# Git as Memory & Reversibility

**Track:** Foundations of AI-Native Development — AI-Native Full-Stack Development — complete (30)
**Framework / surface:** web dev (Git / source control)
**Level:** Beginner
**Prerequisites:** Map the Whole System

**In one line:** Commits, diffs, and branches turn agentic code changes into inspectable, reversible experiments.

## Theory, aesthetics & inspiration

Linus Torvalds wrote Git in 2005, in the weeks after the Linux kernel lost its license to BitKeeper, to answer a question agentic development has made urgent again: how do thousands of loosely coordinated contributors change one codebase without losing history or trust? A coding agent can modify ten files in seconds; the chat transcript is not a record of what actually changed, but the diff is. The core vocabulary is small — repository, working tree, diff, commit, branch, merge — and it turns delegation into a sequence of inspectable, reversible experiments: commit a known-good state, ask the agent for one bounded change, read the diff, run verification, commit again. Reverting is not failure; it is the experimental loop working. The history that accumulates is a trace of decisions that stays intelligible long after the model conversation that produced it is gone.

**Builder question:** If an agent made a bad multi-file change right now, could you identify it, explain it, and restore the last known-good state?