Agentic AI frenzy rises as vendors stake claims in workspace and infrastructure

At a glance:

• OpenAI launched workspace agents in ChatGPT, while Microsoft added hosted agents to its Foundry Agent Service.
• Google updated its Gemini Enterprise app and launched the Gemini Enterprise Agent Platform to build, scale, govern, and optimize agents.
• Anthropic’s Claude Managed Agents entered public beta as a suite of composable APIs for building and hosting cloud-hosted agents.

OpenAI, Microsoft, and Google push agentic layers into enterprise workflows

The AI agent introduction frenzy continued at a torrid pace this week, with OpenAI launching what it called workspace agents in ChatGPT and Microsoft adding hosted agents to its Foundry Agent Service. Both launches arrived on the same day that Google updated its Gemini Enterprise app to provide new ways for office workers to build, manage, and interact with AI agents, and separately unveiled the Gemini Enterprise Agent Platform, which the company said is designed to build, scale, govern, and optimize agents.

This wave of announcements follows Anthropic’s early April introduction of Claude Managed Agents, a suite of composable APIs for building and hosting cloud-hosted agents, which is now in public beta. In its announcement, OpenAI said, “workspace agents are an evolution of GPTs. Powered by Codex, they can take on many of the tasks people already do at work—from preparing reports, to writing code, to responding to messages. They run in the cloud, so they can keep working even when you’re not. They’re also designed to be shared within an organization, so teams can build an agent once, use it together in ChatGPT or Slack, and improve it over time.”

Infrastructure and control planes sharpen the enterprise pitch

Microsoft, meanwhile, stated in a blog that its latest move “brings agent-optimized compute and services designed for production-grade enterprise agents.” After its preview of hosted agents last year at Microsoft Ignite, the company said, “this refresh is a fundamentally different experience: secure per-session sandboxes with filesystem persistence, integrated identity, and scale-to-zero economics.”

Jason Andersen, principal analyst at Moor Insights & Strategy, said, “these four announcements are connected, as the frenzy around agents continues. What OpenAI is announcing is the native ability to support the creation and sharing of agents.” He noted that this is new functionality for OpenAI, which is a bit late to the game; Google, Microsoft, Anthropic and others have had this capability for some time, and are in fact moving farther ahead with these other announcements. “What we are seeing with Anthropic and Microsoft is that, as agents become more powerful, they will go to great lengths to solve the problem they are posed with, and sometimes that includes the agent writing code and doing other tasks,” he pointed out. “This increases complexity and concerns about agents and models being well managed while running. The hosting options both of these vendors provide are a more advanced infrastructure for agents to run.”

Delivery models and target users begin to diverge

Right now, he added, “many agents are being treated as simply a more advanced front end. These newer options provide the ability for an agent to do things like spin up a dedicated container, and they can support semi-autonomous and, in some cases, autonomous operations. These two announcements are more infrastructure-related, whereas OpenAI is more about agent building.” He described the Google launch as being “something in between.”

He noted, “OpenAI’s announcement is very similar to last year’s announcement of Gemini Enterprise from Google. This year, Google took steps forward to enable a management control plane for agents called Gemini Enterprise Agent Platform, which enables a much richer sharing experience and a number of management and governance capabilities.” On the whole, Andersen said, “the agent space is getting very hot, and some who have been later to the party are getting on board, and those who have been investing are evolving to provide end customers more scale, operations, and security capabilities.”

Hyperscalers and AI startups pursue distinct positioning

Brian Jackson, principal research director at Info-Tech Research Group, said that with the flurry of announcements “we’re seeing a race to gain critical mass as the agentic platform becomes the daily work interface for the enterprise. Anthropic and OpenAI are coming at it from their AI startup positioning, while Google, Microsoft, and Amazon are leveraging their entrenched hyperscaler and enterprise platform positions.”

Jackson pointed out that the differentiation in what these tech firms offer is most clear in who they are targeting and their delivery model. He noted that OpenAI’s Workspace Agents are designed for non-technical business teams. They provide templates for agents that can automate tasks from lead scoring to vendor research reports. Users can “prompt” their way to work automation without worrying about the behind-the-scenes mechanics—what model is being used, what APIs are being called, how data is retrieved and written, or how permissions are granted.

Composable APIs and model-agnostic platforms split developer strategies

Anthropic is taking a different approach, he said. Rather than going directly to business users, it is providing tools to enterprise development teams to build their own agents and provide a custom interface to their users. Anthropic’s Managed Agents are a group of composable APIs that developers can use. The approach is more flexible, but it requires more effort to produce value.

Microsoft and Google, on the other hand, are both vertically integrated platforms providing an agentic layer on top of an extensive stack. Microsoft’s Foundry is similar to Anthropic’s offering, but offers even more flexibility by remaining model-agnostic and allowing developers to choose their preferred agentic framework.

Observability, cost, and lock-in risks rise with agentic workflows

As the agentic platform market develops, Jackson observed, “we are seeing new problems crop up regarding observability. Detecting and observing agents will be rooted in the identity system used to provision them. However, since each platform uses its own identity system, it will be difficult for any one platform to see all agents created in an enterprise, or worse, those created by a rogue user (Shadow AI).”

Furthermore, he added, “agentic workflows imply significantly higher AI token consumption to complete work. We are already seeing AI capacity constraints and price increases due to high demand. Because agents require multiple ‘reasoning’ steps to complete a single task, it is very hard to predict what a workflow you automate today might cost to run one year from now.” This means that IT leaders need to decide where they will build the agentic layer of their stack. “You don’t want to get it wrong, because becoming entrenched in one platform means significant vendor lock-in,” he said. “We already worry about lock-in with systems and data, but when you add an intelligence layer, you are essentially building a brain with neuronal pathways to your workflows. It is not going to be easy to do a ‘brain transplant’ to another platform later.”