Michael Gerstenhaber, Product VP at Google Cloud, leads Vertex, the company’s unified platform for deploying enterprise AI. His role gives him a clear view into how businesses are using AI models and where the real challenges lie in advancing agentic AI. During our conversation, he introduced a compelling framework for understanding the current state of AI: models are pushing against three distinct frontiers simultaneously—raw intelligence, response time, and cost efficiency at scale. The first frontier is raw intelligence. Models like Gemini Pro are optimized for maximum cognitive capability. For tasks like writing complex code or solving intricate problems, users prioritize accuracy and depth over speed. Here, the goal is simply to get the best possible output, even if it takes minutes or longer. The second frontier is latency. In real-time applications such as customer support, speed is critical. A model might be highly intelligent, but if it takes 45 minutes to respond, the user will hang up. The value here isn’t just intelligence—it’s the ability to deliver the most accurate answer within a tight time window. The challenge is balancing intelligence with responsiveness. The third frontier is cost and scalability. This is where the economics of AI become paramount. Companies like Reddit or Meta need to moderate vast, unpredictable volumes of content across the internet. They can’t afford to run expensive models on every single post. Instead, they need models that deliver strong performance at a price point that allows for infinite scaling. Intelligence is important, but only if it can be deployed affordably and reliably across massive, variable workloads. Gerstenhaber notes that despite the impressive capabilities of today’s models, agentic systems are still slow to take off in practice. While demos showcase remarkable potential, real-world adoption lags. He attributes this to a lack of production-grade infrastructure. There are no standardized patterns yet for auditing agent behavior, managing data access, or ensuring compliance. These are essential for safe, reliable deployment. He also points out that the software engineering domain has advanced faster than others because it already has a mature development lifecycle. There are safe environments for testing, peer reviews, and staged rollouts—processes that minimize risk. For other fields, such as legal, healthcare, or finance, similar safeguards need to be built from the ground up. What sets Google apart, according to Gerstenhaber, is its vertical integration. From custom chips and data centers to its own AI models, inference layers, agent engines, and even consumer-facing interfaces like Gemini Enterprise, Google controls the entire stack. This allows for tighter coordination and faster iteration across all layers of the AI system. In short, the race isn’t just about building smarter models. It’s about creating systems that are intelligent, fast, and scalable—while also being safe, auditable, and deployable in real-world environments. The frontier isn’t just intelligence. It’s the ability to deliver it reliably at scale.