WSO2 Agent Manager: a halo around the agent

Rania Khalaf is Chief AI Officer at WSO2. Luca Berton talked to her at AGNTCon + MCPCon North America 2026 in San Jose on 22 October 2026. WSO2 is building, in her words, “everything you need for your digital agentic enterprise”: agent management, securing AI traffic, agent identity, “all the way down to your APIs and your integrations”.

Agent Manager

What WSO2 has just released is the Agent Manager, a new product that brings together its work on agent identity and AI gateways. It follows an agentic application in steps.

First the traffic: all the interactions the agent has, its calls to LLMs and MCP servers, and what comes in. Then identity: who the agent is, what it can access, and how it is configured. That brings least privilege, and an identity for the agent that is separate from the humans who use it. Then observability and evals, which monitor continuously how the agent is doing. From what they show, you can stop the agent, shut it down or revoke its tokens if it misbehaves. “It’s kind of a 360 view of the agent and the scaffolding around it to make it easier, faster, safer to run.”

Evals for probabilistic systems

Luca asked about evaluation, as software moves from a deterministic world to a more probabilistic one. That is what interests her most. We have run deterministic systems for many years, with the engineering principles and architectural practices to match. Gen AI, and AI in general, bring in a probabilistic factor. That changes how the lifecycle is managed, what testing and versions mean, and how you fix a drop in performance.

Evals are how you understand how well an agent is doing in a way that takes that probabilistic nature into account, and possibly a multimodal one, when the exchanges are in natural language or something else.

Compose as you go

WSO2 Agent Manager has a free trial, fully SaaS, and the project is open source, with a GitHub repository to download and try. It is one platform with one experience, but people come to the space from different starting points. Those who mainly need to control traffic to their LLMs and MCP servers can take the AI gateway on its own; those who need agent identity can take it from WSO2’s identity platform; the whole experience starts with Agent Manager. “So you can kind of compose as you go.”

The agent halo

The most common use case is getting to production faster. Some teams are stuck in pilot because the controls aren’t in place and someone, a CISO for instance, won’t let them go further. Others have already let a lot of agents out, see the problems, and want to bring them back under control.

Rania’s name for the answer is “the agent halo”: “the things around the agent that make it safer and that will make you go much faster”. People think control slows them down. Her view is that a consistent way to define controls and guardrails, across any agent framework or runtime, saves doing it differently for every agent and every platform. “So you can actually go faster.”

More

The episode with Rania Khalaf is being produced and goes out on YouTube first. The day it was recorded: AGNTCon + MCPCon North America 2026: day one.