The second brain as context layer

The belief

Context-over-prompt is an axiom. The second brain is what makes it operational. Plain markdown, git, a kg.json ontology - every future AI session reads the same brain and writes back to it. That bidirectional link is the load-bearing word: the next session inherits the last session's compounding, without rebuilding context from scratch.

How to apply

1. Default to markdown-plus-git as the substrate, not a dedicated app. Notes apps lock you into a platform's schema. Markdown files in a git repo are readable by any AI agent, diffable, portable, and version-controlled. The substrate should outlast any single tool.
2. Wire the brain to every AI session by default. The second brain only compounds if it is present at the start of each session. Set it as system context, not something you manually paste when you remember. A context layer you reach for manually is not infrastructure - it is a habit that will break.
3. Treat writes-back as the compounding mechanism. A second brain that only reads is a static doc. The value is in the writes: when a session produces a decision, a belief update, or a new pattern, write it back. The brain should be measurably more useful in session 100 than in session 1.
4. Use the ontology (kg.json) to solve the retrieval problem, not just the storage problem. Raw markdown accumulates fast. Without a typed ontology - themes, beliefs, people, decisions, projects as typed nodes with typed edges - retrieval degrades to keyword search. Structure the graph so agents can traverse it, not just grep it.
5. Audit maintenance burden quarterly. The reason second brains failed before was the maintenance overhead. AI reduces but does not eliminate that burden. If the brain has not been written to in 60 days, it is not functioning as infrastructure. Prune, refresh, re-index. Static context decays.

What this is not

• Not a productivity system. A productivity system is a workflow manager: tasks, deadlines, tracking. The second brain is a context substrate: beliefs, decisions, patterns, structured knowledge. The two can coexist but should not be conflated. Treating the brain as a GTD tool is the failure mode that killed PKM for most people.
• Not a one-time setup. The ten-minute seed is the start, not the deliverable. The deliverable is a live context layer that accumulates over months. A brain seeded once and never touched again decays to zero within a few sessions.
• Not a substitute for taste. The brain stores what you know and how you think. It does not replace the judgment call about what matters. Taste-over-execution applies here: a well-structured brain with bad taste produces well-organized noise.

Argues against

• "A personal knowledge base is a nice-to-have for knowledge workers, not infrastructure for AI use."
• "Prompt engineering skill is sufficient - you do not need persistent context across sessions."
• "Second brains failed in the PKM era and the AI era just makes them a shinier version of the same problem."

Where to go from here

If you want the upstream axiom that this belief operationalizes, read context-over-prompt. The second brain is the artifact that makes "context is the lever" testable rather than theoretical.

If you want the parent theme that situates the second brain in the broader agent-first stack, go to the second-brain theme. It holds the full architecture: wiki, kg.json, /enter v3.

If you want the adjacent belief about keeping the underlying substrate honest - curating what goes in rather than accumulating everything - read self-instrumentation. The retrospection practice that populates the brain is as important as the brain itself.