In this video I react to that viral claim: “20 million developers used to gatekeep software, now AI makes it available to everyone.” I get why people say it—AI tools like Claude absolutely make it easier for high-agency non-technical folks to build basic apps, prototypes, and simple integrations that used to require hiring someone. But calling developers “gatekeepers” is just wrong, both technically and culturally. We built the ecosystem: docs, courses, blogs, open-source, FreeCodeCamp-style learning—most of it free—and a lot of that knowledge is what these models were trained on in the first place. The real point is: AI is great at instruction following and brainstorming when you already know your stuff. It’s not great at making end-to-end decisions, especially in the “unknown unknowns” zone—things you don’t even realize you’re missing. I use a concrete example: you can prompt an AI about scaling to a million users and it’ll happily recommend AWS, then Hetzner to cut costs, then self-hosting—without truly owning the real-world failure modes and operational tradeoffs. My takeaway: AI replaces subpar, low-skill output, not experienced engineering judgment. You can build an MVP faster than ever, but shipping a real product, repeatedly, at scale still needs technical supervision and real-world experience. Don’t get fooled by marketing narratives—upskill, learn the fundamentals, and use AI as leverage, not as a substitute for thinking.