Oceans of Wisdom

Capitalism is **mostly a bazaar at the level of the economy**, but it contains many cathedrals inside it. The capitalist argument is that no central planner possesses enough information to design the whole economy. Millions of people and firms independently experiment, respond to prices, copy successful ideas, abandon failures, and discover what consumers value. Prices act as compressed signals: scarcity raises prices, demand attracts investment, and losses tell producers that resources may be better used elsewhere. In that sense: * **Central planning is the cathedral:** a small group attempts to design production and allocate resources coherently from the top. * **Competitive capitalism is the bazaar:** decentralized actors try different approaches, with outcomes emerging from competition and exchange. This is close to Hayek’s argument that markets solve a **knowledge problem**, not merely an incentive problem. Relevant knowledge is dispersed among individuals and often cannot be fully collected or articulated. A functioning market allows people to act on their local knowledge without anyone understanding the entire system. But the analogy has an important twist: **the individual capitalist firm is usually a cathedral**. Amazon, Toyota, or a university does not normally allocate every internal resource through an open market. Management sets strategy, assigns budgets, hires people, and imposes architecture. Ronald Coase’s insight was that firms exist partly because using markets for every small decision would be costly. So capitalism can be pictured as: > **A bazaar of competing cathedrals.** Entrepreneurs centrally organize firms, but those firms are externally tested by customers, competitors, investors, employees, and suppliers. The market does not determine how a company should be built; it determines whether the company’s choices survive. There are also differences from the open-source bazaar: * Open-source collaboration is often cooperative and reputational. * Capitalist competition is driven more strongly by ownership, profit, and exclusion. * Open-source participants often share discoveries freely. * Capitalist firms frequently protect discoveries through secrecy, patents, and proprietary control. Capitalism therefore gains some of the bazaar’s evolutionary power—parallel experiments, rapid feedback, decentralized knowledge—but may lose some of its openness. The strongest version of capitalism is not “markets always know best.” It is: > No one knows in advance which ideas are best, so society should permit many experiments and make it possible for successful ones to expand. Its characteristic strength is **discovery**. Its characteristic weakness is that market outcomes reflect purchasing power, ownership, bargaining strength, and externalities—not simply social value. A profitable outcome can still be harmful, monopolistic, environmentally destructive, or based on costs imposed on others. Modern capitalism can also become less bazaar-like over time. Successful firms may accumulate enough capital, data, political influence, or network effects to suppress experimentation. A competitive bazaar can gradually turn into a small number of private cathedrals controlling the marketplace itself. So the most productive political question is not simply capitalism versus planning. It is: > How do we preserve decentralized experimentation and feedback while preventing wealth and power from becoming so concentrated that the bazaar closes?

or capitalism itself

**“The Cathedral and the Bazaar”** is Eric Raymond’s argument for open-source software development, based largely on his experience with Linux. ### The two models **Cathedral model** * Software is designed by a small, centralized expert team. * Development is private and tightly controlled. * Releases are infrequent and polished. * Architecture is planned coherently from the top down. **Bazaar model** * Development happens publicly. * Many contributors inspect, test, modify, and improve the code. * Releases are frequent, even when imperfect. * Direction emerges through experimentation, feedback, and distributed collaboration. ### Raymond’s core argument For many software projects, the bazaar can outperform the cathedral because large communities discover bugs, edge cases, and better solutions faster than a closed team can. His most famous principle is: > “Given enough eyeballs, all bugs are shallow.” This does not mean every contributor can solve every bug. It means that, with enough diverse reviewers, someone is likely to recognize the problem and propose a fix. ### Main principles * **Release early and often.** * Treat users as potential collaborators. * Listen closely to feedback. * Expect good ideas to come from outside the core team. * Be willing to discard or redesign code. * A strong project leader coordinates rather than controls everything. * Open source works best when contributors find the project personally useful or interesting. ### The deeper lesson The essay is not merely about publishing source code. It contrasts two ways of organizing knowledge: * **Cathedral:** intelligence is concentrated, planned, and hierarchical. * **Bazaar:** intelligence is distributed, iterative, and evolutionary. Raymond argues that complex systems can often be built more effectively by creating rules and infrastructure that allow many people to contribute, rather than trying to specify the entire solution centrally. ### Important limitation The bazaar is not automatically superior. It works best when: * the project can attract capable contributors; * components can be divided and tested independently; * feedback is rapid; * leadership can maintain coherence; * contributors share enough technical standards. Cathedral-style development may still be preferable for tightly integrated systems, safety-critical products, secret projects, or work requiring a single strong architectural vision. **In one sentence:** the book argues that software development can function less like constructing a monument from a master plan and more like running a lively marketplace where useful solutions emerge through open participation.

a version of this too

ends up with a positive upward spiral for all

I want them to be motivated to do things, supercharged by agents, and the net effect of what they do

and the hundreds of other people who care here

i do not want to 'coordinate' per se @[HugoFaz]

like markets

it is more like alignment

it is not coordination per se

the question for us is something different

ok it is fine, you can say "you, mike, work on the chat FE this week and we will stay out of your way"

that work for the same organization

on a team of 10

of a network

there are lot of harnesses etc the harder part I think is the higher level alignment

where you do not have a hierarchical developer structure

is what it means when 400 people show up here, each with their claude code and dozens of subagents

the one I do not even have a framework yet

and the even more interesting point here

and what is keeping you from an 100x acceleration

but in the small instance, what happens when you have 10 people on your team using codex and claude code

not "claude code'

it is all about the 'meta' harness

but having a non-stupid CTA on our website is much easier

harder

so if you are doing anything less complex than the informaton theory of physics and you think the models are not good enough to do it, I regret to inform you that the problem is your prompting :slightly_smiling_face:

and they are all literally rocket scientists

anyway he and his team are checking

and he was like "umm, I think it is right"

and was like "so, can you check if I am a moron'

i was sitting around last friday night on the couch and was wondering if possibly and by Tuesday I was done, sendt it to someone who knows 1000000x more phsyics than me

and I did it on a lark more or less, I mean not really I am deeply interested in how we can deeply them

so if the models can do that, the models can improve our landing pages, our CTA, our ERC-721 contracts etc etc, all of this is literally layers of abstraction easier than what I did

on each revision

to edit a PDF

it was runing 1,250 checks to write a PDF

but to give you a sense of what i am talking about

it is not going to win a nobel prize but it is also if it holds non-trivial, the number of people find any theorem whatsoever are smol

possibly I have discovered a theorem, tbd

by properly harnessing Sol and Fable

100x harder than anything we are doing here

i did something extraordinarily advanced in mathematical physics last week

constructor / reviewer

how the repo works, how the CI works

it is super important the meta harness

seeing this on all projects

in principle, if we were sufficiently well structured, we could ship the next two years of roadmap in a week

we have all the intelligence we need