1995 · Texas's first public ISP

I built the DNS, the frame relay, the mail, and the support desk. Provisioning was physical — you handed a customer a floppy and walked them onto the internet, one at a time.

The interesting part was what came next: teaching other, larger ISPs to run provisioning as an API against their own systems. That is the first time the pattern shows up. I do it by hand, I understand where it hurts, and then I build the thing that means nobody has to do it by hand again.

EDS → Nortel Shasta 5000

Enterprise-scale infrastructure access at EDS, through the Y2K cutover from a network operations center, then network security at Bank of America — the CCIE era. That work turned into the Nortel Shasta 5000 BSN: broadband subscriber provisioning, productized, sold, and supported.

Same instinct as the floppy disks, three altitudes up. A subscriber gets connected without a human touching the path.

CNCF · the Dan Kohn years

Back in the mid-nineties, while I was building that ISP, Dan Kohn was doing one of the first secure commercial transactions on the internet. We did not know each other. Two people building foundational internet infrastructure at the same time, twenty years before it mattered.

Dan went on to be COO of the Linux Foundation, then the first executive director of the brand-new CNCF. A few months in, needing people who could execute on his vision, he asked his network. Bob Wise recommended me.

How the work actually arrived

Dan would say: I have got a problem, can you figure it out. I would take it all the way through — research to execution. He called the team Strategic Initiatives. That was his name for it, not mine.

Cross Cloud

Intel had donated a thousand-node cluster that was sitting underutilized. Dan asked whether I could deploy something on it. I was excited to get real hardware running, so that became the first demo, before I shifted to Packet because that system was far easier to work with.

Once I could consistently demo across multiple clouds, we realized cross-cloud was the real story. Cross Cloud is the name we coined and ran with.

It grew well past where it started — most major clouds plus OpenStack, IBM and Oracle, testing Envoy, Jaeger and Linkerd alongside Kubernetes. I handed day-to-day operation to Denver Williams and Taylor Carpenter, and CNCF contracted their cooperative to own it long-term.

The CI dashboard

The dashboard made all of it visible, and the surprise was the direction of the pull. People got excited to fix things in public. Providers wanted in. Projects wanted in. No reporting structure over any of them, no authority to compel anyone — just visibility, and the wish to be part of something that worked.

Cloud Native Credits

That enthusiasm produced a problem: the community now needed real testing infrastructure to keep up with itself. That demand became the Cloud Native Credits Program.

I worked with Jay Pipes at Amazon to get AWS to match Google's three million dollars a year, and eventually brought Oracle to the table — their executive committing another three million out loud, in a public board meeting.

Then I stayed. Being the person who raises the donation and vanishes is the easy version. I was steward of those donations across AWS, Oracle and Packet, meeting weekly with SIG K8s Infra to review configuration changes and utilization, using tooling I wrote for exactly that visibility.

Conformance

In parallel, someone surfaced the initial coverage numbers showing how little of Kubernetes was actually conformance-tested. That finding went to the TOC in Berlin, and I stepped in to build the infrastructure and tooling underneath the entire conformance program — across every SIG, over multiple years.

This is the work I am proudest of. Not because the tooling was clever, but because "what is Kubernetes" went from a matter of opinion to a matter of evidence, and the evidence was public.

The era closed with one more conformance program, this time for Envoy Proxy, on the same underlying approach.

Recently

Still on bare metal. PXE-booting 8×A100 systems on rented cloud hardware to compare immutable-OS approaches against each other — the same provisioning lineage as Hanlon and Razor at Chef, and a netboot rack of Banana Pis in New Zealand.

And running a small cooperative's infrastructure out of an RV: a k3s cluster, a dozen DNS zones, and a hard-won respect for what happens when the uplink is a satellite.

What I am looking for

The seat where hardware meets software and somebody has to define the interface between them — then prove it holds across every environment it ships into. I have built that boundary three times now, at increasing scale, and the part I care about is the proof.