A developer argues that git is fundamentally broken for today's AI-assisted coding workflows, proposing a new database-centric approach. The author contends that as developers increasingly collaborate with LLMs rather than writing code manually, git's blob-based architecture creates friction through merge conflicts, rebases, and lack of semantic understanding.

Current git limitations include difficulty managing monorepos, inability to handle overlay branches for prompts and plans, and absence of syntax-aware queries. The author notes that major tech companies either build custom solutions on top of git or abandon it entirely. The proposed alternative uses AST-like trees with deterministic merge algorithms and a structure-aware query language, built on the RDX format - a JSON superset with CRDT merge semantics.

The vision represents a fundamental shift from git's filesystem approach to treating code as a database. This would enable features like tracking changes to specific functions over time, managing collaborative AI agents, and maintaining semantic context through merges. The author has been developing this system for years, drawing on CRDT research and prior experience with JetBrains IDEs and LLVM internals.