Agents as Enterprise Control Plane

Google and Snowflake are remaking AI agents into managed business software, not just chatbot demos. Their pitch this week was control over fleets of agents, with security, monitoring and policy checks built in. (reuters.com, finance.yahoo.com) An AI agent is software that can take actions, not just answer prompts. In business tools, that means pulling data, triggering workflows, writing code, or handing work to another system after a user gives it a goal. (reuters.com, finance.yahoo.com) Google used its Cloud Next event in Las Vegas on April 22 to launch Gemini Enterprise Agent Platform. Google said the platform is meant to “build, scale, govern, and optimize agents,” and added an Agent Designer, an Inbox to manage agent activity, long-running agents, Skills and Projects. (cloud.google.com, reuters.com) Snowflake said on April 21 that it expanded Snowflake Intelligence and Cortex Code to become what it called a control plane for the “agentic enterprise.” The company said Snowflake Intelligence acts as a personal work agent for business users, while Cortex Code is the builder layer for developers creating and operating AI systems. (finance.yahoo.com) A control plane is the management layer that tells many systems what they are allowed to do and shows operators what they are doing. In the agent market, that means permissions, logs, approvals, orchestration, and a way to update or shut down agents without touching each one by hand. (cloud.google.com, finance.yahoo.com) That is a shift from the last year’s agent race, which centered on single assistants that could book meetings or summarize documents. Google and Snowflake are now selling the plumbing around agents: the data connections, security rules, developer tools and operating dashboards needed to run them inside large companies. (reuters.com, cloud.google.com, finance.yahoo.com) Google tied that software push to infrastructure. Thomas Kurian said Google’s first-party models now process more than 16 billion tokens per minute through direct customer API use, up from 10 billion last quarter, and said the company was also introducing eighth-generation Tensor Processing Units, or TPUs, plus new storage and networking for agent workloads. (blog.google, cloud.google.com) Google also said nearly 75% of Google Cloud customers are using its AI products, and that 330 customers processed more than 1 trillion tokens over the past 12 months, with 35 topping 10 trillion. Those numbers help explain why Google is pairing agent software with custom chips and cloud capacity at the same event. (cloud.google.com, blog.google) The other piece is interoperability, or getting one agent to work with another across vendors. The Linux Foundation said on April 9 that the Agent-to-Agent, or A2A, protocol had support from more than 150 organizations and production deployments across multiple industries. (linuxfoundation.org, opensource.googleblog.com) Snowflake’s case is that the data platform should sit in the middle of that system. It said customers including Capita, Logitech, Telenav, United Rentals and Wolfspeed are moving AI from experimentation to production on its governed platform. (finance.yahoo.com) Google’s case is that the full stack matters more: models, chips, cloud, security and workplace apps in one package. Reuters reported that Alphabet used the conference to show investors that agents are central to how it plans to make money from artificial intelligence in enterprise software. (reuters.com, cloud.google.com) The contest is no longer over which company can show the smartest single agent onstage. It is over who can run thousands of them inside a company without losing track of cost, data access, or human oversight. (reuters.com, finance.yahoo.com, cloud.google.com)