The Pulse
IBM Adds Cross-Platform Controls for Enterprise AI Agents
IBM’s TechXchange announcement describes new watsonx Orchestrate tools for discovering, evaluating and controlling agents across enterprise environments. A separate IBM release reported that Amazon Bedrock agent discovery and registration, along with specified evaluation features, were generally available in August.

IBM announced new controls for enterprise AI agents on October 27, 2026, adding cross-platform discovery, isolation and performance evaluation to watsonx Orchestrate. The updates, unveiled at IBM TechXchange 2026, aim to give companies one place to oversee agents built across different cloud and development environments.
Source prnewswire.com
The announcement addresses a practical management problem: organizations can have agents created by different teams and running on different platforms, while visibility and controls remain scattered. IBM’s additions cover more than inventory. They include ways to test agent behavior, limit access and inspect activity.
Watsonx Orchestrate reaches beyond IBM-built agents
IBM’s AI Gateway can discover third-party agents, models and Model Context Protocol servers on Amazon Bedrock, Microsoft and Google platforms. The company describes the gateway as an enforcement layer, intended to apply consistent oversight across connected systems. Users can also interact with third-party agents through watsonx Orchestrate chat and channels.
The product’s catalog is also changing. The catalog will present approved agents and tools alongside evaluation results, risk indicators and compliance information. For selected agents, the catalog can show performance, reliability and latency measures; for selected MCP servers, it can surface protocol conformance, tool-schema integrity and input-validation checks.
IBM is also introducing Premium Runtime, a hosting and isolation layer for agent workloads. It uses hardware-virtualized container isolation and resource containment, which the company says limits the risks that can arise when agents execute code or connect to external tools.
Testing and tracing join the control set
The expanded product adds custom “LLM-as-a-Judge” evaluation criteria, which teams can define in natural language, as well as simulated user journeys with success conditions tied to tool calls and parameters. Trace Inspector is designed to show the steps and tool interactions behind an agent’s work.
The AgentOps Agent can investigate failures, generate test cases and optimize agent performance through a conversational interface. Together, the evaluation and tracing features are intended to help teams assess agents before deployment and examine their behavior in production, rather than relying on a one-time test or a monitoring dashboard alone.
Identity and availability set boundaries
IBM’s announcement pairs Orchestrate with IBM Agent Identity and HashiCorp Vault. Agent Identity supplies identity and authorization context, while Vault manages access credentials. Orchestrate can also connect to Microsoft Entra and IBM Verify as identity providers for human and non-human identities.
“The winners in enterprise AI are changing how their business runs, not simply adding more AI to it,”
— Dinesh Nirmal, Senior Vice President, IBM Software
Availability information in IBM’s September 3 release differs by feature. It said Trace Inspector was generally available on August 17, 2026, and Custom LLM-as-a-Judge and discovery and registration of Amazon Bedrock agents were generally available on August 31. IBM’s October 27 TechXchange announcement does not specify availability dates for the broader cross-platform discovery, containment and evaluation capabilities it describes.
Sources
- Original source · Press release PR Newswire prnewswire.com
Other pages read
Companies in this story
Related stories
- IBM and Microsoft announce Imagine Enterprise agent platform
- AWS publishes seven-layer framework for enterprise AI agents
- OpenAI and Oracle Expand ChatGPT Integration for Enterprise Apps
- OpenAI Cuts Prices to Win Replit and Rippling Work From Anthropic
- OpenAI Brings Three Models and Codex to Google Cloud