Ask HN: Self Sustaining Codebases | Hacker News Reader

Ask HN: Self Sustaining Codebases

4 pointsrs5458376 comments

I've been thinking about where codebases are heading as agents write more of the code.

Right now the pipeline is: agent writes code, opens a PR, another agent or human reviews it, CI runs all 4000 tests, someone clicks merge. This is the human workflow copy-pasted onto machines. It doesn't scale.

What if a codebase could maintain itself? Agents fix bugs as they appear, add features from specs, refactor when complexity grows, update dependencies. All verified automatically, shipped continuously. Humans set policies ("anything touching auth needs my approval", "cosmetic changes auto-ship") and only see the exceptions.

The pieces that make this possible: a dependency graph of every function, class, and method in the codebase. Not files, not lines. Entities. If you know what every entity depends on, you can compute blast radius instantly, detect real conflicts (not the false ones git creates), run only the 4 tests that matter instead of 4000, and score confidence per change. High confidence ships automatically. Low confidence routes to a human.

The PR was invented because humans can't be trusted to push directly to main. If automated verification is more thorough than human review (and benchmarks suggest it is), the PR becomes a receipt, not a gate.

Has anyone else been thinking along these lines? Curious if this resonates or if there are fundamental blockers I'm not seeing.

The experiments with model collapse would seem to indicate that doing this will go very poorly. See also examples of claw spam for what this looks like today

Model collapse happens when AI output feeds back into AI training data. That's a real problem for models.

But self-sustaining codebases aren't using AI to verify AI. The verification layer is deterministic: dependency graphs, targeted test suites, blast radius computation. These are structural checks, not generative ones. The graph doesn't hallucinate. Tests either pass or they don't

The claw spam problem is what happens when you have no verification at all.

Model collapse happens outside of the training loop as well, at the agentic level, when left to their own devices for too long.

The failure mode you're describing is agents verifying themselves. The alternative is structural verification that exists outside the agent's reasoning loop entirely.

The failure mode I'm talking about has them degrading to the point the output gibberish

Reply on news.ycombinator.com