What is a polite and intelligent response to people who say that they track AI agents their enterprise uses using Excel? [this question is very vulnerable to 'wrong answers only' treatment, but please send both kinds, ideally be funny...]

What would actually have to happen for the market to decide the frontier labs just aren’t that great of businesses? Models converge and are commodities? That’s expected. Growth slows? That’s ok, huge installed base. Margins suck? Investing for the future. Capex explodes again and again? Moat. Delay the IPO? Too focused on the mission. These might end up being incredible businesses. It’s just a little wild that every piece of bad news somehow becomes evidence for the bull case.

I think open weights are becoming the tool library the frontier agents reach for. Case in point: I asked Meta's Muse (closed, frontier) to drive my robot rover. To judge distance it went to GitHub and installed DepthAnything V2, an open-weight model.

Most people are treating AI agents like smarter browser tabs. They're not. They are workers with network access, identity, permissions, and the ability to take action at machine speed. You need to give them a defined network. Not the whole internet.