Google bets on AI agents

Google used its Cloud Next conference in Las Vegas on April 22 to sell companies on AI agents as the product, not just the model behind them. (reuters.com) Google Cloud Chief Executive Thomas Kurian said the company’s new Gemini Enterprise Agent Platform is built to “build, scale, govern, and optimize” agents, with tools including Agent Designer, an inbox for agent activity, and support for long-running tasks. (cloud.google.com) Alphabet Chief Executive Sundar Pichai said customers’ direct use of Google’s first-party models has risen to more than 16 billion tokens per minute, up from 10 billion last quarter, and said more than half of Google’s machine learning compute investment in 2026 is expected to go to the Cloud business. (blog.google) An AI agent is software that can carry out a job across several steps, like pulling data, making a decision, and taking an action in another system. Google has been laying the groundwork for that pitch since April 2025, when it introduced an open-source Agent Development Kit for building multi-agent applications. (developers.googleblog.com) The company is pairing that software push with more infrastructure. Google said at Next that its eighth-generation Tensor Processing Units, or TPUs, are aimed at the heavier computing load that comes with running large numbers of agents, while its seventh-generation Ironwood TPUs are now generally available. (blog.google; cloud.google.com) Google is also spending to make sure outside consultants and software vendors build on its stack. On April 22, Google Cloud announced a $750 million fund for its 120,000-member partner ecosystem to support agentic AI development, adoption, and training. (googlecloudpresscorner.com; cloud.google.com) The sales pitch is already tied to large customer contracts. Merck said on April 22 that it plans to invest up to $1 billion with Google Cloud over several years for AI infrastructure, engineers, and licenses for Gemini Enterprise. (reuters.com; merck.com) Google is also trying to make those agents useful inside the software companies already run. Salesforce and Google Cloud said on April 22 that expanded integrations will let AI agents work across both platforms, with access to data and workflows in Salesforce, Google Workspace, and Google Cloud. (salesforce.com) That leaves Google competing on a broader bundle than raw model performance: chips, cloud capacity, developer tools, partner incentives, and enterprise software hooks. At Cloud Next, the company’s message was that businesses will pay for digital workers that finish tasks, and Google wants to supply the full stack underneath them. (reuters.com; cloud.google.com)