A year into widespread Claude Code adoption, development teams are splitting into two camps: those producing faster, cleaner output, and those buried in AI-generated code they half-understand and can't maintain. The difference usually isn't which model they're running.
ApostropheCMS, the company behind an open-source content management platform, published the second installment of their practitioner guide on using Claude Code without producing what the software industry has started calling "slop" - code that runs but is fragile, bloated, or incomprehensible to the next developer who touches it.
The Mental Shift Most Developers Skip
The core argument is about mental model, not prompts. Developers who treat Claude Code like a vending machine - type in what you want, receive working code - tend to ship things they can't debug later. The more productive framing treats it like a fast but inexperienced junior developer: capable of high output, but requiring constant review and redirection from someone who actually understands the codebase.
That shift changes what you pay attention to. The question stops being "did it work?" and becomes "does this fit the architecture? will this be readable in six months? am I okay owning this?"
What Slop Actually Costs
The productivity math on AI coding tools usually focuses on the upside - tasks completed faster, boilerplate eliminated. The downside gets less attention. Code you don't fully understand accumulates quietly. Each time something breaks in it, you're either back in the AI asking it to fix its own work (which often just moves the problem) or spending the debugging time you saved earlier.
Claude Code can iterate fast, but it doesn't have context about why your codebase is structured the way it is, what constraints you're working under, or what a "good" solution looks like for your situation. That context has to come from the developer. When it doesn't, you get code that solves the literal prompt but misses the actual problem.
The ApostropheCMS guide being part 2 of an ongoing series signals something real: using Claude Code well isn't a technique you learn once. It's a set of habits around review, context-setting, and knowing when to push back on what the model produces. Teams treating it as an ongoing practice rather than a one-time shortcut are the ones getting consistent value out of it.