Google is going all-in on agentic AI. At Google’s Cloud Next conference on Wednesday, the tech giant announced a slew of updates aimed at helping its enterprise customers automate their business processes with AI agents.
Google Cloud reports that 75% of its customers use AI in their businesses, but given how inescapable AI has become in Google products like Docs, Sheets and Gmail, that high figure isn’t surprising. But Google is still doubling down, writing in a blog post that it’s chasing the idea of the “agentic enterprise.”
Agentic AI is the tech that runs agents, or bots, that can autonomously complete tasks, with little human babysitting needed. It’s transforming AI, particularly for coding and workplace tasks. Tech companies have embraced agents this year, placing big bets that tools like OpenClaw, Claude Code and OpenAI’s Codex will fulfill AI’s promise of automating big swaths of tasks. Now, Google is giving its business tech the same agentic makeover.
Google Cloud CEO Thomas Kurian told reporters that agentic AI is where the company sees AI tech going in the future. The company focused this year’s updates on making sure customers have AI processes that are secure, connected to internal systems and “optimize performance, scale and cost of how agents run,” he said.
The Gemini enterprise agent platform is the new, behind-the-scenes tech that businesses can use to oversee all their AI agents. Employees can create and use agents through the Gemini enterprise app, which includes a new agent designer that can be used to schedule tasks to run across different applications.
A full rack of the TPU 8I, meant for inferencing.
Google is also announcing two new, eighth-generation TPUs, the 8T and 8I. These are not your mother’s computer processors; they’re advanced chips solely meant for tech companies doing heavy-compute tasks, like developing AI.
The 8T chip is meant to make training more efficient, with Google saying it has three times the processing power compared with Google’s seventh-gen Ironwood. The 8I is specifically for inference. Google says it has an 80% improvement in the amount of memory you get with SRAM, and it has about 11,152 chips in a single system.
