2 Comments
User's avatar
Elizabeth Kilrain's avatar

Sometimes I feel like companies rushing to deploy AI agents are like the sudden buying up of sensors a decade ago. The lack of limits on what agents can do is more because a lot of companies don't have a good idea of what they want them for, yet. The AI-first mandate came down from leadership before the guidelines on what to use AI for, and while it makes sense for there to be an exploratory period for teams closest to their own work to envision how AI can change their workflows, a lot of that work might be brainstorming the old-school way, without agents, without giving them access yet, and that just feels like heavy work when the pressure and the mandate is given from above.

I'm currently outlining an essay on this right now, so apologies if my thoughts feel a little half-formed, but the TL; DR of it is that it's hard to set explicit permissions, roles, and restrictions on agent identities because many organizations haven't decided what their roles are supposed to be yet.

AI Delivery Brief's avatar

The $150,000 agent is the story I would pin on the wall. One day of unchecked spending, and the first question is who even notices. A daily ceiling with a named person alerted is the cheapest control in the whole briefing. A ceiling does not slow the agent down, it is what lets a team trust it at all. https://aideliverybrief.substack.com/p/the-insurance-exclusion-says-generative