I pay for Cursor, Claude Code, and Aider. They're not substitutes. The "X is dead because Y" posts are usually written by someone who only uses one of them. Here's the decision tree I run on every task.
Short version
Cursor for inline coding in a codebase I already know. Best DX for "edit this, refactor that, add a test."
Claude Code for autonomous multi-step work. Best when the task is "go figure out how to add feature X, run tests, commit."
Aider for terminal-native work, and when I want maximum control over which files the model sees.
