Google Cloud AI Lead Reveals the Three Frontiers Reshaping Model Development
The AI race is not just about building smarter models anymore. According to Michael Gerstenhaber, VP of Product at Google Cloud Vertex AI, the industry is simultaneously pushing against three distinct frontiers: raw intelligence, latency, and scalable cost.
This framework offers a fresh perspective on how enterprises are actually deploying AI—and why some applications remain stubbornly out of reach despite rapid model improvements.
The Three Frontiers Framework
Gerstenhaber, who joined Google after 18 months at Anthropic, describes a nuanced view of model capabilities that goes beyond the typical “bigger is better” narrative:
1. Raw Intelligence
Use case: Code generation, complex analysis, research
“Models like Gemini Pro are tuned for raw intelligence. Think about writing code. You just want the best code you can get, doesn’t matter if it takes 45 minutes, because I have to maintain it, I have to put it in production.”
When quality is paramount and time is secondary, enterprises reach for the most capable models available.
2. Latency
Use case: Customer support, real-time decisions, interactive applications
“If I’m doing customer support and I need to know how to apply a policy, you need intelligence to apply that policy. Are you allowed to transact a return? Can I upgrade my seat on an airplane? But it doesn’t matter how right you are if it took 45 minutes to get the answer.”
For time-sensitive applications, the most intelligent model isn’t always the best choice.
3. Scalable Cost
Use case: Content moderation, large-scale analysis, high-volume processing
“Somebody like Reddit or Meta wants to moderate the entire internet. They have large budgets, but they can’t take an enterprise risk on something if they don’t know how it scales.”
When dealing with unpredictable, massive scale, cost per inference becomes the limiting factor.
Why Agentic AI Is Moving Slower Than Expected
Despite impressive demos and capable models, widespread agentic AI adoption has been slower than many predicted. Gerstenhaber points to a familiar culprit: missing infrastructure.
“This technology is basically two years old, and there’s still a lot of missing infrastructure. We don’t have patterns for auditing what the agents are doing. We don’t have patterns for authorization of data to an agent.”
The Production Gap
Production deployment is always a trailing indicator of technological capability. Two years hasn’t been enough time to develop the operational patterns needed for safe, auditable agent deployment.
Key missing pieces:
- Auditing patterns for agent actions
- Data authorization frameworks
- Compliance and governance tooling
- Human-in-the-loop workflows
Google’s Vertical Integration Advantage
Gerstenhaber’s move from Anthropic to Google was driven by a strategic observation:
“Google is I think unique in the world in that we have everything from the interface to the infrastructure layer. We can build data centers. We can buy electricity and build power plants. We have our own chips. We have our own model.”
This full-stack control spans infrastructure, models, inference, agentic layer, APIs, and interfaces—positioning Google to optimize across all three frontiers simultaneously.
Key Takeaways
- Three frontiers: Intelligence, latency, and scalable cost each require different optimization strategies
- Infrastructure gap: Agentic AI adoption is limited by missing operational patterns, not model capability
- Production reality: Two years isn’t enough time to develop enterprise-grade deployment patterns
- Vertical integration: Google’s full-stack control offers unique advantages for multi-frontier optimization
The organizations that succeed won’t be those with the smartest models—they’ll be the ones that build the infrastructure, patterns, and trust mechanisms needed to deploy AI safely at scale.