By Russell Brandom
Publication Date: 2026-02-23 19:18:00
As a product VP at Google Cloud, Michael Gerstenhaber works mostly on Vertex, the company’s unified platform for deploying enterprise AI. It gives him a high-level view of how companies are actually using AI models, and what still needs to be done to unleash the potential of agentic AI.
When I spoke with Michael, I was particularly struck by one idea I hadn’t heard before. As he put it, AI models are pushing against three frontiers at once: raw intelligence, response time, and a third quality that has less to do with raw capability than with cost — whether a model can be deployed cheaply enough to run at massive, unpredictable scale. It’s a new way of thinking about model capabilities, and a particularly valuable one for anyone trying to push frontier models in a new direction.
This interview has been edited for length and clarity.
Why don’t you start by walking us through your experience in AI so far, and what you do at Google?
I’ve been in AI for about two…
