As the Vice President of Product at Google Cloud, Michael Gerstenhaber is heavily involved in the development of Vertex AI, Google’s unified platform for deploying enterprise AI. This role provides him with valuable insights into how companies are utilizing AI models and the challenges that still need to be addressed in order to fully unleash the potential of agentic AI.
During my conversation with Gerstenhaber, he introduced a novel concept that caught my attention. He explained that AI models are pushing against three key frontiers simultaneously: raw intelligence, response time, and a third quality related to cost-effectiveness. This third aspect focuses on whether a model can be deployed at a low enough cost to operate efficiently at a massive scale. This perspective offers a fresh and valuable approach to understanding the capabilities of AI models, especially for those looking to push the boundaries of AI innovation.
Gerstenhaber shared insights into his background in AI, highlighting his experience at companies like Anthropic before joining Google. In his current role, he oversees Vertex AI, Google’s developer platform, which caters to engineers building their own applications. The platform provides access to agentic patterns and smart models for inference, empowering developers to create innovative solutions in various domains.
The decision to join Google was influenced by the company’s unique vertical integration across different layers of technology infrastructure. From data centers and chips to model development and inference, Google offers a comprehensive ecosystem that sets it apart in the industry. This integrated approach enables Google to provide a wide range of services, from data processing to consumer interfaces like Gemini chat.
When discussing the capabilities of AI labs, Gerstenhaber emphasized the existence of three distinct boundaries that AI models must navigate. Models like Gemini Pro prioritize raw intelligence, aiming to deliver the best possible outcomes regardless of processing time. For applications requiring real-time responses, latency becomes a critical factor, as intelligence must be delivered within a specific timeframe to maintain user engagement. Additionally, scalability and cost-effectiveness are essential considerations for organizations seeking to deploy AI models at a massive scale, such as content moderation on platforms like Reddit or Meta.
The conversation also delved into the challenges hindering the widespread adoption of agentic systems. Despite the impressive advancements in AI technology, there is still a lack of infrastructure for auditing agent behavior and securing data authorization. Implementing these critical components into production environments remains a complex task, as organizations grapple with integrating AI into existing workflows. While software engineering has embraced AI relatively quickly due to established development processes, other professions and industries require tailored solutions to leverage the full potential of AI technology.
In conclusion, Gerstenhaber’s insights shed light on the evolving landscape of AI and the intricate challenges that accompany its deployment. As companies strive to harness the power of AI for innovation and efficiency, addressing the frontiers of raw intelligence, response time, and cost-effectiveness will be crucial in unlocking the true potential of agentic AI.

