Git vs GitHub: What Each Does When an AI Writes Code
Git and GitHub get used as if they were one thing. They are not, and the difference decides what an AI coding agent can break. Here is the split.
Git vs GitHub comes down to one sentence: Git is the tool that tracks changes to your files, and GitHub is a website that hosts Git projects online so people can share and review them. Git works on your own computer with no account and no internet. GitHub is a company's service built on top of Git. When an AI coding agent "commits" your work, that is Git. When it "opens a pull request", that is GitHub.
GitHub's own documentation puts it the same way. It says "Git is a version control system that tracks changes to files" and that "GitHub builds on Git by hosting your Git projects, called repositories, in the cloud." The Git project describes itself as free and open source, released under the GNU General Public License version 2, and notes that hosting services like GitLab, GitHub and others are why it became so widespread. In other words, GitHub is one of several places to keep a Git project, not the only one.
Git vs GitHub side by side
Git | GitHub | |
|---|---|---|
What it is | A version control tool | A hosting service and collaboration site built around Git |
Where it runs | On your computer | On the web, with an account |
Needs internet | No | Yes |
Core idea | A history of snapshots (commits) you can return to | A shared copy of the repository plus reviews, issues and automation |
Who makes it | An open source project | A company |
Alternatives | Rarely replaced | GitLab, Bitbucket and others host Git too |
What an AI agent does with it | Saves, branches, undoes, compares | Opens pull requests, reads issues, runs checks, can publish code |
What Git actually does
Git keeps a history of your project. Every time you save a snapshot, called a commit, Git records what changed, who changed it and a short message. You can look back through the history, compare two versions, or return the whole project to an earlier state.
Three words cover most of the vocabulary an AI agent will use:
Repository. The project folder plus its full history.
Commit. One saved snapshot with a message.
Branch. A separate line of work, so an experiment does not touch the main version until you decide it should.
Git on its own never sends anything anywhere. A commit stays on your machine until someone deliberately pushes it somewhere.
What GitHub adds
GitHub gives the repository a home on the internet, and then adds the parts that make teamwork possible:
A shared copy that several people, or several agents, can work from.
Pull requests, where a proposed change is reviewed before it joins the main branch.
Issues, a to-do list that agents and people can both read.
Automated checks that run tests when code is pushed.
Permissions, so you decide who can change what.
None of that is Git. If you deleted your GitHub account, the project history would still exist in every copy of the repository stored elsewhere.
Why the split matters when an AI agent is writing code
The practical difference is reversibility and visibility.
Local Git actions are forgiving. An agent that commits to a branch on your laptop has changed only your laptop. You can inspect the result and discard it. That is why most good agent workflows tell the agent to commit in small steps, and why how to use Git with AI coding agents recommends a branch per task.
Pushes and GitHub actions are public, or at least shared. Once code is pushed, other people can see it, automations can run on it, and anything sensitive in it has left your machine. That includes a pasted password. What to do when an AI coding agent commits a secret to your repo shows why deleting the file afterwards is not enough, because the secret stays in the history.
Review is where mistakes get caught. The pull request is the natural checkpoint for AI-written code, and reading the diff is the habit that matters most. See how to review an AI agent's git diff before merging.
Protection settings live on GitHub, not in Git. Requiring a review before changes reach the main branch is a GitHub rule. How to set up branch protection for an AI coding agent covers the setting that stops an agent pushing straight to your live code.
A simple way to remember it
A useful comparison is a document and a shared drive. Git is like the version history inside a document editor, working on the file in front of you. GitHub is the shared drive where several people open the same document, leave comments and approve changes. You can have version history with no shared drive. You cannot have the shared drive's review features without something underneath tracking versions.
Do you need both?
If you build alone and never share the code, Git alone is enough, and it is still worth using, because it gives you an undo button for everything an AI agent does. If you want a backup off your machine, a place for a client to review work, or automated checks, you add a host, and GitHub is the most common one. Some AI app builders manage the repository for you, in which case you may never type a Git command, but understanding the layers still tells you what is happening to your code.
To see where this tooling sits among everything else an AI-assisted developer uses, the pillar article on AI coding tools is the map.
Frequently asked questions
Is Git the same as GitHub?
No. Git is the version control software that runs on your computer. GitHub is a website that stores Git repositories online and adds review and collaboration features.
Can I use Git without GitHub?
Yes. Git works entirely on your own machine, and you can also host repositories on other services such as GitLab. GitHub is a popular choice, not a requirement.
Do AI coding agents use Git or GitHub?
Usually both. Local commits and branches are Git. Opening a pull request, reading an issue or triggering an automated check is GitHub, or whichever host you use.
Is my code public if I use GitHub?
Not automatically. A repository can be private or public, and you choose when you create it. Check the setting before pushing anything, especially if an agent created the repository for you.
Is Git hard to learn?
The core of it is small: commit, branch, push, pull. You can get far with those and let an AI tool handle the unusual cases, as long as you review what it did before you share it.
How did this land?
About the author
Staff Engineer, Platform
Carlo works on the platform that turns prompts into running apps. He writes the engineering deep dives and the changelog notes worth reading.
