Git Is Unprepared for the AI Agent Coding Surge

Mitchell Hashimoto, co‑founder of HashiCorp, publicly announced moving the popular open‑source terminal emulator Ghostty off GitHub, citing frequent service interruptions and increasingly slow pull‑request handling that waste developers' time.

He defended Git itself, arguing that the real issues lie in the surrounding infrastructure—issues, PRs, Actions—especially as AI‑generated pull requests surge. According to a GitClear study, AI‑generated code creates an average of 10.83 problems per pull request versus 6.45 for human‑written code.

These trends raise fundamental questions about the evolution of the software development lifecycle (SDLC) and whether Git must adapt. Autoptic co‑founder Peco Karayanev warned that the influx of AI agents is pushing GitHub toward collapse, emphasizing the need for more continuous Git operations.

GitButler, founded by Scott Chacon and others, offers a powerful client with virtual‑branch technology, richer file metadata, and tools to eliminate “merge‑hell” by keeping user code synchronized with upstream branches. While many of its features can be replicated with native Git commands, its usability improvements aim to reduce error‑prone workflows.

Other alternatives include Diversion, a game‑industry‑born distributed version‑control system, and Jujutsu (jj), a Git‑compatible VCS that adds an undo button and allows commits to continue despite conflicts. Gitoxide, a Rust‑based project to rebuild Git, promises multi‑core performance gains and memory‑safety benefits.

Git’s core maintainers, including Junio Hamano and GitLab engineer Patrick Steinhardt, discussed upcoming Git 3 changes at FOSDEM. A major focus is improving how Git manages commit references: the current packed‑refs approach becomes a bottleneck at scale (e.g., a GitLab repository with over 20 million refs). The new Reftable format, inspired by Eclipse JGit, stores references in an indexed binary format, supports block‑level updates, and offers faster reads, paving the way for larger, more complex codebases.

Despite two decades of dominance, the article questions whether Git can remain attractive to the next generation of autonomous developers without addressing these architectural and scalability challenges.