@snowboat84: There was once a classic in software engineering that swept through the field, called *The Mythical Man-Month*. It drew…

There was once a classic in software engineering that swept through the field, called *The Mythical Man-Month*. It drew from author Brooks's experiences in large-scale software projects in the 1960s and 70s, with the core discussion centering on this: software development isn't simply a matter of adding up "man-months" of work. The larger the scale of an engineering project, the higher the costs of communication, coordination, learning, and system integration between people—meaning that adding more programmers to a project that's already delayed might even make it slower. A natural follow-up question arises: in the era of AI programming, do the conclusions of *The Mythical Man-Month* still hold? Today's AI can already whip up a runnable app in a day, or even just a couple of hours. But there's still a massive chasm between a "runnable program" and a true software product, and from there to a large-scale software system. Features like login, payments, databases, security, error handling, long-term maintenance—and in large systems, dependencies and integration—aren't something you can just fix by "generating more code." There's still a huge gap between AI programming and actual software engineering. Of course, one highly appealing answer is harness engineering: by breaking a large project into tons of small tasks with clear boundaries that can be automatically verified—through modularity, clean interfaces, tests, CI/CD, sandboxes, and agent orchestration—you let AI agents tackle them independently. If this path pans out, then the biggest bottleneck from *The Mythical Man-Month*, namely the costs of human communication and coordination, could indeed be dramatically reduced. But the problem might just get shifted elsewhere: in the past, the bottleneck was collaboration between people; in the future, it could be specification and verification. AI can generate code more and more cheaply, but can we fully define "what's correct" and automatically verify a large system riddled with countless dependencies, edge cases, and unknown failure modes? In other words, AI can quickly churn out *something*, but the verification side remains a bottleneck. So the real issue with AI coding might not be "will AI write code," but rather: can we use software engineering to tame the complexity of large software enough for AI to handle it autonomously? That's still a huge open question right now. If we can, *The Mythical Man-Month* might truly become a relic of a bygone era; if not, what AI might wipe out is mainly the cost of coding, not software engineering itself.