Google Cloud: Thomas Kurian on the Rise of AI Agents

Thomas Kurian and the Evolution of Google Cloud

This week, Thomas Kurian, the CEO of Google Cloud, gave an interview to Stratechery. Kurian, who took the helm of Google's cloud division in 2018 after a 22-year career at Oracle, shared his thoughts on the recent and future developments of Google Cloud. Since his arrival, he has regularly spoken at Google Cloud Next, an annual event where he presents the company's advancements and strategic directions. The latest interview took place on April 15, prior to his keynote address.

Kurian joined Google in 2018 to lead the cloud division after spending over two decades at Oracle as president of product development. Since his arrival, he has been a key player in transforming Google Cloud, bringing his expertise and vision to evolve the company's cloud infrastructure. His regular interviews, including those in March 2021, April 2024, and April 2025, have allowed for tracking the evolution of his strategies and priorities.

A Unified Architecture in Action

During his speech, Kurian highlighted the shift from a unified architecture, a recurring theme since the previous year, to a large-scale concrete implementation. He emphasized that use cases are no longer merely theoretical but are now being deployed for real users. Sundar Pichai, CEO of Google, also underscored the importance of this investment, specifying that half of Google's capital expenditures were directed towards Google Cloud. This shared infrastructure between Google and Google Cloud is a key point of the company's approach, with a particular focus on security.

Kurian explained that the unified architecture allows Google Cloud to operate on the same infrastructure as Google itself, which enhances the efficiency and security of the services offered. Sundar Pichai highlighted this aspect during his remarks, emphasizing that Google's capital investment was largely dedicated to strengthening this common infrastructure, demonstrating the company's commitment to its cloud offering.

The Era of AI Agents

The interview also addressed the role of AI agents, with a particular focus on the Gemini tool. Kurian expressed confidence in the quality of this tool, highlighting Google's advantage in terms of integration. The discussion also touched on the balance between Google's internal needs and those of its external clients, such as Anthropic. Kurian explained that Google's software ecosystem relies on strong partnerships, and the company is ready to seize the opportunities presented by AI, largely thanks to its vision.

Kurian detailed how AI agents, and particularly the Gemini tool, are designed to seamlessly integrate Google's internal needs with those of its clients. He emphasized that this integration is crucial for maximizing the efficiency and impact of the AI solutions offered by Google Cloud. Furthermore, he stressed the importance of partnerships in Google's software ecosystem, which help to enhance the service offerings and accelerate innovation.

A Framework for Innovation

This year, Kurian presented a framework that highlights the evolution of AI models. These models no longer just answer simple questions; they now automate tasks and processes within organizations. This automation not only improves efficiency and productivity but also transforms how companies introduce new products and services to the market. To support this transformation, a world-class agent platform is necessary, along with a robust infrastructure that enables agents to understand and interact with enterprise data.

Kurian explained that the evolution of AI models now allows for the automation of complex processes, representing a significant shift from simple chatbot-like responses. This advancement opens up new possibilities for businesses, enabling them to rethink how they introduce new products and services. For this, a world-class agent platform is essential, supported by infrastructure capable of managing complex interactions with enterprise data.

Progress of the Gemini Models

Kurian described several major advancements since last year. The Gemini models have significantly improved their reasoning capabilities and can now maintain long-term memory, which is essential for automating complex tasks. Additionally, interactions with external tools and systems have been enhanced through advanced abstractions, allowing agents to handle complex tasks efficiently.

He emphasized that the Gemini models have made significant progress in reasoning capabilities, enabling them to manage increasingly complex tasks. The ability to maintain long-term memory is a major asset for automating complex processes, as it allows agents to keep track of task states across multiple steps. Furthermore, improvements in interactions with external tools and systems enhance the agents' efficiency in managing complex tasks. MCPs (Model Context Protocols) are abstractions that help agents reason and interact effectively with the rest of a company's systems.

Customer Testimonials

Kurian preferred to let Google Cloud's customers speak for themselves regarding the effectiveness of the Gemini agents. At the Next event, around 500 clients, including Citigroup, Bosch, eBay, Virgin Voyages, Walmart, the Food and Drug Administration, Comcast, and Unilever, shared their experiences. For example, Citigroup uses the agents for wealth management advice, researching clients' specific investment priorities, such as funding children's education, and providing personalized recommendations. Comcast, on the other hand, employs the agents to optimize consumer services, including repairs, appointment scheduling, and dispatching technicians on-site. These testimonials illustrate the agents' ability to solve specific business problems through automated workflows.

Kurian highlighted that customer testimonials are the best way to demonstrate the effectiveness of the Gemini agents. At the Next event, companies from various sectors shared their experiences, showing how they use the agents to solve specific business problems. For instance, Citigroup explained how the agents help them provide wealth management advice, while Comcast detailed the use of agents to optimize consumer services, thus illustrating the positive impact of the agents on their operations.

The Complexity of Workflows

Kurian emphasized that managing the complexity of workflows requires even smarter AI models. Agents must be capable of handling a multitude of variants and unforeseen situations without prior programming. For example, when scheduling appointments, an agent must be able to navigate among 19 different conditions without deterministic instructions. This sophistication allows models to seek optimal outcomes by following high-level instructions.

He explained that the complexity of modern workflows necessitates AI models capable of adapting to a variety of unforeseen situations. Agents must be sophisticated enough to manage multiple conditions without prior programming, which is essential for tasks such as appointment scheduling. This ability to adapt and seek optimal outcomes by following high-level instructions is crucial for maximizing the efficiency of AI agents in complex environments.