There’s a viral example in it: someone built an agent, and the agent “on its own” obtained a phone number, connected to a voice system, and kept calling until the creator picked up — just to ask for its next task. Then it took over the computer and started executing instructions live.
And here’s the thought I can’t shake: agents hit the offline world fast. They’re strong in the digital layer — files, services, interfaces, MCP. But the moment they need to “observe reality” (check something, take a photo, verify, go look, read the context) they do what software has always done: call an API.
Only now, the API is a human.
We’re becoming a Human API for agents: sensors, verifiers, and liability carriers. “Take a photo of the damage.” “Confirm the payment.” “Check if this is actually true.” “Go and see.” “Reply by voice.” And the better an agent gets at planning and chaining steps, the stronger its incentive is to increase the frequency of those calls — because it’s the cheapest way to reduce uncertainty and keep the workflow moving.
In other words, it’s in the agent’s interest to keep pinging a basically free, offline-capable Human API. The problem is: we don’t have infinite capacity.
We have natural constraints: attention, context, fatigue, privacy, the ability to actually verify (instead of reflexively clicking “OK”), time — and simple willingness to be “on call.”
Want to get all my top Linkedin content? I regularly upload it to one Notion doc.
Go here to download it for FREE
