How Anthropic’s safety first approach won over big business—and how its own engineers are using its Claude AI

Anthropic engineers mostly use Claude for debugging, not writing new software, but they worry about deskilling