Generate in Small Units, Not a Whole App at Once

A scoping rule: ask ChatGPT for one testable unit at a time — a function, a component, a module — and verify it before moving on, rather than requesting an entire app in a single prompt.

A giant generation is impressive and unverifiable: you can't tell which of 300 lines is subtly broken, and one bad assumption early propagates everywhere. Small units keep each piece within what you can actually read and test, so errors stay isolated and cheap to fix.

Any non-trivial build. The bigger and more interconnected the feature, the more it pays to slice it into units the model gets right one at a time.

Genuinely boilerplate scaffolding (a config file, a standard CRUD stub) where the whole thing is predictable and low-risk to generate at once.

We're building <feature>. Let's go one piece at a time. Start with just <first unit> — its signature, implementation, and a test. I'll confirm it works, then we'll do the next piece.

Instead of 'build a todo app', you generate the data model, then the store, then one component at a time — each verified — and never hit the tangled-blob debugging spiral.

• Prompting 'build the whole thing' and then debugging an opaque wall of code.
• Moving to the next piece before the current one is verified.
• Losing the thread of how pieces connect because none were reviewed in isolation.