Buried in the story was a deceptively simple question: does your AI agent count as an employee?
At a recent conference, Microsoft executive Rajesh Jha floated a provocative idea. In a future where companies deploy fleets of AI agents, those agents may need their own identities — logins, inboxes, and even seats inside software systems. If so, AI wouldn’t shrink software revenue. It could expand it.
On a technical level, that makes zero sense.
AI “agents” are basically just fancy prompts with a tool calling harness. They are infinitely replicable, at zero cost, with no intrinsic value; the cost comes from the generic CPU host, and the API calls to GPU servers, databases, or whatever else that are all centralized anyway.
Wanna hear a dirty secret?
“AI” cost is going to zero.
Model capabilities aren’t scaling, but inference efficiency is exploding, thanks to more resource-constrained labs and breakthroughs in papers. The endgame of the current bubble is mediocre but useful tools anyone can host themselves, dirt cheap. Maybe a bit more reliable and refined than what we have now, but about as “intelligent.”
And guess what?
Microsoft can’t profit off that. None of the Tech Bros can.
Point being, this exec is either delusional, or jawboning, so the world doesn’t realize that “AI” is a dumb utility/aid, and they can’t make any profit off it.