As a product VP at Google Cloud, Michael Gerstenhaber works totally on Vertex, the corporate’s unified platform for deploying enterprise AI. It provides him a high-level view of how firms are literally utilizing AI fashions, and what nonetheless must be performed to unleash the potential of agentic AI.
After I spoke with Michael, I used to be notably struck by one thought I hadn’t heard earlier than. As he put it, AI fashions are pushing in opposition to three frontiers directly: uncooked intelligence, response time, and a 3rd high quality that has much less to do with uncooked functionality than with price — whether or not a mannequin might be deployed cheaply sufficient to run at large, unpredictable scale. It’s a brand new mind-set about mannequin capabilities, and a very precious one for anybody making an attempt to push frontier fashions in a brand new course.
This interview has been edited for size and readability.
Why don’t you begin by strolling us by means of your expertise in AI thus far, and what you do at Google?
I’ve been in AI for about two years now. I used to be at Anthropic for a 12 months and a half, I’ve been at Google virtually half a 12 months now. I run Vertex, Google’s developer platform. Most of our prospects are engineers constructing their very own functions. They need entry to agentic patterns. They need entry to an agentic platform. They need entry to the inference of the neatest fashions on the earth. I present them that, however I don’t present the functions themselves. That’s for Shopify, Thomson Reuters, and our varied prospects to offer in their very own domains.
What drew you to Google?
Google is I feel distinctive on the earth in that we now have all the pieces from the interface to the infrastructure layer. We will construct knowledge facilities. We will purchase electrical energy and construct energy vegetation. We have now our personal chips. We have now our personal mannequin. We have now the inference layer that we management. We have now the agentic layer we management. We have now APIs for reminiscence, for interleaved code writing. We have now agent engine on prime of that that ensures compliance and governance. After which we even have the chat interface with Gemini enterprise and Gemini chat for shoppers, proper? So a part of the rationale I got here right here is as a result of I noticed Google as uniquely vertically built-in, and that being a energy for us.
Techcrunch occasion
Boston, MA
|
June 9, 2026
It’s odd as a result of, even with all of the variations between firms, it looks like all three of the large labs are actually shut in capabilities. Is it only a race for extra intelligence, or is it extra difficult than that?
I see three boundaries. Fashions like Gemini Professional are tuned for uncooked intelligence. Take into consideration writing code. You simply need the perfect code you will get, doesn’t matter if it takes 45 minutes, as a result of I’ve to keep up it, I’ve to place it in manufacturing. I simply need the perfect.
Then there’s this different boundary with latency. If I’m doing buyer help and I must know apply a coverage, you want intelligence to use that coverage. Are you allowed to transact a return? Can I improve my seat on an airplane? But it surely doesn’t matter how proper you’re if it took 45 minutes to get the reply. So for these instances, you need probably the most clever product inside that latency funds, as a result of extra intelligence not issues as soon as that particular person will get bored and hangs up the telephone.
After which there’s this final bucket, the place any individual like Reddit or Meta needs to average your complete web. They’ve massive budgets, however they will’t take an enterprise danger on one thing in the event that they don’t know the way it scales. They don’t know what number of toxic posts there will likely be in the present day or tomorrow. In order that they have to limit their funds to a mannequin on the highest intelligence they will afford, however in a scalable option to an infinite variety of topics. And for that, price turns into very, essential.
One of many issues I’ve been puzzling about is why agentic programs are taking so lengthy to catch on. It feels just like the fashions are there and I’ve seen unimaginable demos, however we’re not seeing the type of main modifications I’d have anticipated a 12 months in the past. What do you suppose is holding it again?
This know-how is principally two years outdated, and there’s nonetheless plenty of lacking infrastructure. We don’t have patterns for auditing what the brokers are doing. We don’t have patterns for authorization of knowledge to an agent. There are these patterns which can be going to require work to place into manufacturing. And manufacturing is at all times a trailing indicator of what the know-how is able to. So two years isn’t lengthy sufficient to see what the intelligence helps in manufacturing, and that’s the place persons are struggling.
I feel it’s moved uniquely shortly in software program engineering as a result of it matches properly within the software program growth lifecycle. We have now a dev surroundings by which it’s secure to interrupt issues, after which we promote from the dev surroundings to the check surroundings. The method of writing code at Google requires two individuals to audit that code and each affirm that it’s adequate to place Google’s model behind and provides to our prospects. So we now have plenty of these human-in-the-loop processes that make the implementation exceptionally low-risk. However we have to produce these patterns elsewhere and for different professions.
