What is the MIT license, and what does it actually let people do?

Most developers pick MIT because it is the default in the dropdown and because everyone else picks MIT. That is a fine reason. It is also the reason almost nobody has read it.

It is worth reading. It is one paragraph.

What the license actually says

Strip the boilerplate and the MIT license makes two statements.

The first is a grant. Anyone who obtains a copy of your software may use it, copy it, modify it, merge it into something else, publish it, distribute it, sublicense it, and sell it. There are no carve-outs. Not for commercial use, not for competitors, not for anyone building something you would hate.

The second is a condition, and there is only one. They have to include your copyright notice and the permission text in copies or substantial portions of the software.

Then a disclaimer of warranty in capital letters, which is there to make sure nobody sues you when your code breaks their production database.

That is it. Compare that to any commercial license agreement you have signed and you will understand why MIT won.

What it costs you

Nothing, if you understood what you were agreeing to. Quite a lot if you did not.

You cannot charge for it. You cannot stop a company from wrapping it, closing the source, and selling it. You cannot require attribution anywhere a user will ever see. You cannot stop a competitor using it. You cannot add conditions later to code that is already out there.

That last one is the one people get wrong. Changing the license in your repo changes the license on future versions. Every copy already distributed under MIT stays MIT, permanently, including the right to fork it and carry on. A license is a grant, and you cannot ungrant it after the fact.

If you are fine with all of that, MIT is an excellent choice. It is short, well understood, compatible with practically everything, and it will never cause an argument in a procurement review.

Where this connects to the thing nobody tells you

Here is the part that matters if you have been shipping code for more than a few years.

AI labs need real production code to train coding models. Not tutorials. Not synthetic examples generated by another model, which turn out to be a poor substitute. Real repositories, with commit history, failed approaches, ugly workarounds, and the specific texture of code written under a deadline by somebody who had to make it work.

They already have your public code. Every permissively licensed repository on a public host has been in training corpora for years. Whether every instance of that was lawful is still being fought over. Practically, it is done. There is no licensing market for something already collected, because nobody pays for what they have.

So permissive plus public equals no licensing value. Not because your code is bad. Because the scarcity is gone.

The code that still has scarcity is the code that was never published.

your repo
  │
  ├── public + MIT ───────────► already scraped, no value
  │
  ├── private + client owns IP ► not yours to license
  │
  └── private + you own IP ────► licensable

Almost every working developer has repositories in that bottom branch and has never once thought of them as anything other than clutter. Client projects that shipped and went quiet. Internal tools. The product that did not find a market. Abandoned side projects with two years of commits in them.

┌─ WORTH CHECKING ─────────────────────────────────────────────────────┐
│                                                                      │
│  Private, finished, owned by you, and doing nothing.                 │
│                                                                      │
│  [ See what your repos are worth ]                                   │
│                                                                      │
└──────────────────────────────────────────────────────────────────────┘

Why private code is the valuable kind

Three reasons, and none of them are about code quality.

It is scarce. Nobody has it. That is the entire economic argument and it is sufficient on its own.

It is different. Public code skews heavily towards libraries, frameworks, tutorials, and the kinds of projects people publish. Private code is line-of-business software, integrations held together with knowledge nobody wrote down, and long-lived systems maintained through five framework migrations. Models trained only on public code are noticeably worse at that category of work, because they have seen far less of it.

It comes with history. A repository with years of commits shows how code evolves. That is training signal a snapshot cannot provide, and it is specifically what makes a model useful at maintaining software rather than generating it.

The part that surprises people

Licensing is non-exclusive. You are not selling the code. You are granting permission to train on it, while keeping every right you had before. You can keep using it, keep shipping it, keep maintaining it, and license it again to somebody else.

This is the opposite of how developers instinctively read the word “sell”, and it is why the idea gets dismissed before it is understood. Nobody is asking you to hand over your work. They are asking for permission to read it.

So what should you actually do

Keep using MIT for the things you want people to use freely. It is the right license for that job and this is not an argument against it.

But separate the two piles in your head. Public code is a contribution. It buys you reputation, reuse, and the occasional job offer. It is not an asset and it will not become one.

Private code you own is an asset that has been sitting on your account depreciating for years, and the only reason it has not been valued is that nobody told you it could be.

The next question is whether you actually own it. If the repo came out of client work, the answer lives in a contract you signed and probably did not read carefully.

What to read next

• Who owns the code you wrote for a client, which is the question that decides whether any of this applies to you
• What AI labs pay for code, the only page here with figures on it
• Open source licenses compared, if you are choosing a license rather than auditing one