OpenShift 5 moves Lightspeed from chat to approved agentic operations

Red Hat has detailed how OpenShift 5 turns OpenShift Lightspeed from a conversational assistant into an agent that can investigate clusters and prepare changes for administrator approval. The release separates capabilities ready for production use from more autonomous workflows that remain previews, giving platform teams a clearer line between assisted analysis and agent-led execution. Red Hat’s product engineering post was published November 9.

What is generally available

Update risk analysis is generally available in OpenShift Lightspeed. Administrators can use it to run automated pre-update checks and monitor a cluster update while it proceeds. Red Hat also made augmented troubleshooting generally available: an operator describes a symptom or anomaly in natural language, and Lightspeed traces likely causes and proposes a fix for review. Red Hat describes both capabilities and their support levels.

Lightspeed can now also generate Perses dashboards from chat as a generally available feature. The workflow uses the Cluster Observability Operator and Red Hat’s build of Perses, replacing a manually written PromQL dashboard query with a natural-language request. The announcement identifies dashboard generation as GA.

What remains in preview

Agentic updates are a Technology Preview. Lightspeed evaluates pre-update checks, identifies blockers and assembles an “agentic run” that an administrator reviews before approving the update. Autonomous troubleshooting is also a Technology Preview: the agent investigates active alerts, proposes a resolution and verifies the fix, but a person must approve the change before it reaches the cluster. Red Hat says both workflows retain human approval.

AI-assisted installation is available as a Developer Preview. It pairs an external coding agent—including Claude, Codex, Cursor or Goose—with the cluster-creator skill from the Plugin for Red Hat OpenShift to guide initial cluster setup. The post labels the installation workflow Developer Preview.

The wider architecture

The OpenShift 5 design connects Lightspeed with an MCP gateway supplied through Red Hat Connectivity Link, MCP servers, agent skills, temporary execution environments and customer-provided model endpoints. Red Hat says state-changing agents use short-lived permissions and that the gateway applies human approval and model guardrails to prompts, payloads and tool calls. Those controls are outlined in the architecture section.

For platform teams, the practical distinction is support status: update-risk checks, assisted troubleshooting and conversational dashboard creation are GA; agent-driven updates, autonomous remediation and AI-guided installation are not. Teams evaluating the previews should therefore treat them as controlled experiments rather than production automation, and preserve the approval boundaries Red Hat has built into the workflows.