Rules of Engagement: How SQLite is Managing the AI Invasion

What happens when autonomous artificial intelligence decides it wants to help build one of the world's most critical pieces of software? You write a rulebook specifically for the machines.

Recently, SQLite—a foundational database technology quietly powering everything from your smartphone to web browsers—added a curious new file to its system: AGENTS.md. In the programming world, read-me files are standard guides for human developers. This one, however, functions as a strict set of rules of engagement for AI agents and the people deploying them.

As AI coding assistants transition from simple autocomplete tools into autonomous "agents" capable of analyzing systems and suggesting changes, open-source communities are facing an unprecedented wave of machine-generated contributions. SQLite’s response offers a fascinating glimpse into the future of human-AI collaboration.

Their first rule is a hard boundary: SQLite does not accept AI-generated code. The development team is so serious about this that they recently edited the document to remove the word "(currently)," doubling down on their refusal to merge agentic code into their pristine codebase. If an AI suggests a fix, human developers might look at it as a proof-of-concept, but they will rewrite the actual code themselves.

However, the project isn't shutting the door on AI entirely. While machines aren't allowed to write the software, they are highly encouraged to break it. SQLite explicitly welcomes bug reports generated by AI agents, provided they come with a reproducible test case.

This open invitation led to an immediate, chaotic result: the project's main forum was quickly flooded with AI-generated bug reports of varying quality. To manage the deluge, the community had to quarantine these machine-made submissions into a brand-new, dedicated bug forum. Far from ignoring them, SQLite’s creator, D. Richard Hipp, has been actively working through this new forum, fixing the legitimate issues discovered by the AI bots.

This scenario matters because it perfectly illustrates a pragmatic approach to the AI revolution in the workplace. Instead of fearing replacement or blindly accepting machine output, the SQLite team established a clear division of labor. They relegated AI to the role of a tireless, hyper-vigilant quality assurance inspector, while reserving the creative, authoritative act of writing code strictly for humans. As AI agents begin to enter other industries—from legal research to financial auditing—this model of "AI as auditor, human as author" might just be the blueprint we all need.

Key Points

• SQLite introduced an AGENTS.md file to establish rules for AI agents interacting with its codebase.
• The project strictly bans AI-written code, removing the word '(currently)' to emphasize their firm stance.
• AI is welcomed to act as a quality assurance tool, provided bug reports include reproducible test cases.
• A flood of machine-generated reports led to the creation of a separate, dedicated bug forum.
• This establishes a practical "AI as auditor, human as author" collaboration model for the future.

Why It Matters

It provides a real-world blueprint for integrating AI into complex workflows by setting firm boundaries that keep humans in control of creative tasks while leveraging AI for rigorous testing.

Sources:

• sqlite AGENTS.md — Simon Willison's Weblog