For example, vibecoding seems to shift a lot of the emphasis in software production to code review. The LLM agent produces code, and a human programmer is meant to be extra diligent in checking that code to ensure that it's minimally garbage. Code review has always been a part of collaborative software design, but previously, you were reviewing the work of a human programmer. That's a form of social connection — you see how that person addresses problems, bits of their personality in how they comment or arrange code. At points in the process, you likely discuss the project with them — direct human-to-human contact. If AI is handling the initial production of code, then much of that connection is lost. Not just alienation from the act of working, but also from other people.