The Better Claude Code Becomes, the Less I Want to Use It
The Better Claude Code Becomes, the Less I Want to Use It
There is a paradox in AI tool development - as the tool gets smarter, it gets more opinionated. And as it gets more opinionated, it starts making decisions you did not ask it to make.
The Opinionation Creep
Early Claude Code was a blank slate. You told it what to do, it did it. You were in control. The tool was powerful but neutral - it amplified your decisions without substituting its own.
Newer versions come with built-in preferences. It wants to structure code a certain way. It has opinions about error handling patterns. It adds comments you did not ask for. It refactors things that were fine. Each individual change is arguably an improvement, but collectively they shift control from you to the tool.
When Tools Think for You
The danger is subtle. When the tool consistently makes good-enough decisions without being asked, you stop making those decisions yourself. Your design muscles atrophy. Your codebase starts reflecting the tool's aesthetic instead of yours.
Six months later, you look at your code and it all looks the same - not because you found a consistent style, but because the tool imposed one. Every project has the same patterns, the same structure, the same comments. The tool's fingerprint is everywhere.
What We Actually Want
The ideal AI coding tool is capable but deferential. It should be able to do everything Claude Code does, but only when asked. It should not reorganize your imports unless you say to. It should not add error handling unless you specify the pattern. It should not refactor unless you point at the code and say "fix this."
Maximum capability with minimum opinion. That is the sweet spot. An agent that does exactly what you tell it and nothing more - until you give it more autonomy on purpose.
Fazm is an open source macOS AI agent. Open source on GitHub.