Google is pitching its Gemini Enterprise Agent Platform as a governance layer for businesses deploying AI agents across internal data, tools and cloud services, while OpenAI has begun charging Enterprise customers for some agent workloads.
Google launched the platform on April 22 as the successor to Vertex AI. According to the company’s announcement, future Vertex AI services and roadmap updates will be delivered through Gemini Enterprise Agent Platform rather than as a separate product. The new platform combines model access, agent development, runtime hosting, observability and controls intended for large agent deployments.

AI security operations center visualizing agent networks, identity controls, databases, analytics, and monitoring dashboards.Google’s governance pitch​

The notable additions are Agent Identity, Agent Registry and Agent Gateway. Google says Agent Identity assigns each agent a cryptographic ID, creating an auditable record tied to its authorization policies. Agent Registry is a catalog for approved agents, tools and skills, while Agent Gateway is the enforcement layer between agents and their connected tools and data sources.
That is the useful distinction for IT teams: Google is trying to make agent identity and tool access first-class cloud controls, rather than leaving each development team to implement them inside an application. The gateway also integrates with Model Armor, Google’s protection service for threats including prompt injection and data leakage.
Gemini Enterprise Agent Platform supports Google models and third-party options through Model Garden. Google says the platform provides access to more than 200 models, including Anthropic’s Claude family alongside Gemini and Gemma. Its updated Agent Development Kit offers graph-based orchestration, while Agent Runtime is designed for long-running workflows and persistent agent state.
The claims are largely Google’s own product positioning, and enterprises should still validate logging, identity boundaries, data residency and tool permissions in a pilot before allowing agents to act on production systems.

OpenAI adds a usage bill​

OpenAI’s Enterprise and Edu release notes say that, as of July 6, ChatGPT for Excel/Sheets tasks use token-based credits, and Workspace Agent runs are likewise metered for Enterprise workspaces. Charges are based on input, cached input and output tokens rather than a fixed price per run.
That changes the operational calculation for organizations that had treated agent usage as part of a flat subscription. Agent workloads can now incur variable costs based on the volume and complexity of their tasks, so administrators will need to apply usage limits, monitor credit analytics and decide which workflows justify autonomous execution.
OpenAI’s July 9 launch of ChatGPT Work also has a direct Windows angle. The company says its new ChatGPT desktop app for Windows combines Chat, Work and Codex, with Work able to use local files and desktop apps when users grant permission. OpenAI also made its ChatGPT for PowerPoint integration generally available for Enterprise and Edu customers on July 6, with deployment through Microsoft 365 and role-based controls where applicable.
For Windows-heavy organizations, the immediate decision is less about benchmark results than control-plane fit. Microsoft 365 and Entra ID shops should assess how new agent tools inherit their existing permissions, while Google Cloud users should test whether Gemini’s Agent Identity and Gateway controls actually reduce the need for custom oversight code.
Admins should inventory existing agent integrations and set budget, logging and approval controls before their next automated workflow becomes a standing production cost.

References​

  1. Primary source: Tech Times
    Published: 2026-07-19T19:11:10+00:00