Skip to content

homenotes from the workbench

Journal

Short notes from the workbench: project updates, things I learned, and links worth a comment. Lower stakes than the newsletter, more frequent.

Filter by tag

  1. I like 80% of Obsidian, so I'm building the other 20%

    Obsidian gets me most of the way; B2 is my Rust-and-Tauri take on the rest, with hybrid keyword + semantic search built in.

    • rust
    • pkm
    • tooling
    • learning-in-public
  2. Two DeepSeek models are building me a markdown editor for pocket change

    A worker model, an evaluator model, and a macOS markdown editor taking shape for about forty cents.

    • tilth
    • models
    • agents
    • ai-assisted-dev
  3. The four-hour course is delivered, demo fails and all

    A delivered course (demo fails and all), a clean rename out of lsp4agents, and a meetup demo to plan.

    • ai-assisted-dev
    • agents
    • tooling
    • teaching
  4. Measuring transcription quality is harder than comparing strings

    Exact string match punishes the very thing a good transcription model does: clean up what was actually said.

    • ios
    • models
    • tilth
    • learning-in-public
  5. Relay Notes is heading for the App Store

    A choice of transcription engines, a cleanup pass, and Relay Notes is into field trials.

    • ios
    • ai-assisted-dev
    • tilth
    • models
  6. A model too good to just kick the tires with

    Claude Fable 5 is impressive enough that I can no longer afford to just kick the tires.

    • ai-assisted-dev
    • models
    • ios
    • learning-in-public
  7. Describe the what, let the agent navigate the how

    The thing that always blocked me from building for iPhone was never the idea, it was the how.

    • agents
    • ai-assisted-dev
    • ios
    • tooling
  8. Autonomous agents need different machinery than the ones you watch

    Most coding agents assume someone is watching. The interesting problems start when no one is.

    • agents
    • tilth
    • ai-assisted-dev
  9. Give the agent the tools you already use

    We hand agents a text editor and act surprised when they reason about code like it's text.

    • agents
    • tooling
    • rust
    • learning-in-public