Bram Cohen Outlines What Version Control Should Look Like After Git

Bram Cohen — yes, the BitTorrent creator — has published a detailed vision for the future of version control in a post titled 'Mañana,' and it's drawing serious attention on Hacker News (270+ points).

Cohen isn't new to this space. He's been working on version control problems for years, most notably through the Chia project's internal tooling needs and his earlier work on merge algorithms. His core argument: Git won, but Git also calcified. The tool that freed us from centralized version control has itself become an obstacle to further progress, not because it's bad, but because its internal model constrains what we think version control can do.

The post lays out a coherent design philosophy rather than just a feature wishlist. Cohen has historically focused on three areas where Git falls short for modern development:

Merge quality. Git's three-way merge is adequate for simple cases but produces unnecessary conflicts in scenarios involving moved code, renamed files, and concurrent refactors. Cohen has previously published work on improved merge algorithms that track semantic changes rather than just line-level diffs. If your team has ever spent an afternoon resolving merge conflicts that a smarter tool could have handled automatically, this is the problem space.

Large repository performance. Git's content-addressable storage model means every clone carries full history. Solutions like partial clone and sparse checkout are bolted on. A ground-up design can treat large-scale repos as a first-class concern rather than an afterthought.

User model simplicity. Git's index/staging area, detached HEAD states, and reflog archaeology are power-user tools masquerading as core workflow. Cohen's vision likely pushes toward a model where the common operations are genuinely simple, not just wrapped in aliases.

The timing matters. We're in a moment where AI-assisted development is generating more code changes faster, making merge conflicts and history navigation more frequent pain points. A version control system designed for the era of AI pair programming — where dozens of speculative branches might be explored in a single session — would look meaningfully different from one designed for 2005-era kernel development.

What makes this more than blog-post futurism: Cohen has a track record of actually building distributed systems that work (BitTorrent is still the most efficient large-file distribution protocol ever designed). Whether this becomes a real tool or remains a design document, the technical analysis of where Git's model breaks down is worth reading for anyone who's accepted Git's quirks as immutable laws of nature.

They're not. They're design choices, and different choices are possible.