Your AI agent will act like a second self, so ask who it works for

Mike Mignano, a general partner at the venture capital firm Union Square Ventures, was describing a near future in which an AI assistant stops answering questions and starts handling money, messages, and daily plans. The prospect led to the question behind a post he published a few weeks ago, titled “Who is your agent working for?” His answer was direct: “I want that thing to work for me. I want that thing to be completely aligned with my incentives.”

The concern, as Mignano put it, is that the incentives of models built by the big AI labs are to make those labs’ models smarter, better, and faster. He says he wants an agent that shops, books, and plans for him to work for him and stay aligned with his incentives.

A web browser learned about our interests, but it did not do much on our behalf and left control with us. Agents will be different because they will spend, send “very personal messages to family members and loved ones,” as Mignano put it, and act as what he called a “second self.”

Mignano predicts comfort with that kind of handover will become normal within five years. He said “people will get incredibly comfortable handing over the keys in the same way” they grew comfortable “putting our credit cards online using Apple Pay” and finding a spouse online. Harry Stebbings, who was questioning him, put the privacy point more bluntly, saying he did not think anyone outside hardcore technologists cares about privacy. Mignano said there was a good chance Stebbings was right, citing what he called “the history of the internet and personal computing.”

Mignano’s hope is narrower than changing everyone’s mind. He argued it may be enough if a couple of good actors offer human aligned alternatives, meaning agents built to follow the user’s goals, not the company’s, and thereby keep the rest of the market in check.