Enterprises are discovering how many AI agents they already run, usually while trying to answer a simpler question: who authorized this one?
WSO2 has taken Agent Manager to general availability, an open-source platform meant to answer that once for every agent regardless of which model or framework produced it. The design separates governance from agent logic, so identity, permissions and lifecycle rules do not have to be rebuilt per stack, and it runs in cloud, on-premises or hybrid setups.
Agent identity is the core problem the release addresses. Unlike ordinary software, agents call tools, reach APIs, hand off work to each other and act on behalf of users, which creates authority questions that conventional access control was never shaped around. The platform keeps a central inventory with verifiable identity, role-based access, delegation, token exchange and revocation.
Since a June beta, WSO2 added governance for Model Context Protocol interactions and a Kubernetes-native sandboxed runtime. Lifecycle controls move agents from development through staging to production and can suspend a live one. More than 40 built-in controls cover matters such as masking personally identifiable information and rate limiting.
The bet is that oversight becomes shared infrastructure rather than a per-project rebuild, an argument likely to get louder as agent counts keep climbing.
