Who gets to shape the self-driving company?
The interesting question is who gets to turn judgment into a shared way of working.
Most companies already depend on judgment their systems cannot express. Someone knows which exception should stop a process, which number in a report is misleading, or which customer needs a call instead of another automated message. Human Vibe-* (CSM/sales/service/HR/fill in the blank)
I've talked about this before but from an engineering POV: when every change has to pass through a software team, the person who sees the problem has to translate their judgment into a request and hope it returns intact. They explain what happens, why the normal process fails, and what the software should do instead. Someone else turns that explanation into a specification, a priority, and eventually a feature.
Each handoff creates another chance to lose the part that mattered.
The promise of agents is that the person closest to the work can describe a better approach, try it, and correct it without handing the idea through several layers first. Their judgment can become a way of working that someone else can use.
Marc Andreessen’s recent comment about Replit’s idea of an “Agent Habitat” gets at this: enough structure to make bottom-up agent use visible and governable without the shenanigans of management layers, motives, politics, etc.
If every improvement still has to originate with a central AI team, the company has adopted new technology without changing who gets to shape its operation. The people doing the work remain sources of requirements rather than authors of the system.
Does that make the central team unimportant? Someone still needs to build the habitat, define its boundaries, and make sure one person’s useful automation does not become everyone else’s invisible risk. But who that is, oy, is changing rapidly, isn't it?
The promise (that I think is reality now in pockets of the world) is that someone can notice a better way to work and just make it happen without first becoming a software developer or waiting for a place on a roadmap.
So I would ask a simpler question about any self-driving company: can the people who see a better way make it happen? Or do they still need permission from people who don’t know how to put agents to work?
