Can you actually make passive income from your dead code?

Most GitHub repos are cluttered with at least 10% obsolete code that just sits there taking up space. We all do it—leaving old utility functions, deprecated API wrappers, or half-finished experiments in a branch because "we might need it someday." Usually, that code just becomes technical debt or a confusing relic for the next developer who joins the project. But there is a growing argument that this "trash" is actually a goldmine for LLM training.

The core idea is that AI labs like OpenAI or Anthropic are constantly hungry for diverse, real-world datasets to improve their reasoning and coding capabilities. Even code that is obsolete for a specific business application can still be structurally valuable for a model learning how to solve a particular logic problem or handle a specific edge case. Instead of letting it rot in a private repo, you could essentially license that data to AI companies.

If you think about the current state of prompt engineering and the push toward more specialized LLM agents, the demand for high-quality, human-written code is only going to increase. Licensing obsolete code transforms a maintenance burden into a financial asset. Instead of spending a weekend doing a manual cleanup or paying for storage and overhead on legacy modules, you turn that codebase into a stream of passive income every time the data is utilized in a training set.

From a practical standpoint, this shifts how we view the "cleanup" process. Usually, we think of a hands-on guide to refactoring as the only way to handle old code. But if a licensing model exists, the "deployment" of your old code moves from a server to a training cluster. It makes a lot of sense for startups or independent devs who have built dozens of prototypes over the years; those repositories are essentially untapped data lakes.

For anyone curious about how this works in practice, there are platforms starting to bridge the gap between private repos and AI labs. You can check out the specifics of how companies handle these licenses here:

https://pangea.ai/code-licensing/companies?id=1135

It is a weird concept at first, but in an era where data is the new oil, even the "leaks" and "waste" in our repositories have a market value. It turns the chore of repository auditing into a potential revenue strategy.