Key Takeaways
- Quinn Slack, co-founder of Sourcegraph and founder of Amp, warns that developers retreating into isolated private git environments threatens the open software commons.
- AI-generated throwaway code, or "jelly wear," replaces battle-tested public libraries with unvetted private code built from scratch inside corporate silos.
- As the The Changelog hosts observed, locking software into private repositories starves future frontier models of the public training corpus that made AI coding possible.
- The shift from traditional line-by-line code review to artifact verification accelerates the decline of public code literacy, ending the era where an 11-year-old can inspect Linux or Chromium to learn the craft.
Private Repositories Threaten Code Quality
Quinn Slack spent over a decade building developer tooling at Sourcegraph and Amp. During that time, software development relied on a single shared assumption: developers built in public, stood on each other's shoulders, and shared solutions across companies. Now, AI coding agents are breaking that social contract. Instead of importing a standard, battle-tested open-source package, engineers tell an agent to write custom logic from scratch. Developers increasingly push that code into isolated, self-hosted git environments behind private corporate networks.
Slack sees this fragmentation as an active danger. “I worry about that going away if everyone goes to their own little huts and all their Git repositories are disconnected, and everyone is building things from scratch rather than reusing,” (play from 1:03:23) Slack said. “I worry about the quality going down...” (play from 1:03:33)
When developers build in private huts, they stop contributing fixes back to shared foundations. The open source model succeeded because millions of eyes inspected edge cases in projects like Linux and Chromium. Replacing proven community modules with disposable, agent-generated "jelly wear" creates brittle systems. Each company ends up maintaining an accidental private fork of common computing patterns without the security fixes, community auditing, or optimization that made open source reliable in the first place.
Disposable Code Threatens Future Model Training
The loss goes deeper than code quality. It strikes at developer education and the raw data that powers artificial intelligence. For twenty years, GitHub functioned as an open workshop. Anyone could read production code written by the best engineers in the world. As Slack recalled: “I think of a kid who's 11 years old getting interested in programming realizing that Linux and Chrome and so much of the software that they use, they can go look at the code there. That's something magical, and I'm I'm worried about that going away.” (play from 1:04:05)
As teams move away from line-by-line code review toward automated artifact verification, developers spend less time reading source code. If proprietary AI agents write code that lives exclusively in private repositories, the feedback loop breaks. As the The Changelog hosts pointed out, “Quinn worries that as we all move our code into private corners, our little worlds, we lose the shared commons that open source built, and dare I say it, the corpus these models have been trained on.” (play from 1:22:14)
Modern coding agents exist only because decades of developers shared their work openly under permissive licenses. If engineers retreat behind corporate firewalls to generate isolated codebases, future models will run out of fresh, high-quality public human software to study.
What to Do With This
Audit your engineering team's current development cycle this week. Pull three internal utility scripts or helper tools your engineers generated with AI agents over the last quarter. Strip the proprietary credentials, package them with tests, and open source them on GitHub under an MIT license.