Ideas
Development logs, theories, and concepts
An archive of technical writing about software, systems, product thinking, and the ideas that emerge while building in public.
You Can't Unit Test an LLM Pipeline. Try Chaos Engineering Instead.
Multi-turn AI orchestration fails in ways that no assertion will ever catch. The fix is not better metrics, but chaos engineering for LLM pipelines.
Architect v0.3: tightening the discipline loop
The latest release makes the planning loop stricter, clearer, and much harder to skip when the urge to vibe-code kicks in.
Why Architect stays desktop-first
Keeping the product local changes the trust model, the UX, and the kind of features worth building.
SQLite as the memory layer for a local tool
A small local database is often enough for the kind of continuity an AI-assisted desktop app actually needs.
Writing the build log while the product is still messy
Documenting a product before it feels finished is uncomfortable, but the notes are more valuable when the edges are still visible.