AI agents, automation, and what they actually cost
Network engineer, dual CCIE, 15 years of large-scale networks. I build automation and AI agents that are safe to let act, measure what they save, and write it down.
Featured
When automating triage quietly costs you money
I measured AI triage on 1500 closed GitHub issues across three real codebases. The cost case is real. But on one repo, automating would have silently mislabeled one issue in three — and the accuracy number gives you no warning.
Recent Posts
SNMP vs gNMI: poll and hope vs subscribe and know
SNMP repeatedly asks the network for state. gNMI lets a client subscribe and receive a stream of updates. Why that reversed conversation is the real difference, and why automation is easier on the side that gets told.
What a Network Engineer Needs to Know About ML Training
Training across many GPUs changes the role of the network. A network engineer needs to understand the training loop, how work is divided, and why GPUs must synchronize before they can continue.
The EVPN underlay has three jobs
EVPN gets the attention, but it moves nothing on its own. The layer underneath has exactly three jobs, and there are two very different ways to do them. I built both.
BGP for EVPN: Type 2 and Type 3
How BGP carries EVPN routes using Type 2 and Type 3 advertisements, and why even a perfect control plane depends on a healthy underlay.
Why does EVPN/VXLAN exist? The problems that broke the old way
EVPN/VXLAN didn't appear because engineers wanted more acronyms. It appeared because the old network model broke under the demands of modern data centers.