Uses

Agents

Since April 2026 I haven’t opened an IDE. Agents write almost all of my code, mostly Claude Code, and my job is to give them good context, review what they produce and keep them unblocked.

• Mage, my orchestrator, takes Linear tickets through to mergeable pull requests, running agents across several machines.
• Repo instructions. Every repository gets an AGENTS.md setting out the workflow, the generation steps and the style rules, so an agent behaves the same way whichever one picks up the work. This site’s is in its source.
• Checks over trust. Tests, linters and formatters are what make delegating safe; anything an agent can’t verify, I review myself.

Terminal

zsh with oh-my-zsh, tmux, vim (for the rare file I still edit by hand), git and ssh, on GNU/Linux desktops and servers. I’ve used BSD in the past. My dotfiles, including a few useful scripts, are at WillMatthews/dotfiles.

Languages and ML

Python, Go, C++ and TypeScript most days. For vision work: PyTorch, YOLO, TensorRT and NVIDIA Jetsons. For LLM work: vLLM, LoRA fine-tuning, evals and MCP.

This site

A single Go binary with HTML templates, SQLite through sqlc and goose migrations, and styles written in SCSS. Task runs the build and generation steps. Agents helped write the code; the posts I write by hand. There’s more on the about page.