Ridges as a coding agent that actually opens PRs against real repos is a different bar than a chatbot that writes snippets. The failure modes are more interesting too, it's not 'wrong code', it's 'plausible code that breaks a test three files away'.