Agents Need Infrastructure to Reach Full Potential

The recent Google Cloud Next '26 conference highlighted a significant shift in the development of AI agents. While previous events focused on showcasing the capabilities of agents, this year's conference emphasized the need for infrastructure to support these agents. As Sundar Pichai noted, the question has moved from 'Can we build an agent?' to 'How do we manage thousands of them?'

The Evolution of Agents

The introduction of the Gemini Enterprise Agent Platform marked a significant milestone in the evolution of agents. This platform provides a foundation for building, scaling, governing, and optimizing agents, moving the conversation away from 'Can I build a cool AI demo?' to 'Can I run thousands of agents safely inside a real organization?'

The platform consists of various components, including the Agent Development Kit, Agent Studio, Agent Runtime, Agent Registry, and Agent Identity, among others. These components work together to provide a comprehensive architecture for agent development and management.

Practical Applications

The developer keynote demonstrated the practical applications of agents in a real-world scenario, such as planning a marathon in Las Vegas. The system used multiple specialized agents, including a planner agent, evaluator agent, simulator agent, and supply chain agent, to propose and evaluate marathon routes, simulate the impact on the city, and handle logistics.

This approach highlights the importance of breaking down complex tasks into smaller, manageable components, and using multiple agents with clear jobs to achieve a common goal.

The Importance of Boring Systems

One of the most underrated aspects of the conference was the emphasis on 'boring' systems, such as logs, identity, access control, and observability. These systems are essential for making agents production-grade and trustworthy.

Features like Agent Identity and Agent Gateway provide trackable identities and control points for agent-to-agent and agent-to-tool traffic, while Agent Registry and Agent Observability provide organizations with a way to know what agents and tools exist, and to debug and optimize their performance.

Future Developments

The community's reaction to the conference announcements suggests that developers are excited about practical workflows, such as the Agent Development Kit, Model Context Protocol, and Cloud Run updates. These developments have the potential to simplify the process of building and managing multi-agent systems, and to provide developers with the tools they need to create production-grade agents.

As the field of AI agents continues to evolve, it is likely that we will see a greater emphasis on infrastructure and 'boring' systems. By providing a solid foundation for agent development and management, organizations can unlock the full potential of AI agents and achieve significant benefits in terms of efficiency, productivity, and innovation.