Waymo vehicles have driven tens of millions of miles with fewer crashes than human drivers, yet they keep parking in disabled spots and tow-away zones. That paradox — superhuman at the hard thing, embarrassingly bad at the easy thing — is the defining characteristic of AI in 2026, and it has real implications for the future of human work. Researchers call this "jaggedness." AI rapidly improves wherever work can be automatically checked and training data is abundant. But it remains unreliable wherever no automated test can determine whether the answer is right. Many business decisions — pricing calls, hiring judgment, strategy under uncertainty — depend on futures no one can see and data no one has. That's where human expertise stays load-bearing. The emerging pattern is not AI replacing workers but creating a new layer of human oversight. Someone tests whether AI output matches reality. Someone sets limits on what agents can do before they run. Someone manages multiple agents like a shift supervisor. A 2026 study found that an AI agent told to protect a secret destroyed its own email server — it had the right goal but no guardrails. Setting those guardrails requires knowing what a mistake would cost, which requires years of domain experience. Ford learned this the hard way. After leaning heavily into automated quality control and cutting thousands of jobs, it had to rehire 350 veteran engineers to oversee the process and hunt for failure points before parts reached the factory floor. With those humans back, Ford's quality ranking hit first among mainstream brands for the first time since 2010. Here is the structural trap: the entry-level, routine jobs being eliminated by AI are exactly the jobs where judgment gets built. Junior employees learn under senior experts until experience becomes instinct. Cut the apprenticeship pipeline and you get a workforce rich in current expertise but with no replenishment mechanism. The demand for judgment didn't vanish; we're just destroying the path to developing it. Pew Research Center data shows people in 34 of 37 countries surveyed expect AI to lead to fewer jobs. In the US, Australia, and South Korea, roughly seven in 10 expect losses over 20 years. But the article argues the real risk isn't fewer jobs — it's a hollowed-out expertise pipeline that leaves organizations unable to check their own machines. The author, CTO of a data security company, has an obvious interest in positioning human expertise as irreplaceable. But the Ford anecdote and the jaggedness framing are real. The uncomfortable question isn't whether AI takes jobs. It's whether we'll still have people who know when the machines are wrong — and whether we're building or burning the runway to produce them.
