Autonomous but Not Alone

The next frontier in AI agents is knowing when to ask for help.

We now have agents that can work for hours, process large amounts of information, and execute long-running tasks. But when something breaks, the computer misbehaves, or screen mirroring decides it has other plans, the agent can get stuck without an effective way to reach me.

So we gave Bob a phone.

I added a simple instruction to our evaluation workflow:

“If you run into an issue, send me a text message saying: ‘Hi Cris, this is Bob. I need your assistance.’”

That might be one of the most human skills we have taught an AI agent so far: do the work independently, but do not fail silently. Escalate when you need help.

Now Bob can work autonomously for hours and, when necessary, send the digital equivalent of:

“Hey, can you come look at this for a second?”

The future of agentic AI may look less like a robot uprising and more like a highly productive coworker who knows when to text the boss.

---

This writing reflects my personal perspectives on product management, AI, and content discovery. It does not represent the official position of my employer or any affiliated organization.