It is only six months since I wrote about my agentic AI worries.
Since then, things have moved on to a new level, and I have new advice.
The dominant safety assumption in AI has been "human-in-the-loop" - if a person is watching, the system remains controllable.
But as agentic AI systems grow more complex, faster, and more interconnected, something important is becoming clear: in the most consequential scenarios, humans can no longer realistically stay in the loop.
We may need a new governance model that relies on other machines being in the loop while humans move above it.
I call this HOMIL.
You can find the LinkedIn newsletter version here: https://www.linkedin.com/newsletters/the-hornet-7376193049816625152/
