Why I'm still bearish on LLMs after Navier-Stokes

The article argues that despite impressive headline demos, frontier LLMs are still not reliable autonomous replacements for knowledge workers because they require heavy oversight, precise specifications, and robust validation to avoid failure or reward hacking. For CIOs and technology leaders, the strategic implication is that near-term value will come less from full automation and more from constrained, well-governed use cases where outputs are narrow, risk is low, and human review remains feasible; in most enterprises, current AI deployments should be treated as productivity tools, not unsupervised operators.