Do we need a new GitHub for AI coding era?
1–3 of 3 posts
Re: Do we need a new GitHub for AI coding era?
#2Over the last year, coding agents have changed how developers work.
But GitHub hasn’t changed much: it’s still built around the idea that humans write code, and commits are the primary unit of progress.
I’m building MemoV, and I keep running into this question:
What if the core artifact of software development is no longer the code commit, but the conversation + agent plan + diff?
Today’s workflow looks more like: 1. you talk to a coding agent it generates a plan 2. the plan produces code 3. you iterate conversationally 4. many changes never become Git commits
Context is scattered across prompts, diffs, tools, terminals, editors, memory layers… And GitHub simply can’t represent this.
Re: Do we need a new GitHub for AI coding era?
#3Problems with the GitHub model in the AI era:
A commit doesn’t explain why the agent wrote that code.
Critical reasoning (prompts, plans, tool traces) is lost.
Hard to reproduce or continue work across agents/humans.
Version control is code-first, not context-first.
Agents can’t collaborate because they can’t “see” each other’s reasoning.
What Memov tries to explore: Not “a better GitHub,” but a different abstraction: Every interaction (prompt + agent plan + diff) becomes a traceable unit. Memory-first versioning, not file-first. Reproducibility: any agent can resume from context, not code only.
Time machine for reasoning: “why was this written?” not just “what changed.”