kauls.ai
← All posts
May 23, 2025

From reactive to proactive: real-time network intelligence

Agentic AIAIOpsSRv6Operations

Originally shared as part of a series on LinkedIn. Read the original →

In Part 1 I described wiring Claude to a live SRv6 network through an MCP server. Here’s what actually changed in day-to-day operations.

The traditional cost of “what’s the status?”

Getting a complete picture of the lab — 18+ logical systems across two platforms — traditionally meant logging into each device, running dozens of commands, and mentally correlating ISIS adjacencies, BGP sessions, SRv6 locators, and service state. Call it 30–45 minutes, with real risk of missing an interdependency.

The agentic version

With the MCP integration, the same request — “give me a complete operational status of our SRv6 network” — returns a synthesized, structured answer in under 30 seconds: infrastructure uptime, SRv6 locator health, ISIS adjacency counts, BGP session state, L3VPN route exchange, and inter-POP connectivity.

What I measured

Across common tasks in the lab, the pattern held:

  • Network status check: ~45 min → ~30 sec
  • BGP session analysis: ~20 min → ~45 sec
  • SRv6 locator verification: ~30 min → ~1 min
  • Service health check: ~25 min → ~45 sec

Average time reduction landed around 90%, with a meaningful drop in missed relationships because the model correlates across protocols instead of relying on a human to hold it all in their head.

A caveat worth stating plainly: these are results from a controlled lab, not a production network. But the direction is unmistakable — the work moves from gathering information to acting on it.

Adapted and summarized from my original LinkedIn series.