← Writing

The backlog problem was never ideas

July 11, 2026

Every engineer who has shipped for a while ends up with the same graveyard: a long list of tools and side projects that sounded good the day they were captured and never got touched again. Not because the ideas were bad. Because trying any one of them meant a weekend of pure setup before you got anywhere near the interesting part.

The bottleneck was never conceiving the thing. It was the plumbing: standing up the infra, wiring one API to another, reading a config format you'd use exactly once, debugging an auth flow that had nothing to do with what you actually wanted to build. That tax is what kept most of the backlog permanently deferred against a day job, or a job search, or two small kids.

Agents change that ratio directly. Something that can read the docs, write the glue code, run the thing until it works, and tell you honestly when it doesn't, takes the setup tax that used to gate every new idea and makes most of it delegable. The idea didn't get better. The cost of finding out whether it works dropped by an order of magnitude.

This site is the proof, and so is most of what's on /projects: a Nix control plane running five machines, a dozen MCP servers wiring home, finance, and infrastructure into tools an agent can drive end to end, a self-hosted homelab that used to be three separate half-finished attempts. All of it sat in a backlog for years before this one, not because I didn't want it, because I never had a free weekend that also had the patience for the plumbing.

The backlog is still long. It's just moving now, item by item, for the first time. More of it: projects.