Welcome to my personal website. You may find something interesting here, especially if you are working with coding agents actively.
Does using an LLM actually improve your writing? I found no direct answer, but three concepts (anchoring, design fixation, and automation bias) help explain why the question is so hard to pin down, and give some dimensions worth judging AI-assisted writing by.
Coding agents are much less useful without credentials. Two moves — assume the leak, then make the leaked thing worthless — and what that looks like in practice with a sandbox, a local proxy, and expendable tokens.
Per-project Colima VMs on macOS, plus a host-side credential broker so the VM only ever holds inert handles instead of real secrets.
Go · Lima/Colima · macOS
A website builder for small businesses — an agent writes the site, and it ships as static pages on our own CDN.
TypeScript · Astro · Agents
A PostgreSQL wire-protocol proxy that lets you hand an agent a database connection: PII hashed or redacted, writes and big reads behind approval.
Go · PostgreSQL · SQLite
Streams the PostgreSQL WAL into graphile-worker jobs, so table changes become background tasks and the business logic stays in the tasks.
Go · PostgreSQL · graphile-worker
