When to use Claude Code and when to just write it yourself

Claude Code saves time on some tasks and costs time on others. The mistake most people make is using it by default, not by choice. Here's how to tell the difference before you start.

Use Claude when the pattern is clear

Claude does well when there's an obvious correct answer and the work is mostly execution. Adding a new field to an existing API that already has 10 other fields. Writing tests for a function whose behavior is documented. Migrating a config file to a new format. Converting between two data structures.

These tasks have a clear shape. Claude can see the pattern and replicate it. The output is predictable, the review is fast, and it's faster than doing it yourself.

Write it yourself when the decision matters

Anything where you're making a choice that will be hard to undo deserves your own thinking. Database schema design. Authentication architecture. The data model for a new feature. How to structure a module that other things will depend on.

Claude can help you think through options for these. But having Claude make the choice for you, then reviewing what it chose, is usually slower and riskier than thinking it through yourself first.

Use Claude for boilerplate you'd rather not write

Form validation logic. CRUD endpoints for a new resource. Converting a class component to a function component. Error handling wrappers. Documentation for existing functions. These are tedious but mechanical. Claude handles them well and the review is fast because you know what it should look like.

Write it yourself when you need to understand it

If you're going to maintain this code, you need to understand it. Using Claude to write something you don't fully follow means you've taken on a maintenance burden without the understanding to discharge it.

Learning by reading Claude's output is slower than learning by writing. If understanding the code is part of the goal, write it yourself and use Claude as a reviewer instead.

The good heuristic

Would you be able to write this yourself in 10 minutes? If yes, it might still be worth using Claude — 10 minutes is 10 minutes. But the review overhead is low because you understand it well enough to spot problems quickly.

Would it take you an hour? Claude probably saves real time here, and the review overhead is worth it.

Would it take you most of a day? Use Claude, but stay closely involved. The longer the task, the more ways it can go sideways.

The context switch cost

Starting a Claude session, writing a good prompt, reviewing the output, fixing what's wrong — that has a fixed cost. For a 5-minute task, that fixed cost might be more than just writing it. For a 2-hour task, it's negligible.

Small tasks that you know how to do: write them yourself. Medium tasks that are mechanical: use Claude. Large tasks: use Claude but plan for active review time.