I spent a full day on Google's San Jose campus with a small group of AI creators.
We went building to building for product demos and panels, ran a workshop where the humans and the agents all had to pick the same coffee shop location, and I pulled a couple of Google engineers aside to talk about where AI actually is right now.
The same themes came up in almost every conversation: trust your data, give agents a clear definition of done, and keep a human in the loop for judgment.