Share with your CTO
WSO2 is betting that the agent governance layer will be infrastructure, not a feature, and it’s moving to own that layer before the hyperscalers do. The company this week released WSO2 Agent Manager under the Apache 2.0 license, a control plane for governing AI agents across any framework, model, or deployment environment. Gartner puts the average Fortune 500 firm on track for 150,000 agents by 2028, while only 13% of organizations believe their current governance is adequate. Agent Manager covers identity, guardrails, observability, sandboxed runtime, and full lifecycle management in one platform, and it’s deployable self-hosted or as managed SaaS.
What this means for your business
The sharpest indicator of where your organization sits on this is whether your agents today share credentials with your service accounts. Most do, because agent identity was never designed for agents. The recurring failure mode looks like this: a team ships an agent quickly, wires it to an LLM and a few tools using existing OAuth tokens, and then nobody can answer who authorized it to access what, or shut it down cleanly when something goes wrong. If that pattern is live in even one production environment, the sprawl problem is already yours, not an abstract 2028 concern.
WSO2’s core architectural argument is that governance should be separated from agent logic entirely, the way a network firewall doesn’t care which application generated the traffic. That’s a defensible position, and it mirrors how API governance matured over the last decade. The companies that solved API sprawl by buying gateways bolted to specific vendors eventually rebuilt from scratch when they needed to swap infrastructure. WSO2, pitching from its twenty-year open-source integration pedigree, wants to be the vendor-neutral layer that survives those swaps. The risk in that framing is that it assumes framework heterogeneity persists. If enterprises standardize on one agent runtime the way many standardized on Kubernetes, a universal control plane loses its core value proposition.
The identity piece is where this becomes a genuine architecture call, not a tooling preference. WSO2 co-authored the OAuth 2 extension for Model Context Protocol (MCP, the emerging standard that lets agents call external tools) with the OpenID Foundation, which means their governance model is built on standards they helped write. That’s either a competitive moat or a conflict of interest depending on whether those standards achieve broad adoption independent of WSO2’s commercial interests. The leading indicator to watch is whether competing control plane vendors, named in the same Forrester landscape report, adopt the same OAuth-MCP extension or fork their own. Convergence validates the standard; divergence means WSO2 bet on a standard it controls, which is a different product than the open foundation it’s advertising.
Concept deep-dive: Agent control plane
An agent control plane is the management layer that sits above AI agents and handles everything except what the agent actually does, think of it the way an air traffic control tower relates to the planes it guides. It enforces identity (which agent is this, what is it allowed to touch), applies guardrails like rate limits and data masking, provides observability so you can see what agents are doing in real time, and gives operators a kill switch. Without it, each agent team builds their own version, inconsistently.
Based on reporting from WSO2 Agent Manager Brings Sovereign AI Governance to Enterprise Agent Sprawl – IT News Africa, originally published 2026-09-20 16:26:00.
