Welcome back to In General: one useful thing I tried, built, or figured out this week. AI workflows, productivity systems, dev tools, side-project mechanics. All road-tested before they show up here. If someone forwarded this to you, subscribe here.
I have a laptop where I do most of my work, an always-on Mac Mini running agents and scheduled workflows, and a phone and iPad for when I’m away from my desk.
They all need a default place for work to land.
David Allen calls that an inbox, the one trusted place to capture everything, so your head can let it go. I like that analogy. The difference is that my inbox has to work for more than one head.
Mine is an Obsidian vault. If you haven’t used Obsidian, it’s a note-taking app built on plain Markdown files that live in an ordinary folder on your computer. That folder is what Obsidian calls a vault. Nothing is locked in a database or a proprietary format — it’s just text on disk, which is exactly why Claude, Codex, my scripts, and I can all read and write the same files. GitHub syncs the vault between my computers, and Obsidian Sync syncs the same vault to my phone and iPad.
That one decision pays off in five ways.
1. I can open the same work anywhere
A note I write on my laptop is there when a workflow runs on the Mac Mini. A report that workflow writes is there when I open Obsidian later. Everything’s on my phone and iPad too.
The first benefit is boring: I know where the current version is.
That matters more now that work happens across several machines. I don’t want a project decision trapped on one computer, a meeting note stuck inside the app that recorded it, or a thought buried in whatever Git branch or worktree I happened to have open at the time.
2. Agents don’t start from zero
When I start a new Claude or Codex session, I point it at the vault: project notes, current backlog, past decisions, open questions. It gets up to speed by reading, the same way a new collaborator would.
I use project folders to leave that context behind on purpose. One session writes a plan or records a decision. A later session picks it up without asking me to reconstruct the whole story.
That saves me from repeating context I've already given elsewhere.
3. The handoff survives the chat window
Chats are where the work happens. The vault is where the work lives. By the time a session ends, whatever matters has already landed somewhere the next one can find it.
This article is an example. The source notes, interview, outline, and draft all sit together in the vault. Different agents can work different stages without sharing a chat history, and I can see what each one leaves behind in the same place I edit the piece.
Client work and product work go the same way. The durable output might be a decision, a report, an updated issue, a set of open questions. Different format, but the same approach.
4. One file moves between me, an agent, and another system
My engineering backlog used to live in Linear. Now it’s Markdown files with a little metadata.
I see those files as an Obsidian Base, a spreadsheet-style view over a folder of notes. An agent sees them from the command line. The dev workflow can take one issue, create a worktree, write a plan, build the feature, run the tests, open a pull request, and update the issue when the work is done.
Issues for various projects I'm working on go through that loop. I reviewed the issue and the plan. The workflow built the feature. PRs are merged with security, lint, and tests all green, and cleanup marked the issue as done.
The issue never got translated into a prompt or copied into another project tool. I read the same file the agent used to do the work.
5. I can see what changed
A shared place only works if I can trust what lands there.
Git gives me history. Obsidian Bases give me views across projects. AGENTS.md, CLAUDE.md, and various skills tell agents where files belong, what front matter they need, and what they’re allowed to touch. Linting and cleanup catch some of the times they get it wrong anyway.
It’s also why I want workflows to write back instead of just reporting success in chat. An updated issue, a processed meeting note, a generated report, those are things I can check. They’re also more useful to the next agent than just some claim that the cron job finished.
The vault is the default, not the whole system
I still use specialized tools. Granola records meetings. GitHub runs CI and holds the pull requests. OmniFocus manages tasks. Glance watches my systems. They’re all better at their jobs than just a folder of text files would be.
The vault holds the part that has to survive the handoff: context, decisions, current state, and output I can review. Machine state and other operational noise stay local.
None of this is particularly clever. It’s a folder of text files with sync on top. But it means whoever shows up next, me on my phone, a workflow on the Mac Mini, a fresh Claude session, starts from what’s already there instead of from zero.
One place for the work to land.