The Autonomous Telco: AI Agents as First-Class Citizens in Cloud-Na...
summary
KubeTrace is a distributed packet-capture system with one tracer pod per node, a discovery API, and an AI agent that composes kubectl, curl, jq and tshark to diagnose 5G core faults. A 400-line skill file encodes workflows and tribal knowledge so the agent stitches AMF–SMF–UPF pcaps, correlates timestamps and IDs across three nodes, and produces a full registration + PFCP session story in tenths of a second. A cluster-wide scan over 150 targets spawns six subagents, runs ~200 tool calls and finishes in 16 minutes. Lessons: agents fail with confidence and need human review and tight guardrails.
AI-written summary · about
chapters (15)
- 0:00AI agents in telco infrastructure
- 1:24Observability in 5G core networks
- 2:48Building AI agents in infrastructure
- 3:55MCP fails for dynamic telco ops
- 4:57Designing skills instead of tools
- 5:41Real-time packet capture infrastructure
- 7:24Decoding signaling protocols with AI
- 9:10Correlating multi-node sessions
- 11:39Scaling agents for network health
- 13:23GitHub operator skill workflows
- 16:09Ad-hoc config queries with AI
- 18:14Merge request evaluations automation
- 19:59Lessons from production AI agents
- 21:43Testing with agent-driven scenarios
- 23:32Q&A
related talks
22mThe Agent Control Plane: Managing Thousands of AI Agents
06mFrom Containers To Agents: The Next Cloud Native
17mAI TechOps: Data and Infra for AI Foundation to Scale and Manage 1000s of Cohesive Agents
26mManufacturing Alerts To ReActive Agents: Self-Evolving AI Agents on K8S for Predictive…
30mAI Agents Unchained
38mAI as a Platform Engineer: Explaining Kubernetes Failures, Not Just Detecting Them
27mFrom Chatbots to Agents: Orchestrating Agentic AI on Existing Enterprise Kubernetes Infrastructure.
08magentctl: A CLI for Managing AI Agents & MCP Servers Like Infra