

A lot of modern open source is more about creating standards than anything else. Kubernetes, OpenTelemetry, Rust, Go, whatever. Home Assistant. Linux. ATProto! Some of these are actual bona fide standards projects with specs to match, others are creating de facto specs through competition with proprietary alternatives. The thing they offer isn’t just software, though. It’s a shared lexicon, a way of describing solutions and defining problems in order to build a shared reality and acting as a flag for humans to rally around. The interesting counterpoint to these are the communities I’ve seen pop up around AI projects, especially buzzy ones - OpenClaw, Gas Town, etc. These seem to devolve into angry hives of robots buzzing around building work-shaped output in response to the individualized demands or wishes of operators. AI allows infinite customization, and the norms of the community seem to drift towards embracing this individualism in a somewhat nihilistic pursuit of increasingly fine-grained requests, ledgers, authorities, and evidence. Sound and fury, as it were.

I don’t think one or both of these are inherently bad, although I do tend to think that the latter is unproductive. I also find it unproductive to hide in a corner, throw up Anubis, and tell users to fuck off. Ejecting yourself from a technology unilaterally is not a principled moral stance. It is cowardice at best, a refusal to adapt to a changing world. At worst, well, I’ll point to the witch hunt that took place this week re: Big Walk and “undisclosed AI usage”. We are entering a world where much of the work that we’ve strived for - the ability to build shared understanding and solve problems thru computation - is becoming a truly trivial detail. Implementation is no longer hard, the bottlenecks accrue around verification and validation. The code was never really that important except as a way to transmit these lessons to the future; it is now even less important than it was before.

With this said, I do wonder what the role of open source should be going forward. Ideas - good ones - are at a premium. Code sharing is only really important in the sense that it saves time and gives you a tangible reference to an idea. What matters is what’s always mattered - how do we do the work of coordinating people towards big, audacious goals? How do we learn, how do we share, and how do we even find our people and problems in an increasingly individualized information sphere?

I think there’s a few things that I’m exploring at least:

1. Community of practice. Collaboration gets results, it’s something that we see over and over as humans. We don’t form companies or teams or whatever just for the pleasure of herding cats, we do it because we lift each other up and multiply our efforts. I think having that sort of community is going to be an increasingly valuable part of OSS as a way to bring in more ideas, bring in new people, etc.
2. Removing code from the equation and providing conformance. Less “here’s an sdk” and more “here’s an exhaustive test suite”. Formalizing specifications requires a lot of work and alignment. Traditionally that alignment is expressed in code, but we can/should think about how we can use formal verification DSLs and build tools that make verification easy.
3. Discovering necessary friction. Code understanding is friction; but we are at a point where it is mostly unnecessary in a lot of ways. Consider a PDF - the rendered output is what you care about broadly. The details of what’s in the file, less so. I can create artifacts and outputs that are slop on the inside but it’s disposable, one time use stuff. But I want to be able to find patterns and improve the process, and the process uses code to codify things. So how do we keep systems understanding without the friction of code understanding? How do we transmit that understanding? Is the AI a universal translator? Idk. I think people hold values and stakes here, so understanding is crucial.
4. We need to raise the bar and curate more. This is a discovery problem. Information organization and accessibility are crucial.

I don’t really have answers here, every answer leads to more questions. But I am increasingly convinced that individual polarization on this topic is not only leading to a lot of self imposed ivory tower thinking, it’s also leading to a lot of wishcasting. We’re clapping for tinkerbell in our own rooms.