When to Use AI While Coding (And When Not To)

AI assistants can speed up coding, but they work best when you know what you want. Here’s how I use them.

Good For: Boilerplate and Repetition

Generating test fixtures, mock data, or repetitive UI from a clear spec. You still review and own the result; the tool just saves typing.

Good For: “What’s the Syntax Again?”

Looking up library APIs, regex, or language quirks. Faster than digging through docs. Always sanity-check the output against the official docs for edge cases.

Risky: Architecture and Design

Letting the model design your system or pick patterns can lead to over-engineered or inconsistent code. Use AI to draft, then apply your own judgment and constraints.

Risky: Security and Critical Paths

Don’t paste secrets or sensitive logic into external tools. For auth, payments, or access control, write and review the code yourself. AI can help with tests and comments, not with the core logic.

Use AI as a lever: it makes you faster when you know the direction. It doesn’t replace understanding or ownership.