AWS unveils agent registry
Amazon previewed a cloud‑agnostic registry to inventory and manage fleets of AI agents, arguing enterprises need a catalogue of what agents exist, what they can do, and where they run. The registry is pitched as a control plane for agent metadata, permissions and observability — the sort of infrastructure that turns agentic systems from prototypes into auditable services. (siliconangle.com), (thenewstack.io)
Most companies do not have one artificial intelligence agent. They have a growing pile of them: one for customer support, one for code, one for finance, and another built by a different team on a different cloud. (aws.amazon.com) Amazon Web Services is trying to solve that sprawl with a new preview service called AWS Agent Registry, announced on April 9, 2026. The service sits inside Amazon Bedrock AgentCore and acts as a private catalog for agents, tools, skills, Model Context Protocol servers, and custom resources. (aws.amazon.com) A registry is basically a company phone book for software workers. Instead of asking around to find which agent already knows how to book travel or summarize contracts, a team can search one place and see what exists. (thenewstack.io) Amazon says the registry is cloud-agnostic, which means it is meant to list agents even when they run outside Amazon Web Services. AWS said many organizations already split their agent systems across Amazon Web Services, other cloud platforms, and on-premises data centers. (thenewstack.io) Each registry entry is not just a name. AWS says records can include who published the agent, which protocol it speaks, which services it exposes, and how to invoke it. (forbes.com) That protocol detail matters because companies are starting to connect agents to other agents and to outside tools. AWS says the registry supports Model Context Protocol and Agent-to-Agent descriptors, plus custom schemas for systems that do not fit those standards yet. (forbes.com) Amazon is pitching this less as a chatbot feature and more as control-room infrastructure. The company describes the registry as a governed discovery layer, which means administrators can decide what gets published inside the organization and what other teams are allowed to find and reuse. (aws.amazon.com) This fits a bigger shift in artificial intelligence spending. Amazon Bedrock AgentCore was introduced as a platform for deploying and operating agents at scale, with services for memory, identity, tool access, monitoring, and policy controls rather than just model access. (aws.amazon.com, aws.amazon.com) The bet is that companies are moving from building one impressive demo to managing hundreds of small software workers with owners, permissions, versions, and logs. AWS told The New Stack the problem is not that agents are hard to invent anymore, but that they are becoming hard to track, share, and govern once every team starts making its own. (thenewstack.io) That is why a registry sounds boring and still ends up being strategic. In cloud computing, the winners are often the companies that build the directory, permission, and monitoring layers that turn scattered experiments into systems a large company can actually audit. (siliconangle.com, theregister.com)
