A strange fact: a thing can be better in the ways that matter most, and still be terrible at becoming the default.
- Permaculture is better for soil, water, biodiversity, and long-term food security. But we practice monoculture on 80% of the world’s arable land.
- Walkable neighborhoods are healthier, cheaper, and overwhelmingly preferred; yet most cities can’t convert because the infrastructure is already built wrong.
- Open protocols gave us email and the web; closed platforms captured both.
Why? And what would it actually take for the better version to win?
[draft]
Why beautiful things lose
It’s tempting to assume that if something is genuinely better, it should simply win. But that assumes a shared game with a shared objective. In real systems, there usually isn’t one. Different people and institutions are pursuing different goals under different constraints and timescales.
Ecologists have a frame for what happens when different strategies compete in the same environment:
- r-selected species — bacteria, dandelions, weeds — reproduce fast, colonize aggressively, and die easily.
- K-selected species — elephants, old-growth trees, whales — grow slowly, invest heavily, and are resilient but terrible at colonizing.
After Mount St. Helens erupted in 1980, the first colonizers were wind-dispersed lupines and spiders ballooning on silk threads. Within 5 years, grasses. Within 20, shrubs. The full forest recovery will take 200+ years.
Weeds are not failed trees; trees are not successful weeds. They solve different problems at different stages.
The ugly-but-scalable thing is solving the colonization problem. The beautiful thing solves the maturity problem. The tragedy is when the colonizer prevents the mature system from ever arriving.
This plays out in many ways — sometimes through distribution lock-in, sometimes through institutional failure, sometimes through deliberate tradeoff:
-
Ted Nelson’s Xanadu had two-way links, built-in attribution, version history, and micropayments — in 1960. The Web shipped in 1991 with one-way links that break and no rights management. Nelson:
“HTML is precisely what we were trying to PREVENT.”
But Berners-Lee had already built two-way links in his earlier system ENQUIRE (1980) — he knew exactly what he was giving up. He later called one-way links:
“the principle compromise made in the W3 architecture, which then, by allowing references to be made without consultation with the destination, allowed the scalability which the later growth of the web exploited.”
Nelson met him around 1989 and said:
“he’d done this very simple thing, and it sounded too trivial to me… and the next thing I knew suddenly the thing had caught on.”
-
JavaScript was created in 10 days in 1995, with type coercion bugs that can never be fixed (
typeof null === "object"is a day-one error). It won because it shipped in Netscape Navigator at 80% market share — the only scripting language in the browser. -
Python has the GIL, a 12-year migration from 2 to 3, and is 10-100x slower than compiled languages. It won because universities made it the default teaching language, which fed a generation of developers into the data science ecosystem (NumPy, pandas, PyTorch — all C/Fortran under the hood).
-
Better-designed alternatives to both exist. None of them matter. (mostly).
-
Xerox PARC built the GUI, the mouse, Smalltalk, ethernet, and laser printing by 1973. Xerox — one of the largest corporations in the world — couldn’t scale any of it. They shipped the Star at 2,495. A company a fraction of Xerox’s size took PARC’s beautiful ideas and actually got them into millions of people’s hands. Having the beautiful thing and having a scaling strategy for it are completely different capabilities.
-
Before Stripe, accepting a credit card online meant weeks of merchant accounts, payment gateways, PCI compliance, and banking relationships. Stripe reduced it to seven lines of code. The complexity didn’t disappear — Stripe absorbed it on behalf of developers, handling the banks, the fraud detection, the regulatory maze, so that the interface stayed simple. Patrick Collison:
“More of Stripe’s success than one would think is downstream of the fact that people like beautiful things because a beautiful thing tells you the person who made it really cared.”
-
Bret Victor’s Dynamicland is the most compelling vision of what computing could be — spatial, communal, physical, humane. It has been one building in Oakland for years.
-
OpenAI was founded in 2015 as a nonprofit with fiduciary duty to humanity. By 2025 it had restructured into a $300 billion for-profit corporation, because the nonprofit form could not generate the capital needed to compete. The mission statement has changed six times.
The common thread: whatever wins is better fitted to the selection pressures of the current environment, whether that means being simpler to deploy, cheaper to adopt, or easier to finance.
Succession, not teleportation
So the beautiful thing and the ugly thing are adapted to different environments. But even when the beautiful thing is viable, it often can’t appear in its mature form on day one — it has to be grown through stages.
You cannot deploy a forest.
First come the pioneer species. Then grasses, shrubs, nurse trees, fungal networks, and only much later the dense interdependent thing people actually want when they say “forest”.
A lot of beautiful systems seem to fail because they are being asked to appear in their mature form, all at once, inside an environment that cannot yet support them.
Walkable cities are a good example. Everyone loves them after they already exist, but they are not just an architectural style — they depend on land use, finance, transit, political coalitions, zoning, norms, and path dependence. Amsterdam was car-dominated until 3,300 people died in traffic in 1971 (450 of them children) and parents blockaded streets with overturned car wrecks. Most cities can’t convert because the infrastructure is already built wrong, and retrofitting costs orders of magnitude more than building correctly the first time.
The same is true of open protocols and permaculture — each depends on a dense web of preconditions (clients, norms, adoption thresholds, local knowledge, long feedback loops) that can only accumulate over time.
Many beautiful things are climax communities — and climax communities, by definition, can’t be exported.
Chinese mapping apps are a striking example. They can show you:
- live red-light countdowns
- real-time bus tracking
- shade-percentage routing for pedestrians
- shortcuts through parking garages
They are so far ahead of Google Maps that using them feels like time travel. But they can’t expand outside China, because their quality comes from an ecosystem — 19 competing domestic survey companies, a massive user base generating real-time data, and the Beidou positioning system which is more precise than GPS. The app is a product; the quality is an ecology. You can export the app, but the ecology has to grow where it is. (James C. Scott’s Seeing Like a State is relevant here — standardization strips away exactly the kind of local, context-sensitive knowledge that makes these systems work.)
So this is an ecology problem
Taken together, these failures have a common shape — and even when something does get big, the surface pattern often spreads while the generative logic dies, which is closer to embalming than scaling. The harder test is whether a thing can travel well. Can it survive contact with incentives that would cheapen it? Can it grow without skipping the stages it needs? Can it regenerate in the hands of strangers without losing its essence? The more a thing depends on local knowledge, dense relationships, patience, and integrity, the more vulnerable it is — and the winning systems actively reshape the landscape behind them. Monoculture absorbs $9.3 billion a year in subsidies that make transition nearly impossible. Platforms destroy the habit of the open web. The colonizer locks in an ecology that is hostile to whatever might follow.
Sometimes someone finds a way through. Jobs took PARC’s ideas and got them into millions of hands while Xerox — orders of magnitude larger — couldn’t move them past a $100,000 price tag. The open internet did spread, for a while. Amsterdam did transform, after enough people died. But these are exceptions, and most of them required crisis, unusual taste, or both.
The question I keep coming back to is: what would an ecology that selects for maturity look like? What has to be true — about incentives, institutions, norms, ownership — for a better pattern to spread without being hollowed out?
I don’t have an answer yet. But I think this is one of the hidden subproblems beneath a lot of the problems I care about.