A quick update on my setup, because nobody asked: it’s mostly off-the-shelf, and I feel fine.
There’s a lot of talk right now about everyone building their own custom harness. I’m not entirely sold. Maybe I’m underestimating the value.
I do have a custom one at home for personal things, and I like tinkering with it. But bringing it into the workplace, where integrations matter so much, seems like a pain in the butt. Every bespoke piece is one more thing to secure, keep alive, and explain to someone else.
So at work I run boring parts, wired together:
- Obsidian as my canvas. Notes, projects, call notes, drafts: all markdown, all in one vault. All local. If a tool can read a folder, it can read my work.
- The community Terminal plugin . A terminal inside Obsidian, so I never leave the canvas to talk to an agent. Stacked windows to see changes to my work, right away.
- Claude Code in that terminal. This is where most of the actual work happens.
- Call notes sync on their own. Granola
notes land in the vault through a Granola-to-Obsidian plugin. I haven’t copy-pasted a meeting summary in months.
- I’ve tried local / open-source alternatives - stability has been a trade-off.
- Everything from the root. Claude Code runs from the top of the vault, so pulling in context is just naming a file.
- Consistent YAML front matter. Every note carries the same small set of fields. It’s the cheapest structure I know.
- Self-cleaning, self-healing skills. A handful run weekly. They tidy the vault, archive what’s gone stale, and fix what’s drifted, so I don’t have to.
- Publishing from the same place. This site, jadams.fyi , is the same markdown, moved into the website folder and pushed. Still the same boring setup as day one .
None of it is clever, and that’s the point.
A custom harness is a great hobby. At work, I’d rather run something boring enough that other people can trust it.