AI Agent Governance Faces OpenStack-Era Fragmentation

The market for AI agent governance tools resembles cloud infrastructure circa 2010—everyone recognizes the problem, but no standard solution has emerged, according to Boris Renski, CEO of startup Apelogic.

Renski brings relevant experience to the comparison. He co-founded Mirantis, an OpenStack distribution company that later pivoted to Kubernetes, and subsequently launched FreedomFi before starting Apelogic. His new venture builds an AI governance platform based on Nvidia's open source OpenShell runtime.

The challenge is straightforward: enterprises, cloud providers and telecommunications companies are deploying autonomous agents that interact with code repositories, customer data and business systems. These agents require isolation and constraints, but the industry has not converged on which software should provide those guardrails.

The OpenStack parallel

In the early cloud era, AWS held an early advantage while Google and Microsoft built proprietary alternatives. Open source projects including CloudStack, Eucalyptus, OpenNebula and OpenStack competed for adoption. The market eventually consolidated around clear leaders.

OpenStack itself found success in a specific niche—private cloud infrastructure for telecommunications companies and government organizations—rather than displacing AWS. The Open Infrastructure Foundation projects that VMware migrations and public cloud repatriation will quadruple OpenStack deployments by 2029.

Today's agent governance landscape shows similar fragmentation. AWS released Policy in Amazon Bedrock AgentCore in March, using its Cedar policy language to intercept and validate agent tool calls. Microsoft offers Agent 365 and Entra Agent ID for agent discovery and identity management. Google made Agent Sandbox on Google Kubernetes Engine generally available in May.

These hyperscaler tools integrate tightly with their respective platforms. Renski argues that organizations operating outside a single cloud environment—including telcos, alternative cloud providers and enterprises running hybrid infrastructure—will require an open alternative built on the infrastructure agents already use: Kubernetes.

OpenShell's approach

Nvidia introduced OpenShell at its GTC conference in March. The open source software runs each AI agent inside an isolated container or virtual machine, enforcing written policies that specify which files the agent can access, which network destinations it can reach, which credentials it can use and which AI models it can invoke.

The project has attracted support from Red Hat, which plans to integrate OpenShell into its Red Hat AI platform, and Canonical, which released an Ubuntu package in June. Cisco, CrowdStrike, Google, Microsoft Security and TrendAI announced they were building OpenShell compatibility into their security tools.

First use case: coding agents

Apelogic has identified a repeatable initial application: governing fleets of coding agents embedded in software development workflows. Engineering teams are moving beyond individual developers using tools like Claude Code, instead deploying agents directly into Jira or Linear as developer, tester and deployer roles, and into CI/CD pipelines for tasks like writing release notes and reviewing pull requests.

Atlassian made agents in Jira generally available in May for its Cloud Standard, Premium and Enterprise tiers, allowing teams to assign work items to agents including GitHub's Copilot. The risks are real—Microsoft researchers demonstrated in June that hidden instructions in GitHub pull requests could manipulate CI/CD agents into leaking secrets, according to Red Hat engineers.

Why it matters

The governance layer for AI agents will determine which organizations can safely deploy autonomous systems at scale and which vendors capture that market. The OpenStack comparison suggests proprietary platforms may ultimately dominate, with open alternatives serving specific segments. For enterprises already managing Kubernetes infrastructure, a governance layer that aligns with existing tooling could reduce complexity compared to adopting hyperscaler-specific solutions. The market remains early—Nvidia describes OpenShell as "alpha software" with multi-tenant support still in development.

Apelogic currently has seven employees and operates on funding from Renski's previous exits plus customer revenue. The company is betting that the agent governance market will follow cloud infrastructure's path toward consolidation, with Kubernetes-based open source emerging as the alternative to hyperscaler platforms.

These details were first reported by Fierce Wireless.