Still Readable

An AI-generated illustration was under fire online.

A piece that took dozens of hours to draw, and an image that appeared in seconds from a few lines of prompt. Indistinguishable. In the CG and illustration world, the debate continues: is that creation? Provenance is being questioned. Process is being questioned. Whether time was spent is becoming part of how work is valued.

In the world of code, this debate barely exists.

The reason is simple. Git. Who changed which line, when, and why. All recorded. Whether AI wrote it or a human did, the commit log and the diff remain. Provenance is traceable. Process is visible. Code had proof baked in from the start.

The first version control I touched was Microsoft SourceSafe. Lock a file to edit it, unlock when done. Two people could not touch the same file. Exclusive locking. Primitive, but hard to break. Hard to break — or rather, it stopped before it broke.

Subversion arrived and established the central repository. Check out, edit, commit back. Resolve conflicts by merging, not locking. A step forward. But if the central server went down, everyone stopped. The fate of centralization.

Then git appeared. Linus Torvalds had been using BitKeeper to manage the Linux kernel. A licensing dispute erupted, and he started writing a replacement. Four days later it was self-hosting. In an era without AI, he designed and implemented distributed version control at the same time. Beyond jealousy at this point.

Git is distributed. Everyone holds a complete copy of the repository. The server goes down, the history is still in your hands. Branches are cheap. Merges are smart. Where SVN was centralized, git treats everyone as equal. The architecture itself was a philosophy.

Think about it. Version control is a system for recording text diffs in chronological order. What changed. What was removed. Who did it. Not limited to code. Legal statutes, constitutional amendments — at their core, they can be told through diffs.

In an age where AI writes code, git's value is rising. Even when AI's name never appears in the commit log, the diff tells you. Code in volumes no human could write, packed into a single commit.

But what git truly records is not the author. It is the trail of trial and error. Of tradeoffs. Why this design was chosen. Why that implementation was discarded. Trace the commit log and the diffs, and the human intent behind AI-written code is still readable.