I've spent 14 years building and running cloud infrastructure — the unglamorous layer of pipelines, Terraform, Kubernetes clusters, and 3am pages that everything else sits on top of.

In 2026 I started building PandaStack: an open-source cloud where AI agents can safely run the code they write. Every workload gets its own Firecracker microVM — a real kernel boundary, not a shared one — and it boots in about 179 milliseconds because every create restores a memory snapshot instead of cold-booting a machine.

I built all of it alone: the agent that orchestrates VMs, the pre-allocated network namespace pool, copy-on-write forking, managed Postgres with point-in-time restore, git-driven app hosting, the SDKs, the docs, the billing. Roughly 400 Go files and 55 Terraform files later, 300+ organizations have signed up.

The part I didn't expect to enjoy: operating it. Being the only person on call for a multi-tenant platform teaches you things no architecture diagram will — that config which only exists in a first-boot script will silently disappear when a host is rebuilt, that a single transient object-storage error can kill a VM if you treat it as fatal, and that detection matters more than prevention because the failure you never see is the one that hurts.

I write about that on this site, and I take on consulting work in the same territory: platform engineering, cloud architecture, and making systems survive their worst day.

Compute & isolation

Firecracker, KVM, Linux namespaces, cgroups, userfaultfd, XFS reflink

Cloud

GCP (GKE, MIG, GCS, Cloud SQL), AWS, Cloudflare

Automation

Terraform, Ansible, GitHub Actions, systemd

Languages

Go, Python, TypeScript, Bash

Data & observability

PostgreSQL, ClickHouse, Prometheus, OpenTelemetry

Product surface

Next.js, REST/OpenAPI, WebSockets, Stripe metering