Google has made its Gemini Enterprise Agent Platform generally available to Global 2000 companies, moving from preview to commercial product with a managed multi-agent infrastructure and an open protocol that connects its agents to those built by Salesforce and ServiceNow.
The announcement came Wednesday from Google Cloud CEO Thomas Kurian, and the message was blunt: the copilot era is over. What Google is selling now is not a smarter autocomplete but a workforce of autonomous software agents capable of completing complex, multi-step tasks without a human holding their hand through each decision. The platform, called GEAP, is built around a core service named Agent Fabric, which lets enterprises deploy fleets of specialized agents , a Supply Chain Agent, a Code Refactoring Agent, a Customer Resolution Agent , each with its own tools, data access, and objectives, all of them capable of communicating with each other to get work done.
The technical numbers Google is leading with are hard to ignore. Agent Fabric reportedly cuts latency in complex query resolution by 40% compared to standard LLM wrappers, and task completion reliability sits at 92%, driven by Gemini 2.5 Flash running in the background. Those figures come from Google’s own benchmarks, so independent validation will matter, but they signal the company is confident enough in production performance to put specific claims in a press release.
The detail that will unsettle enterprise software vendors most is the Open Agent Network. Google is not asking companies to rip out their existing tech stacks and replace them with Gemini. Instead, GEAP introduces a protocol that allows Google-built agents to interoperate with agents from third-party platforms, starting with Salesforce and ServiceNow. That is a calculated move. Enterprise AI adoption has stalled in places precisely because companies are sitting on years of investment in tools they cannot abandon. By bridging those ecosystems rather than demanding loyalty to a single vendor, Google is lowering the barrier to entry and, frankly, making it harder for procurement teams to say no.
Pricing reinforces that strategy. At $295 per active agent per month, Google is undercutting what analysts had been projecting for enterprise-grade agent orchestration. This is not a premium positioning play. It reads more like a land-grab: get agents running inside as many Global 2000 organizations as possible, establish the management layer as the default, and expand from there. The per-active-agent model also gives CFOs a usage-based structure they can rationalize , you pay for agents doing work, not for capacity sitting idle.
What Google is really competing for
The broader competitive context here is infrastructure ownership. Microsoft has been building out its autonomous agent ecosystem through Copilot Studio and Azure, and OpenAI’s custom GPTs have given developers a consumer-grade taste of agent behavior. What neither has shipped at this scale is a managed orchestration layer purpose-built for enterprise operations and explicitly open to third-party agent networks. Google is trying to be the operating system underneath all of it , the layer that coordinates AI labor regardless of where individual agents were built.
That framing matters for how you read the rest of 2026. The hyperscaler AI race has shifted from who has the biggest model to who owns the deployment and management infrastructure. Model benchmarks still matter for sales conversations, but the stickier competitive moat is in orchestration: logging, monitoring, agent-to-agent communication standards, and the integrations that make switching costs real. Google is betting that GEAP, not Gemini the model, is where that moat gets built.
For enterprise technology buyers, the immediate question is whether the Open Agent Network protocol is genuinely open or open-in-name-only. If Salesforce and ServiceNow agents integrate smoothly and the standard gets adopted by other vendors, Google will have seeded something close to an industry protocol. If the integrations prove shallow or the network stays small, this becomes another proprietary platform wearing an open-source costume. Watch which additional vendors announce compatibility over the next ninety days. That list will tell you more about GEAP’s long-term trajectory than any benchmark Google publishes itself.
Also read: OpenAI is in talks to deploy up to $1.5 billion into a private equity joint venture • Google now generates three quarters of its own code with AI and the rest of the industry is watching closely • An unauthorized group has reportedly breached Anthropic’s internal cyber tool Mythos raising urgent questions about AI security from within