I've been experimenting a lot with AI coding tools such as Claude Code, Codex, and others.
Although I can get functionally correct results, I'm always left with a feeling that the code design is not exactly what I would like.
But given the abundance of compute and the revolution that has been going on in software development, I start to wonder: Does it actually matter?
We might be reaching a point where it is cheaper to just let the model run for a bit longer on a potentially messy codebase than having the human in the loop spending their time doing things that don't produce any economic value (at least at first glance).
You see, the point of good software design was to reduce the cost of adding new features to software. If the design was good, it would help us make changes to extend and evolve the software—and that's how we justified to our leaders that we should spend a week refactoring before a new big feature.
But with LLMs completely dominating the production of code in the near future, it might be the case that it just doesn't matter as much.
We might find that things like Clean Code and good design just don't matter as much since the model can "figure it out."
Or even worse, it might be easier for the models to work with code that we find hard to understand.
But as long as the code is compliant with specification, are we good?
Maybe, I guess.
At the end of the day, we are all experimenting and learning about how far we can go with these agentic coding tools.
There's no way back for sure—we are in the middle of a revolution.
But how exactly things pan out from here remains to be seen.