February 15, 2026
Merge wars: bring popcorn
Git is a file system. We need a database for the code
Git is stuck; AI-era coders want a real code database
TLDR: A bold proposal says Git acts like a file cabinet when modern, AI-heavy coding needs a true code database. The community split fast: fans want better queries and monorepo support, skeptics demand working code and warn about control, while others point to tools like Trustfall and lit as proof this matters.
A fiery post claims today’s coding isn’t about typing—it's about browsing, understanding, and chatting with AI assistants (called LLMs). The author says Git, the popular tool that tracks code changes, is basically a file cabinet when we now need a searchable database for code history, merges, and queries. Cue the comments section erupting.
The loudest cheers came from the “Git has always been mid” crowd. One dev declared Git “mediocre-at-best” and said AI is finally forcing a fix, dreaming of big project support and a slick virtual file system. But the skeptics brought the smoke: “Missing 4 out of 5 parts,” grumbled one, accusing the effort of trying to wrest control from open source. Another dropped the classic mic line: “Talk is cheap. Show me the code.”
Amid the chaos, the community started tossing links like confetti. Someone pointed to Trustfall to query files like a database; another plugged a prototype idea called lit. The hot takes hit everything from messy merges and false conflicts to “please let me split and rejoin repos without pain.” Verdict? Half the crowd is ready to ditch Git yesterday, half wants proof and performance today, and everyone’s munching popcorn while the repo wars rage.
Key Points
- •The article argues Git’s architecture is ill‑suited for modern, LLM‑assisted development workflows.
- •The author is building a non‑Git‑compatible replacement, seeking a step‑change akin to Git’s leap over CVS.
- •Git struggles with monorepos and lacks principled mechanisms for splitting/joining code and overlaying auxiliary trees (e.g., prompts/plans).
- •Merges/rebases create friction; merges are non‑deterministic, Git is not syntax‑aware, and manual conflict resolution is common.
- •The author calls for SCM as a database with a query language to track semantic and temporal changes, and criticizes Git’s blob/Merkle, all‑or‑nothing data model.