AWS launches Agent Registry to centralize AI agent governance and combat sprawl

AWS: Agents shouldn't be secret, so we built a registry

Your agent will be pushed, filed, stamped, indexed, briefed, debriefed, and numbered AI agents should not be secret agents, at least in corporate environments. But when companies deploy software automations, they don't always have visibility into what their roboscripts are actually doing. Amazon Web Services aims to illuminate the enterprise agent ecosystem with its Agent Registry, which the infrastructure biz describes as "a single place to discover, share, and reuse AI agents, tools, and agent skills across your enterprise." AWS' Agent Registry is not the only agent registry. Microsoft has its Entra Agent Registry and an Azure Agent Registry. Google Cloud has its Agent Registry. There's an Agent Client Protocol (ACP) Registry. And there are many more third-party registries. The Register notices such things. AWS intends its registry to help make agents more visible within organizations, to provide some measure of control over agent publication and discovery, and to ensure sufficient distribution so that disparate teams aren't reinventing the wheel with redundant agents. Agent Registry is built to work with Bedrock AgentCore, AWS's platform for building, deploying, and managing agent software. It provides a central repository for storing metadata describing agents, tools, MCP servers, agent skills, and associated resources, linked to authorship info, protocol details, services exposed, and invocation instructions. It supports standards like MCP and A2A, in addition to custom schemas, and it's designed to work wherever agents are hosted, whether that's Agent Core, other cloud services, or on-premises. Adding records to the registry can be done through manual metadata submission via the AWS console, AWS SDK, or API. The registry will also fetch records automatically after being connected to a properly configured MCP or A2A endpoint. Querying the registry can be done via AgentCore Console, APIs, or any MCP-compatible client like Kiro or Claude Code. In Amazon-packaged remarks, Pete Hirsch, chief product and technology officer at Zuora, a revenue management biz, talked up the advantages of having a source of truth for agents. "This centralized approach enables teams to find and reuse existing assets rather than rebuilding from scratch," Hirsch said via AWS comms. "Standardized metadata ensures each agent and tool includes consistent details on ownership and capabilities, giving teams end-to-end visibility and accountability across the entire agent ecosystem." AWS expects that agents built using AgentCore, Amazon Quick Suite, and Kiro will be automatically indexed in the registry and that metadata from the operation of those agents will be surfaced to show what exists and how well they function. Organizations can look forward to viewing data about their agents in the AWS Resource Access Manager. The AWS Agent Registry is being offered in preview via AgentCore in five regions: US East (N. Virginia), US West (Oregon), Asia Pacific (Sydney), Asia Pacific (Tokyo), and Europe (Ireland). ®

AWS previews a cloud-agnostic registry for managing agentic fleets at scale - SiliconANGLE

AWS previews a cloud-agnostic registry for managing agentic fleets at scale Amazon Web Services Inc. is trying to make sure enterprises that embrace artificial intelligent agents and automation aren't getting lost in the curse of "agentic sprawl." The cloud computing giant is trying to prevent that from happening with the launch of AWS Agent Registry in preview today. It's a new service that acts like a centralized discovery and governance hub for AI agents, designed to help companies manage the explosion of AI agents, tools and skills across their entire cloud estates. The AWS Agent Registry is part of the broader AWS AgentCore platform, acing as a single source of truth for agents and their operations. Developers can use the Registry to share and reuse existing agents, rather than spending time rebuilding them from scratch every time they want to automate a new business process or task. In a blog post, the company explained that agent sprawl is a byproduct of the rapid adoption of AI agents by enterprises. As organizations scale up their automation drive from a few experimental bots to hundreds or even thousands of autonomous agents, they're inevitably going to struggle to keep track of them, the company said. That will mean a loss of visibility and control, and it also will make it more difficult for them to be reused to perform other tasks. Part of the problem is the siloed nature of AI development. In many organizations, there are different developer teams working on different projects, and they don't always know what the others are doing. One team might have developed a sophisticated payment processing agent, while a team in another department starts working on the same thing. Unless organizations have a centralized directory, these agents are essentially invisible, making them impossible to discover, govern or audit. The result is that teams can waste hours of time on duplicated effort. Compliance is another concern, as the lack of visibility can lead to unauthorized and unvetted agents being unleashed on business critical tasks. The AWS Agent Registry is meant to prevent all that. It stores structured metadata for each agent, tool and agent skill that an organization has built. This metadata lists who built the agent, what its purpose is, the protocols it uses and how to invoke it. What's more, it's platform-agnostic, meaning it can also index agents built on other cloud platforms and in on-premises environments. The metadata means users won't just see a list of agents, but also real-time information on where they're being used in production, their latency and uptime. With the AWS Agent Registry, the company is trying to position itself as a kind of "control plane" for agentic AI. It's similar to having a human resources department for AI agents and it should dramatically accelerate enterprise automation, enabling developers to quickly piece together pre-vetted and reusable skills and create bots for almost any task. One of the highlighted features is a hybrid semantic search capability that makes AI agents discoverable. Instead of just matching keywords, the registry relies on natural language processing to help developers find suitable agents. If someone searches for "billing tools," the Registry will understand that the "invoicing agent" is probably what they're looking for, even if the search term used isn't an exact match. Meanwhile, there are approval workflows that can aid in governance. To prevent agents from running around out of control, each one begins as a "draft" when it's first registered, and it must be vetted and approved before others in the organization can discover it. Developers can manually register each agent they build, or else just point the Registry to an endpoint and have it automatically pull in all of the technical details of each agent. Companies such as Zuora Inc. and Southwest Airlines Co. have already been using the Registry to try to get a handle on their growing agentic fleets, and they claim to have been successful. "This will prevent agent sprawl across the organization while establishing the foundation for scaling thousands of agents with enterprise-grade governance from day one," said Justin Bundick, Southwest Airlines' vice president of AI and intelligent platforms. AWS plans for the Registry to be more than just a simple library of AI agents. It wants it to become embedded into every corner of its cloud ecosystem, spanning the integrated development environments where developers create code to the business workspaces where employees interact with AI systems. In a future release, the company plans to enable automatic registration of any new agent that's deployed via its own agentic services, together with its operational metadata. To begin with, the AWS Agent Registry is being launched in five AWS regions, including US East (N. Virginia), US West (Oregon), Asia Pacific (Sydney), Asia Pacific (Tokyo) and Europe (Ireland).

AWS launches Agent Registry to centralize AI agent governance

AWS launched Agent Registry, a centralized platform for housing, building, and governing AI agents across various frameworks, according to a Thursday announcement. The offering is currently available in preview through the AWS Bedrock AgentCore console and across five compute regions. The registry stores data on agents, tools, and custom resources in a structured, searchable format. Each record includes features such as ownership, protocols, capabilities, and invocation methods. AWS aims to tackle governance and transparency challenges as enterprises increasingly scale AI agents in their operations. Enterprise deployment of AI agents is rising, placing pressure on CIOs to manage growing sprawl. AWS describes Agent Registry as a control layer addressing three critical issues in agentic AI efforts: lack of visibility, weak governance, and tool duplication across teams. The platform incorporates a hybrid search function that utilizes keywords and natural language phrases for efficient results. AWS stated, "Discovery becomes the path of least resistance," allowing teams to search by name, descriptions, and resource type before creating new tools. Other vendors are also developing AI management capabilities to standardize AI agents. In February, OpenAI launched its AI agent management platform, Frontier, which AWS will distribute exclusively as the cloud provider. Meanwhile, Anthropic recently introduced Claude Managed Agents, which offer pre-built infrastructure for developing autonomous systems. Agent sprawl has been a significant concern for CIOs, as organizations struggle to keep pace with rapid tools development. Gartner Director Analyst Autumn Stanish noted that AI agents are among the most disruptive forces for IT operations, as reported in a CIO Dive opinion article. The launch of AWS Agent Registry signifies a shift in CIOs' approaches to AI, moving from initial experimentation to a focus on organization, governance, and scalability.