Google Cloud launches Gemini agent for enterprise work

Google Cloud's pitch rests on a shift in how work is delegated. Thomas Kurian framed it plainly: users give the agent objectives rather than instructions, delegate an outcome and come back to finished work. For that to happen, the agent has to be wired into personal workflows, systems of record and enterprise controls, which is why the launch leans so heavily on connectors to collaboration tools, ticketing systems, databases and desktop files.

The design also follows what the article calls persistent execution. A single set of memories, context and personalization carries across every communication channel, and the agent can spin out smaller coworker agents with dedicated identities, @agents.company.com emails and persistent storage limited to the context a user or team provides. Skills work the same way: Gemini ships with a global library, and users can publish custom skills to a shared company registry or convert a task into a reusable template. Google says the system spends as much time learning what users are doing as it does using tools.

Cost control is the other pillar. Google Cloud notes per-token prices have dropped by about 98% since 2024 even as enterprise AI volume has greatly increased. Multimodel orchestration and smart routing triage workloads to run on the model delivering maximum performance at the lowest possible cost, and since tracking is per project, companies can charge AI costs back to specific departments — a detail likely to matter to finance and IT teams planning budgets by team.