backward compatibility

Backward compatibility means that even when a new version comes out, anything built for the old version keeps working fine without changes. If you move into a new house and the outlets have a different shape, it would be a problem if all your old appliances suddenly stopped working. If the new outlet is designed so old plugs still fit, that's backward compatibility.

The same idea applies to APIs (the channels through which programs exchange data) in AI services. Even when a provider adds features or changes data structures, programs that already connect to that API should keep working correctly without any code changes. This is usually maintained by leaving existing fields untouched and adding new features as optional extras that can be used or ignored.

When backward compatibility breaks, a single update can suddenly cause existing integration code to throw errors or return wrong values. That's why providers try hard to keep the names and shapes of old fields the same even when adding new features.

The article describes an update to GitHub Copilot's usage API by noting that "this change maintains backward compatibility." The newly added per-agent aggregation fields are just optional extras—nice to have but not required—while the structure of existing fields stays the same, so old code keeps working without modification. Contrary to a common misconception, maintaining backward compatibility doesn't mean "nothing changed." New features were clearly added; it just means the addition was done in a way that doesn't harm existing users.