AWS launches Agent Registry
AWS added an Agent Registry inside AgentCore to let organisations register, version and reuse AI agents instead of treating them as one‑off code artifacts. A registry introduces metadata, provenance and ownership controls that make agents discoverable across teams while enabling approval and governance workflows. Treating agents as managed assets sets a practical pattern for multi‑brand platforms that need reuse without centralising all development. (techzine.eu)
Most companies building artificial intelligence agents have the same problem old software teams had with scripts: after a few months, nobody knows what exists, who owns it, or whether two teams built the same thing twice. Amazon Web Services just turned that mess into a product called AWS Agent Registry, now in preview inside Amazon Bedrock AgentCore. (aws.amazon.com) An agent registry is basically an internal app store for agent software. Amazon Web Services says teams can publish agents, tools, agent skills, Model Context Protocol servers, and custom resources into one searchable catalog instead of leaving them buried in code repositories and chat threads. (docs.aws.amazon.com) Amazon Bedrock AgentCore is the part of Amazon Web Services that runs and manages artificial intelligence agents in production. Amazon describes it as a fully managed platform for building, deploying, operating, and monitoring agents with any framework and any model. (aws.amazon.com (docs.aws.amazon.com)) The new registry is aimed at companies with hundreds or thousands of agents, where “agent sprawl” starts to look like shadow information technology. Amazon’s launch post says platform teams usually hit three problems at that scale: visibility, control, and reuse. (aws.amazon.com) Visibility means a company can finally answer simple questions with data instead of guesswork. The registry stores records with metadata like ownership, interfaces, endpoints, and usage instructions so people and other agents can find the right resource with keyword and semantic search. (docs.aws.amazon.com) (theregister.com) Control means not every experimental bot automatically becomes company-wide infrastructure. Amazon says the registry includes approval workflows and curation, so organizations can decide who is allowed to publish, what gets promoted, and what stays private. (docs.aws.amazon.com) (aws.amazon.com) Reuse is the part finance teams will notice first. If one team already built an agent that files expense reports or queries a customer database, another team can discover that asset and use it instead of paying engineers to rebuild the same capability from scratch. (aws.amazon.com) (infoworld.com) Amazon is also betting that agents will not all live in one stack. The registry supports Model Context Protocol and agent-to-agent patterns, and outside coverage notes that it is designed to work with resources hosted in AgentCore, other cloud services, or on-premises systems. (docs.aws.amazon.com) (theregister.com) That matters because the winning enterprise pattern is starting to look less like “one giant company brain” and more like a managed fleet. A central team sets rules for identity, approvals, and discovery, while product teams keep building specialized agents close to their own data and workflows. (docs.aws.amazon.com) (aws.amazon.com) Amazon Bedrock AgentCore itself only entered preview in July 2025 and became generally available in October 2025, so Amazon is adding this registry less than a year into the platform’s life. That timing says Amazon thinks the hard part of agent adoption is no longer just building demos, but keeping a growing catalog of real agents organized enough that a large company can trust and reuse them. (aws.amazon.com 1) (aws.amazon.com 2) (aws.amazon.com 3)
