TL;DR — Key Takeaways
- AI Agent Sprawl Demands Control: IBM is expanding watsonx Orchestrate to give enterprises centralized visibility and governance across third-party and custom AI agents.
- Security Moves Into the Agent Lifecycle: New identity, authorization, testing and observability capabilities aim to help IT teams control what agents access, execute and share.
- Orchestration Becomes Essential at Scale: As enterprises deploy thousands of AI agents, managing them securely will require more than monitoring tools and disconnected policies.
IBM this week extended its portfolio to make it simpler to safely deploy and orchestrate artificial intelligence (AI) agents accessing data via multiple backend data platforms.
Announced at the IBM TechXchange conference, a watsonx Orchestrate AI Gateway has been added to watsonx Orchestrate, a control plane that IBM developed for managing AI agents. The gateway adds an ability to discover and manage third-party agents, AI models and Model Context Protocol (MCP) servers.
In addition, a watsonx Orchestrate catalog can now be used to discover, share, evaluate and reuse approved agents and tools that can be deployed in a hardened runtime environment that IBM makes available. For select MCP servers from, for example, Box, Visier and New Relic, the catalog also surfaces MCP protocol conformance, tool schema integrity, tool description safety, input validation and default credential rejections.
Other capabilities added to watsonx Orchestrate include an ability to customize the criteria a large language model (LLM) uses to evaluate a process, and the ability to define and simulate test journeys, a Trace Inspectors tool for tracking how AI agents executed a workflow and an AI agent, dubbed AgentOps Agent, that investigates AI agents, generates test cases and automatically optimizes performance.
Additionally, IBM announced that it has integrated the instance of the Apache Flink data streaming platform that IBM provides in a Confluent Platform it acquired earlier this year with watsonx.data integration.
Finally, IBM revealed that an IBM Agent Identity control plane that brings registration, delegation, enforcement and governance to AI agents is now generally available. It enables organizations to register agents in a central inventory, capture ownership and metadata, classify risk, document approved tool connections, create audit trails and manage the lifecycle of an AI agent.
Dinesh Nirmal, senior vice president for IBM Software, told conference attendees that it’s critical for IT teams to not just be able to observe AI agents for performance, quality and cost, a goal that is only achievable when a control plane is in place to manage them. Those capabilities are critical for any organization that plans to deploy AI agents at scale, he added.
It’s not clear how rapidly organizations are moving to deploy control planes for AI agents, but at this juncture it’s more a question of when rather than if. Once thousands of AI agents are deployed the only way to effectively manage and secure them will be via some type of orchestration framework. The challenge is determining which framework will provide the best ability to manage both third-party and custom AI agents across highly distributed computing environments.
Hopefully, there will not be multiple security incidents before organizations are able to apply the appropriate controls to AI agents. At this point, the issue isn’t so much whether there are tools available to govern and secure AI agents so much as it is marshalling the resources and skills needed to secure and manage them before the sheer number of them overwhelms the current capabilities of IT teams to manage them.
