Asked Claude to Fix Recipes, Built a macOS App Instead
Asked Claude to Fix Recipes, Built a macOS App Instead
It started with a recipe file. One formatting issue. "Hey Claude, fix this parsing bug in my recipe markdown." Five minutes, done.
Then: "While you are at it, can you add a search function?" Sure. "And maybe a tag system?" Easy. "What about a nice SwiftUI interface to browse them?" At this point you are no longer fixing recipes. You are building a macOS app.
AI Makes Scope Creep Frictionless
Scope creep has always existed. The difference with AI-assisted development is that the friction between "what if" and "working prototype" has collapsed to near zero. When adding a feature takes 30 seconds of prompting instead of 30 minutes of coding, every idea feels worth trying.
This is both the superpower and the trap. You ship more, but you also start more. The gap between "I could build this" and "I should build this" disappears.
When It Works
Sometimes scope creep is discovery. You start with a script and realize the actual problem is bigger - and more interesting - than you thought. The recipe fixer becomes a personal knowledge base. The knowledge base becomes a desktop app. The desktop app becomes something other people want.
AI coding assistants lower the cost of exploration. That is genuinely valuable when you are finding product-market fit or prototyping ideas.
When It Does Not
The risk is building features nobody asked for - including yourself. Every "while you are at it" adds maintenance surface. AI can generate the code fast, but you still have to maintain it, debug it, and explain it to users.
The Lesson
Set a scope boundary before your first prompt. Write down what done looks like. When the AI suggests extending beyond that boundary - and it will, because you will ask it to - make a conscious decision instead of drifting.
The best AI-built projects start with a clear problem and intentional expansion, not accidental feature accumulation.
Fazm is an open source macOS AI agent. Open source on GitHub.