Clean code still matters — even when coding agents do the typing

The gist

It has been reported that keeping code clean helps both humans and agents. A recent thread and linked essay argue that although coding agents behave differently from human devs, they suffer from the same practical constraint: messy structure makes them slower, more error-prone, and more expensive to use. Robert Martin’s Clean Architecture is invoked to separate value (it works) from structure (how it’s organised), and the takeaway is simple: structure matters.

Why structure still matters

Coding agents are limited by context windows — the number of tokens they can keep “in memory” — so sprawling, tangled codeforces them to read more than necessary. Result: degraded performance and higher token costs. Think of it as cognitive load for machines. Readability, simplicity, modularity and testability don’t just help humans; they shrink the slice of a repo an agent needs to touch. The emotional sting? You can automate many tasks, but sloppy structure still steals your momentum and your budget.

Practical moves

The essay doesn’t offer a miracle cure, but it does suggest actionable steps: tell your agent not just what to build, but how the code should be organised; maintain a consistent style (agents pick up repo conventions fast); and always review agent changes — for now, human oversight is non-negotiable. Tests, small modules, and clear interfaces pay off twice: once in developer sanity, and again in fewer tokens burned by your models.

Bottom line

So: will LLMs make clean code irrelevant? Allegedly not. Automation changes the unit of pain from typing to token cost and context clarity, but the remedy is the same old medicine — discipline, modular design, and review. A little housekeeping today saves a lot of chaos (and billable tokens) tomorrow.

Sources: yanist.com, Hacker News