release

sqlite-utils 4.0rc2, mostly written by Claude Fable

For developers building AI workflows, this shows that LLMs can be leveraged not just to create code, but to rigorously review existing code for subtle, data-critical bugs—improving reliability before deployment.

Simon Willison·July 4, 2026·3 min readrelease

releasesqlite-utils 4.0rc2, mostly written by Claude Fable

simonwillison.net

What happened

Simon Willison released sqlite-utils 4.0rc2, a version of his Python library for SQLite databases. The release is notable because much of the final review and bug-fixing was performed by Claude Fable, a large language model accessed via Claude Code on iPhone. According to Willison, Fable's initial report identified five 'release blockers,' including a critical data-loss bug in Table.delete_where() that left the connection in a perpetually uncommitted state. Willison had not yet encountered these issues himself. The involvement of an AI in testing and refining an established developer tool highlights a growing trend: models can serve as rigorous code reviewers, especially for edge cases and breaking changes. For builders of AI workflows, this demonstrates that LLMs can be effectively applied to improve software reliability, not just generate new code. The use of Claude Code on a mobile device also shows how AI-assisted development can be done in low-resource environments.

Key takeaways

sqlite-utils 4.0rc2 release candidate is out, with contributions from Claude Fable.
Claude Fable identified five release-blocking bugs, including a data-loss issue in delete_where().
Willison used Claude Code on an iPhone to run the review, showing AI-assisted development on mobile.
The AI caught issues Willison hadn't yet encountered, such as uncommitted transactions corrupting data.
This case illustrates using LLMs for code review and catching subtle bugs before stable release.