Learning in Public

How AI agent harnesses work, learned by building one in Go: a provider layer that puts many LLM APIs behind one set of types, and an agent core that runs the loop, the tools and the events. pi's pi-ai and pi-agent-core are the reference implementation, read at a pinned commit.

Building a three-node homelab on ThinkCentre M710q Tinys and a TL-SG2008 switch: Proxmox on the metal, Kubernetes in VMs, Terraform driving the VMs, Tailscale and Cloudflare Tunnel for access. Every lesson ends with a change on the real rack and a line in the runbook.

Learning to write design and architecture docs for POCs and features: the one-page design doc, ADRs, the POC recommendation, the two diagrams a doc needs, and running an async review. Every lesson ends with a real section written for a live project.

Learning the command-line tools systems engineers reach for when something is slow, stuck or unreachable: ripgrep and fd, jq and yq, ps/pgrep/lsof, dtrace and bpftrace, du/tar/lsblk, and the networking set (curl, ss, dig, nc, tcpdump, openssl). The aim is to look instead of guess.

Learning PocketBase as the backend for a new product — one Go binary wrapping SQLite, extended with custom Go code and serving a React Router frontend. The aim is a backend I fully understand and can bend, not a black box.

A step-by-step course on taking a SaaS from zero users to a real user base — learned by doing it, on a real product.

Learning analytics from scratch with PostHog — what to measure, how to capture it, and how to read the result. The premise: under every dashboard is one table of events, and every chart is a question asked of it.

Getting fluent with Go — the language, its tooling, and the mindset. The goal is idiomatic, production-ready services, not just passing syntax familiarity.

Temporal makes long-running, failure-prone processes reliable by default. You write ordinary code; Temporal persists its state and replays it on failure.