Solo.io this week at the KubeCon + CloudNativeCon North America 2025 conference added an open source registry for artificial intelligence (AI) agents along with a distribution of a framework created by Anthropic that makes it possible to create context-aware workflows that can be executed more deterministically by an AI agent.
Keith Babo, chief product officer at Solo.io, said the agent registry provides a means to securely publish, discover, version and share AI agents.
Agent Skills, meanwhile, provides a means to programmatically assign complex tasks to an AI agent that ensures instructions, scripts and resources are only loaded when needed. Launched last month, Agent Skills, in effect, makes it possible to convert a general-purpose AI agent into a specialized agent that can now be governed using the networking and security platform that Solo.io developed.
Collectively, these capabilities are critical because while AI agents understand the intent of a request and write the code needed to accomplish a task, there is a need to ensure that policies that limit the scope of the task that any AI agent is allowed to perform are enforced.
It’s still early days so far as adoption of AI agents is concerned but organizations are discovering that in the absence of strict guardrails an AI agent is likely to try and access any and all available data regardless of how sensitive it is. That data will then be incorporated into AI agents’ output in ways that are difficult to predict.
More challenging still, AI agents are likely to be targeted by cybercriminals that have stolen credentials will view them as a means to commandeer an entire workflow. As a result, many organizations are hesitant to deploy AI agents in production environments no matter how easily they can be built and deployed, noted Babo.
The goal, ultimately, is to make AI agents first-class citizens within a secure IT environment, he added.
Solo.io last month added an enterprise edition of an open source agent gateway it previously donated to the Linux Foundation that provides a data plane for natively implementing agentic AI protocols such as the Model Context Protocol (MCP) and Agent2Agent (A2A) framework. Additionally, Solo.io has developed Kagent, an open-source agentic AI framework specifically designed for IT environments running Kubernetes clusters.
There is little doubt that in time there will be multiple approaches to governing and securing AI agents. Solo.io is making a case for an approach that is rooted in the application connectivity portfolio of tools and frameworks it originally developed for Kubernetes environments.
Each organization will need to determine how best to govern and secure what are soon likely to be thousands of AI agents that can be combined to automate an increasingly greater number of complex tasks. The challenge is to refine, as much as possible, probabilistic outputs generated by the large language models (LLMs) that AI agents depend on to make sure they are capable of consistently performing tasks the same way every time. Otherwise, the use cases for AI agents will be confined to a narrow range of tasks where the risk to the business attached to a particular task should anything go awry is not especially high.
