I’ve started to notice something: when I keep my code split into small, focused modules instead of the old-school “big class in a big file” approach, LLMs actually help me better. They think clearer. They make smarter changes. And honestly, I save time.
A huge file forces an LLM to swim through noise before it even sees the method I want to modify. It’s like asking someone to fix a kitchen while standing in the middle of a shopping stairway. But give the model a tiny, tight module with one real responsibility, and suddenly the quality of its suggestions jumps way up. It doesn’t have to guess the context because the context is the file.
This flips the classical coding mentality on its head. Humans could scroll a thousand lines and mentally bookmark things. LLMs operate on context windows. They reason better when the playground is smaller. So instead of stuffing everything into a monster class “because that’s how we’ve always done it,” I break things apart and let the AI work with clean, isolated chunks.
And here’s the bonus:
Small files = fewer tokens = less compute = less energy = saving time and of course less money spent.
It actually helps the environment. Not in a poetic way, but in a real, measurable way.
People love to talk about “AI changing how we code.”
Well, this is it. The code structure itself shifts.
Not for style points, but because an LLM is now part of the team and it performs at its best when the files are lean, sharp, and focused.
I’m not doing microservices for code. I’m doing micro-contexts for LLMs.
And it works.