Paul Kedrosky, a prominent researcher at MIT, is finally saying what the elites have whispered over lunch: the AI coding revolution is a very expensive delusion. Goldman Sachs researchers found that 40 software companies are already overrunning their initial AI budgets by "orders of magnitude," with some firms burning 10 percent of their total labor costs on tools that often produce little more than "vibe code."

In this new era, programmers don't write; they simply describe what they want in plain language to tools like OpenAI’s Codex or Anthropic’s Claude Code. While this “vibe coding” makes for great demos, the technical reality is increasingly filthy. According to the Atlantic, the pace of actual improvement isn’t nearly as fast as the hype suggests. We are witnessing the hollowing out of a professional class, replaced by non-technical users prompting machines to hallucinate infrastructure.

This paper suggests that true luxury is, as ever, biological. As the open-source commons becomes a graveyard of unvetted, AI-generated slop, the only real security will be a human engineer who actually understands the syntax. The current market is a spectacle of administrative arbitrage, where companies spend millions to automate tasks that were better handled by a $150,000-a-year human with a pulse.