Some early morning thoughts on OSS, triggered by reports of projects closing pull requests because of AI slop. Mitchell Hashimoto is considering closing external PRs to his open-source projects. Greenhouse’s 2025 AI in Hiring Report describes a broader loss of trust in recruitment. These feel like two sides of the same shift, and I wanted to think through where that leaves developers who’ve relied on OSS contributions as a way to stand out.
The shrinking long tail
Open source used to be one of the clearest ways for a developer to stand out: ship code in public, let people verify your work, build reputation over time. LLMs are changing that equation, but not by making OSS irrelevant. By redefining what’s valuable.
I expect the long tail of small libraries to shrink. When a team can generate a utility in-house with an LLM instead of pulling in a dependency, some small projects become harder to justify. Supply-chain concerns add another reason to own a little code rather than add a dependency. If that prediction holds, it also means fewer opportunities to build a reputation through casual contributions.
Meanwhile, maintainers of the projects that remain are dealing with AI-generated “slop PRs”. Some prominent maintainers are considering closing external contributions altogether. The casual contribution that once showed initiative now risks getting lost in the noise.
The trust crisis makes signal scarce
This matters because hiring has its own trust problem. Greenhouse’s 2025 report describes recruiters encountering more candidate deception and some companies reintroducing onsite interviews to make AI-assisted cheating harder. Several traditional signals have become less reliable.
That raises a practical question: what still works? How can a developer show they are a good hire even before sending a CV? Which signals retain value?
I don’t think there’s one clean answer, but some signals hold up better than others.
What still works
Sustained, verifiable OSS history still matters. Not just commits, but code reviews, architectural discussions, issue triage, and participation in events. These remain difficult signals to fake. For those who can build and maintain tools others depend on, their value may have increased. Many of these interactions happen in silos such as Discord and Slack, which is not ideal, but they are still visible to people inside those communities.
But we should be honest: that path just got harder and more competitive. For many developers, alternatives will matter more than before: deep domain expertise, technical writing, demonstrable shipped products, or conference speaking.
The case for depth over breadth
Of those alternatives, deep domain expertise may be the most defensible signal. Many ICs (me included) have jumped often across industries, building a broad but shallow understanding of the domains. Fintech for a while, then e-commerce, then data platforms, picking up enough context to be effective and moving on. That worked when the scarce resource was general technical ability. General coding output is easier to reproduce than years of knowledge about why a domain behaves as it does.
But in a world where LLMs commoditise the general-purpose coding layer, the person who deeply understands the domain (the regulatory constraints, the business edge cases, the reasons why the obvious solution doesn’t work) becomes much harder to replace. Breadth got you hired when the bottleneck was “can you code.” Depth gets you hired when the bottleneck is “do you understand what to build and why.”
I’m starting to think the era of the generalist IC hopping across industries every couple of years might be closing. Not entirely, but the calculus has changed.
The emerging skill hierarchy
The skill hierarchy is shifting. “Can use an LLM” is becoming table stakes. “Can use it with judgement” is the baseline. “Can build things others depend on” is where stronger evidence lives, because that combines technical ability with an understanding of what is needed.
OSS isn’t dying. But the version of it that was accessible to everyone as a career shortcut might be. The bar for what counts as meaningful contribution has gone up, and for those who can clear it, the signal is worth more than ever. For everyone else, it’s worth asking: where does your differentiation actually come from, and how defensible is it against the next generation of AI tools?