Some of the most “AI-native” mid-sized and above companies I observe are the same ones that have had dedicated infra platform and dev platform teams, “pre-AI.” These co’s have always had robust dev tooling + release pipelines and so now AI tools have so many verification layers they can use (the existing systems.) For startups starting from scratch, they need to eventually rebuild all those systems: from eg ones for automated testing, manual testing, FF, canary rollouts, o11y,…