Technical
How to Build a Self-Evolving Coding Harness
A practical guide to building a lightweight coding-agent harness that stays fast, learns across runs, and promotes bounded changes through protected evaluation.
Read
Journal · Research and systems
Writing
6 articlesWritten from the work
A practical guide to building a lightweight coding-agent harness that stays fast, learns across runs, and promotes bounded changes through protected evaluation.
How I built a zero-dependency MCP server template in Zig that compiles to 131 KB, with a comptime tool registry and built-in client.
How I built a 1.2 MB package manager in Zig that installs software faster than Homebrew and supports taps and Linux packages.
How I built Bhumi, a client that improves client-side AI inference performance.
What a Rust implementation of Map-Elites taught me about performance assumptions.
Are We Still Thinking for Ourselves? Reflections on AI, Startups, and Creativity