My AI setup
What I use AI for
I use AI to turn scattered notes, repository evidence, and current public signals into clear decisions and durable work. The goal is not to pour all of my context into a prompt; it is to retrieve the smallest relevant slice, verify it against exact sources, and leave the useful conclusion somewhere I can find again.
My everyday setup
- BB: I start a focused project thread, give the work a clear boundary, and let coding agents inspect evidence, use tools, and return reviewable results across local or remote environments.
- Obsidian: This is my durable, human-readable source of truth for notes, research, strategy, plans, and spatial thinking. I treat it as evidence, not as an indiscriminate prompt dump.
- Enzyme: This is the retrieval layer over accumulated work. I use exact search when I know what I am looking for, broader entity and semantic views when I need patterns, and then open the original Markdown before drawing a conclusion.
- Margins: I capture the small judgments around conversations—what mattered, what changed, what was promised, and what should resurface later—so they can become durable context.
- Repositories and tests: These are the source of truth for implementation. Notes explain intent; code and checks show what is actually true.
- Live research: When recency matters, I check current public sources and keep private context out of anything I share.
- Excalidraw: I use a spatial view when a system, plan, or decision is easier to understand as a map than as prose.
How I work
I begin with a bounded BB thread and the instructions for that piece of work. I search directly for known facts, use Enzyme when I need semantic recall or connections across notes, and then read the selected Markdown sources exactly. From there I cross-check the relevant repository, tests, or current public evidence before producing a memo, outline, implementation decision, research note, or drawing. Finally, I save the durable conclusion back to Obsidian and verify the resulting file or diff.
The principle tying it together is simple: capture human judgment, keep it inspectable in Markdown, retrieve only the context the task needs, ground agent work in exact sources and live evidence, and return useful decisions to the archive.