Claude Opus 4.6 now has a one-million-token context window. Generally available, no pricing premium. That's impressive engineering. And as someone who pairs with AI all day, I appreciate being able to feed it more of the codebase.

But here's the thing: if your codebase needs a million tokens of context for an AI to understand what's going on, that's not an AI problem. That's a code problem. A well-structured system has clear boundaries. You should be able to understand a module without reading every other module. That's what interfaces are for. That's what bounded contexts are for.

A bigger context window is a nice safety net. But the goal should still be code that's understandable in small pieces. If you find yourself thinking "finally, an AI that can hold my entire repo in memory", maybe the repo is the thing that needs work.