Challenges of revision control in the LLM era (gist.github.com)

Recent discussions highlight significant challenges in software development brought about by the rise of large language models (LLMs) and the limitations of traditional version control systems like Git. As developers increasingly rely on LLMs for browsing and editing code rather than direct coding, the effectiveness of integrated development environments (IDEs) and Git diminishes. The need for improved methods of understanding and controlling code is paramount, as the collaborative dynamics of coding evolve to incorporate LLMs as essential partners in development. The author argues that it may be time to seek alternatives to Git, as its inherent architectural constraints hinder progress, particularly in handling collaborative development and code restructuring. The core issues identified include the difficulties with modularity in Git’s monorepo approach, the complications arising from merge and rebase processes, and the restrictive nature of Git’s data model. To address these challenges, the author proposes developing a new revision control system that utilizes abstract syntax trees (ASTs) and formal deterministic merge algorithms rather than treating code as simple blobs. By leveraging Replicated Data eXchange (RDX) format with CRDT (Conflict-free Replicated Data Types) semantics, the vision is to create a more intuitive and seamless revision control experience that accommodates the demands of modern development environments shaped by LLMs.