Supercharge Your Second Brain (Without Making a Mess)
For a long time I basically treated a second brain like storage.
Capture more notes. Save more links. Build a bigger vault. It looks productive until you sit down to do real work and realize you still have the same problem: too much unnecessary info, what is the next action, and where does it belong?
When you add agents, the problems get even sharper.
Agents are blazing fast. They can generate mountains of text and scaffolds absurdly quickly. If your system is vague, they will produce vague output quickly. You end up buried under an avalanche of documents with even less clarity than you started with.
The fix for me was treating the vault less like a library and more like an operating system. An operating system has a few properties that are easy to encapsulate:
- there’s a single source of truth for behavior
- inputs get routed into the right place
- active work has a stable state
- review has a clear surface and clear stop conditions In practice, that means the vault needs contracts. Agents need a shared runtime contract: what to read first, where they are allowed to write, and when they have to stop and ask. Without that, every agent run becomes a new interpretation of the vault and the vault drifts.
Projects also need stable surfaces. If a project is active, it should have a small set of notes that do different jobs: requirements, navigation, tasks, decisions, and agent rules. The point isn’t bureaucracy. The point is making the system legible for both a human and a tool.
Stage separation matters too. Not every idea deserves space in your active TODO list. Some things are incubating. Some things need a spec before you commit. Some things are writing pipeline items. When you separate those stages, your TODO list gets quieter and your vault gets calmer.
The last piece is review bandwidth. Once agents are involved, your bottleneck is no longer typing. Your bottleneck is review. A good system makes review cheap: the output lands in the right place, the change is scoped, the assumptions are explicit, and the agent stops when it should. If you build your vault around that loop, agents don’t turn it into chaos. They turn it into a faster version of a system you already trust.