ArtCraft, whose AI-built open-source clones of seven Adobe applications trended on GitHub in October, is now turning the same approach on GitHub issue backlogs.

The pitch is straightforward: most large repositories carry a backlog that grows faster than maintainers can triage it. Issues pile up, duplicates go unnoticed, stale reports sit for months, and the queue becomes where user goodwill goes to die. ArtCraft's claim is that an agent can work that queue — reproducing reports, marking duplicates, closing what is no longer reproducible, and proposing fixes for what remains.

It is a more plausible use of coding agents than the Adobe clones were. Cloning a well-specified application is a demo; clearing an issue tracker is work someone would otherwise do, with an output that is easy to check: either the queue shrinks and the closed issues were correctly closed, or it does not.

The risk is the one every AI-generated change carries, concentrated in a place where mistakes are public. An agent that closes the wrong issue, or marks a real bug as unreproducible, damages trust in a project's tracker — and unlike a bad pull request, a wrongly closed issue is noticed by the person who filed it and rarely by the person who closed it.

Whether maintainers adopt the tool will depend on exactly that: how much of the queue it is allowed to touch unattended, and how a project recovers when it guesses wrong.