Claude Code Review

Claude Code's founding choice is the terminal, and it turns out to be the right one. With no editor to defend, the agent gets the whole repository, a shell and permission to run things, so a task is stated once and it explores, writes, tests, reads the failure and goes round again without being walked through each step. Developer surveys put it well clear of the alternatives on affection, and the reason given is always the same: it holds a large codebase in view and finds the file you had forgotten existed.

It also runs inside VS Code and JetBrains as reviewable diffs, and headless in a pipeline, so the command line is a starting point rather than a cage. The friction is real. Usage caps brought in to stop people running agents around the clock produced a loud backlash, the heavier plans cost as much as a small team's whole tool budget, and a prompt with no visible state asks more of a newcomer than a panel in an editor does. Driven with a diff open, it is the most capable coding assistant on offer.