Open Source Is Starving While AI Makes Coding Free
The Open Source Crisis Nobody's Talking About
Developer costs are plummeting toward zero. AI coding agents can scaffold an app in minutes. A solo founder with Claude can ship what used to take a team of five.
And yet, open source is in crisis.
Maintainers are burning out at record rates. Critical infrastructure projects survive on the goodwill of one or two exhausted volunteers. The xz backdoor wasn’t an anomaly — it was a symptom of a system running on fumes. The “one random person in Nebraska” meme stopped being funny years ago.
We have the cheapest labor in the history of software, and the projects that hold up the internet are still starving for contributors.
How?
Hey, I’m Lakshmi — a Principal SRE building SaaS products on the side. I write about what actually works when you’re shipping solo.
New here? Start with Why Your AI Wakes Up Every Morning With No Memory or Clean Code Is Dead.
The Founding Myth
In 1997, Eric Raymond published The Cathedral and the Bazaar, the essay that became open source’s origin story. The argument was simple: software built like a cathedral — centrally planned, tightly controlled, released when perfect — loses to software built like a bazaar — messy, open, iterated in public by a swarm of contributors.
Linux beat the cathedral. The bazaar won. Raymond’s most famous line became gospel: “Given enough eyeballs, all bugs are shallow.”
Twenty-nine years later, the eyeballs are disappearing. The bazaar is starving. And a new kind of cathedral has risen — one Raymond never imagined.
Cheap Labor Doesn’t Flow to Maintenance
Here’s the disconnect: AI-generated labor isn’t flowing to maintenance. It’s flowing to creation.
Vibe coding doesn’t fix bugs in abandoned logging libraries. It generates new apps. Agents build what they’re told, and nobody tells them “go triage issues on this unglamorous project that 40,000 packages depend on.”
Raymond’s bazaar worked because contributors had intrinsic motivation — they scratched their own itch. Agents don’t have itches. They have prompts.
The result: an explosion of new software built on a foundation that’s slowly rotting.
Linus’s Law Is Breaking
Raymond’s key insight was that open development creates a natural immune system. Bugs get caught because many eyes are watching. The bazaar is self-correcting.
This assumed two things that were true in 1997 and are increasingly false today:
First, that someone wrote the code. Vibe-coded software often has no human author who deeply understands it. The person who prompted it into existence may not be able to read it. The “author” is a model trained on the commons, producing plausible-looking code that works until it doesn’t. I’ve written about this failure mode — code that compiles, passes tests, and is subtly, catastrophically wrong.
Second, that someone reads the code. Open source review depends on humans who care enough to look. But when code is generated at machine speed, the review bottleneck becomes catastrophic. Maintainers are already drowning in AI-generated pull requests — superficially clean, structurally hollow. The immune system is being overwhelmed not by attackers, but by well-meaning slop.
“Given enough eyeballs, all bugs are shallow” only works if the eyeballs are open.
The New Cathedral
Raymond’s cathedral was Microsoft. Proprietary, closed, top-down. The bazaar beat it because openness was a structural advantage — more contributors, faster iteration, better feedback loops.
But look at what the bazaar runs on today.
Every vibe coder, every AI-assisted open source contributor, every agent spinning up code in a terminal — they’re all downstream of foundation models built inside the most cathedral-like institutions imaginable. Anthropic, OpenAI, Google DeepMind — these are cathedrals that would make 1990s Microsoft blush. Billions in compute, proprietary training data, closed weights, trade secrets wrapped in safety rhetoric.
The bazaar didn’t defeat the cathedral. It moved in upstairs.
Open source in 2026 means building with tools you can’t inspect, trained on data you can’t audit, controlled by companies whose incentives you can’t verify. The irony would make Raymond’s head spin: the most “open” era of software creation runs entirely on the most closed infrastructure ever built.
Vibe Coding: Raymond’s Dream or Nightmare?
“Release early, release often.” Raymond preached this as the bazaar’s core advantage. Vibe coding takes it to its logical extreme — release in minutes, iterate in seconds, ship before lunch.
But Raymond’s version had a crucial qualifier nobody quotes: rapid releases were supposed to come with listening to your users. The feedback loop was the point. Ship fast so you can learn fast.
Vibe coding often skips the loop. Ship fast because shipping is easy. If it breaks, generate a new one. Software becomes disposable. Why debug when you can re-prompt?
This creates a bizarre inversion. The original bazaar was messy but convergent — many contributors pushing toward better software over time. The vibe-coded bazaar is messy and divergent — infinite forks, infinite rewrites, nothing accumulating into lasting infrastructure.
Raymond imagined a thousand people improving one thing. We got one person generating a thousand things.
Does Open Source Even Matter the Same Way?
Here’s the uncomfortable question: if anyone can vibe-code a replacement for your library in an afternoon, what does “open source” even mean?
The traditional argument was access. You shouldn’t have to pay Microsoft for a compiler. You shouldn’t be locked into Oracle’s database. Open source was freedom from vendor dependence.
But when the vendor is an AI model and the product is generated on demand, the bottleneck shifts. You’re not locked into specific software — you’re locked into the capability to generate software. The dependency moved up a layer of abstraction.
Open source used to mean: “here’s the code, do what you want.” The new version might mean: “here’s the model weights, do what you want.” And by that standard, most of the AI industry is firmly in cathedral territory.
What Raymond Got Right (That Still Holds)
It’s tempting to write the obituary for the bazaar. Don’t.
Raymond’s deepest insight wasn’t about code — it was about coordination. The bazaar demonstrated that loose networks of motivated people could outperform rigid hierarchies. That insight is more relevant than ever.
The projects that will thrive in the agent era won’t be the ones with the most AI-generated PRs. They’ll be the ones that figure out how to coordinate human judgment with machine labor. Someone still has to decide what’s worth building. Someone still has to say “this PR is slop, reject it.” Someone still has to maintain taste.
The bazaar’s immune system isn’t dead — it just needs to evolve. Instead of “many eyes on the code,” we need “many minds on the direction.” Maintainers become curators. Contributors become reviewers. The scarce resource isn’t writing code anymore. It’s knowing which code to keep.
Raymond was right that openness wins. He was right that central planning can’t compete with distributed intelligence. He was right that scratching your own itch produces better software than building to spec.
He just couldn’t have predicted that the itch would be scratched by a machine that doesn’t know what itching feels like.
The Cathedral and the Bazaar assumed humans on both sides of the screen. We’re entering an era where that assumption breaks down. The principles survive. The implementation needs an upgrade.
What do you think — is the bazaar adapting or dying? Reply and tell me.
A timely and uncomfortable analysis that articulates the growing tension between AI-driven abundance and the sustainability of open-source infrastructure
Excellent analisys! AI's efficiency is a double-edged sword for open source.