At least from the software engineer pleb perspective, their moat is that their tools seem to work well more often than not. I wasn't comfortable with the idea of using GitHub CoPilot as our GenAI solution at work, and apparently that was a widespread feeling, because we switched to Claude Code, and it's been a relatively smooth transition from manual coding to GenAI agentic loops.