A Quiet Remark
An engineer I have worked alongside for twelve years said something over drinks.
“People who can’t tell git from GitHub are now mass-producing code with AI. Doesn’t that scare you?”
It does. Quite a lot.
Not because more code exists. Because no one qualified to judge that code is standing between it and production.
Git and GitHub
For the record.
Git is a version control system. It tracks who changed what, when, and why. It runs locally. No internet required.
GitHub is a cloud service built on top of git. It adds collaboration, code review, issue tracking. Git is the mechanism. GitHub is the venue.
Git is the engine. GitHub is the road.
People who cannot articulate this distinction are now instructing AI to write software.
Code That Runs vs. Code That Is Correct
AI is remarkably capable. Feed it a prompt and it returns working code. It compiles. The screen renders. Tests pass.
The problem is that “runs” and “correct” are not the same thing.
Code that runs responds to today’s request. Code that is correct can be read by a stranger six months later, modified without collapsing, and does not harbor security vulnerabilities.
Whether code runs is a question a machine can answer. Whether code is correct is a question only a human can answer.
That human needs to understand what a change history means, what a design decision implies, and what structural fragility looks like. In other words, that human needs to understand what git actually does.
The Levee Called Code Review
Software development has a process called code review. Someone other than the author reads the code and evaluates it. Not just for bugs. For design coherence, naming intent, long-term maintainability. It is the mechanism by which “correctness” is collectively assured.
AI can assist with this too. Its observations are precise. It catches things humans miss.
But who decides whether AI’s observations are right?
A person who cannot read git history does not have context. They do not know why the code was written the way it was. Without context, all they can do with AI’s feedback is trust it blindly.
That is not review. That is faith.
A Question of Range
A telescope does not give sight to the blind. It extends the range of someone who already sees.
AI works the same way.
A capable engineer using AI turns a ten-hour task into one hour. The freed nine hours go into refining architecture, hardening tests, writing documentation.
An incapable person using AI produces ten times the volume of code that runs but is not correct. And they have no way of knowing.
AI extends your range. It does not create it.
Invisible Debt, Mass-Produced
Code that runs but is not correct is called technical debt.
Like financial debt, it compounds silently. At first, nothing seems wrong. The screens load. The client is satisfied. Then one day, a single change causes the entire system to collapse.
Code written by someone who cannot trace git history is debt that cannot even be identified. You want to pay it back, but you do not know where the balance sheet is.
This is why a twelve-year veteran said the word “scary.”
Those who see it, see it clearly. Those who do not, do not know they are blind.
The Question Is Not About the Tool
AI will continue to improve. Code generation will get more accurate. Review quality will rise. Test coverage will expand.
None of that erases the question.
Can you, the person using AI, evaluate what AI produces?
The difference between git and GitHub is a small piece of knowledge. But not knowing it signals something larger: a lack of resolution on how software is managed, how it accumulates, and how it breaks.
Tools evolve. The precision of the hand that holds them does not evolve on its own.