At the Red Hat Summit in Atlanta, the company unveiled a new dedicated AI skills repository, shifting the focus from larger models to curated agent skills that encode decades of institutional knowledge. This Q&A explores the key aspects of this announcement.
What is Red Hat's new AI skills repository and why is it important?
Red Hat introduced a dedicated repository for agent skills, designed to equip AI agents with specific, curated behaviors rather than relying solely on massive models. This repository bundles task understanding, planning steps, and operational guardrails into reusable building blocks. The importance lies in turning generative AI from a simple chatty assistant into an orchestrator that can handle complex infrastructure tasks with limited supervision. By encoding 20 years of institutional memory—from support cases, knowledge bases, and work capabilities—the repository gives agents the ability to reason, plan, and execute actions within real Red Hat environments, all while respecting subscription, security, and lifecycle rules. This approach moves beyond generic AI to specialized, enterprise-ready intelligence.
How does Red Hat's approach differ from chasing larger AI models?
Instead of competing on model size, Red Hat focuses on productizing a new layer of agent skills and skill packs on top of its platforms—Red Hat Enterprise Linux (RHEL), OpenShift, and Ansible. The company believes that bigger models are not the key to enterprise AI; rather, it's about providing agents with curated institutional knowledge and task-specific skills. This strategy allows AI to manage infrastructure more effectively by combining Retrieval-Augmented Generation (RAG) with agentic capabilities. The result is agents that can perceive, decide, and run end-to-end workflows within enterprise policy, without the computational overhead of ever-larger language models.
What is the "Ask Red Hat" chatbot and what makes it special?
The Ask Red Hat chatbot, now live on the Customer Support Portal, is an interactive assistant trained on over two decades of Red Hat support information, knowledge bases, and work capabilities. What makes it special is that it combines Retrieval-Augmented Generation (RAG) with agentic AI, allowing it not just to answer queries but also to reason, plan, and take action. For example, it can analyze a customer's subscription status, check for vulnerabilities, and suggest remediation steps—all while adhering to existing security and lifecycle rules. This deep institutional memory ensures that the chatbot provides context-rich, accurate responses that a generic model could not match.
How do Red Hat's skill packs enable AI agents to act like seasoned administrators?
Red Hat’s skill packs are curated bundles of behaviors that teach AI agents to perform specific roles. The flagship example is a skill pack that trains agents to function like experienced RHEL subscription administrators. By wiring in Common Vulnerabilities and Exposures (CVE) data, subscription policies, and lifecycle management rules, agents can proactively monitor systems, identify misconfigurations, and execute fixes autonomously. Rather than relying on raw API access, these skill packs encode task understanding, step-by-step planning, and guardrails, ensuring the agent operates within enterprise boundaries. This transforms AI from a copilot into a trusted administrator that can handle complex, multi-step operations with minimal human intervention.
What role does Retrieval-Augmented Generation (RAG) play in Red Hat's AI agents?
Retrieval-Augmented Generation (RAG) is a foundational component, but Red Hat goes beyond standard RAG. By enabling AI agents to work with RAG-enriched Large Language Models (LLMs), the system can retrieve relevant institutional knowledge—such as past support solutions or configuration best practices—right when it's needed. However, the agents themselves can also reason, plan, and execute against real Red Hat estates, combining retrieved data with real-time system information. This hybrid approach allows for more accurate, context-aware decision-making, ensuring that actions are grounded in verified knowledge and comply with subscription, security, and lifecycle policies.
How are skill packs different from just giving AI agents access to tools and APIs?
Simply granting agents raw access to tools and APIs can lead to unpredictable behavior and security risks. Red Hat’s skill packs provide a structured alternative: they bundle task understanding, planning sequences, and policy guardrails into reusable modules. For instance, a skill pack for a RHEL subscription admin doesn't just let an agent call an API—it encodes the logic of how to check a subscription, correlate CVE data, and apply a fix while adhering to lifecycle rules. This ensures that the agent operates consistently and safely, even when tasks involve multiple steps or systems. Skill packs effectively turn institutional expertise into code that AI can reliably execute.
What is the ultimate goal of Red Hat's agentic AI strategy?
Red Hat aims to evolve generative AI from a “chatty assistant” into a full-fledged enterprise orchestrator. By combining last year's Red Hat LightSpeed (which brought AI to DevOps toolkits) with agentic AI, the company enables systems to solve problems and carry out complex tasks with limited supervision. The ultimate goal is to create AI that can perceive, decide, and run end-to-end workflows across infrastructure, all while staying within enterprise policy. This reduces operational overhead, accelerates incident response, and allows IT teams to focus on strategic initiatives. The skills repository is a critical enabler, encoding Red Hat's two decades of experience into trainable agent behaviors.