Google pushes AI agents
- Google put AI 'agents' at the centre of its enterprise cloud strategy, pitching them as task-oriented software helpers. - It also unveiled eighth‑generation TPU chips split for training and inference to scale millions of agent instances. - Google argues lower-cost, agent-focused infrastructure will drive more AI-enabled products that need design for dynamic, automated interfaces. (reuters.com) (benzinga.com)
Google used its Cloud Next conference on April 22 to put AI agents at the center of its pitch to business customers. (usnews.com) At the Las Vegas event, Alphabet Chief Executive Sundar Pichai said companies are moving from chatbots to software that can complete multi-step work, and Google framed that shift as the next stage of its cloud business. (blog.google) (usnews.com) An AI agent is software built to carry out a task, like searching internal files, drafting a response, or triggering another program, with less back-and-forth from a human than a standard chatbot. Google said its new Gemini Enterprise Agent Platform is meant to help companies build and manage those systems. (blog.google) Google paired that software push with new hardware. It unveiled eighth-generation Tensor Processing Units, or TPUs, split into TPU 8t for training models and TPU 8i for running them, a design Google said is tuned for large fleets of agents. (blog.google) Google said TPU 8i is built for fast response times when agents have to reason through steps and act in sequence, while TPU 8t is aimed at training larger models from one shared memory pool. The company is pitching that split as a way to lower the cost of serving AI at scale. (blog.google) (benzinga.com) This is a business fight as much as a technical one. Google is trying to turn demand for generative AI into cloud revenue while competing with Microsoft, Amazon Web Services, OpenAI and Anthropic for corporate customers. (usnews.com) (bloomberg.com) Google has been laying groundwork for that strategy for more than a year. At Cloud Next 2025, it rolled out Agentspace features for enterprise search and custom agents, and it introduced the Agent2Agent protocol to let agents built on different systems work together. (blog.google 1) (blog.google 2) The company is also trying to build an ecosystem around the idea. Google Cloud said on April 22 that it is setting aside $750 million for partners in its 120,000-member channel to speed development and adoption of agentic AI services. (googlecloudpresscorner.com) Google’s argument is that if agents become common workplace software, companies will need both cheaper computing and new tools to coordinate thousands or millions of automated actions. Cloud Next was Google’s bid to sell both at once. (blog.google) (usnews.com)
