Dave Rupert makes a blunt case: when teams prioritize speed, the first casualty is conversation. Consensus takes time. Getting expert input takes time. Resolving disagreements takes time. So under deadline pressure, teams skip all of it. The result isn't faster shipping. It's teams building in different directions, heading toward merge conflicts and incompatible systems that nobody wants to fix later. "Moving fast and breaking things just results in a lot of broken things," as one Hacker News commenter put it.
And then there's AI. Rupert calls LLMs "the ultimate tool in the 'Don't talk to my coworkers' toolchain." Why seek out an expert who might push back when you can ask a model that always says yes? The friction-free interaction feels productive. But it produces products built on entrenched assumptions and higher sunk costs, making the conversations you eventually need to have even harder. Think of the junior developer who spins up a chatbot instead of walking over to ask the senior engineer about the legacy payment system. They get code that works, maybe. But they never learn why the system works that way, what edge cases matter, what failed approaches came before.
Multi-agent frameworks like Microsoft's AutoGen try to automate the friction by assigning conflicting roles to different AI agents. The idea is to force debate and catch problems before code ships. It's a clever technical fix. But delegating consensus-building to agents risks removing humans further from decisions. You get systems that work technically but whose logic nobody on the team actually understands.
The hard truth Rupert points to is that there's no substitute for slowing down and thinking. Engineering management's job isn't pushing tickets across a board. It's making sure the organization rows in the same direction. That requires actual human conversation, even when it's slow and uncomfortable.