Context Engineering and the Second Brain

The second brain post did 389 reactions and about 50 DMs.

The most common question: "How do you stop the AI from forgetting everything?"

That's the real problem. And it's not a prompting problem. It's a context engineering problem.

Here's the difference:

Prompt engineering is what you do in the moment. How you phrase the request. That's short-lived.

Context engineering is what you build over time. The structured knowledge base your agents query before they do anything. That's what compounds.

Most people build their "second brain" as a flat file dump. Every note, transcript, and document in one folder, piped into the agent. The agent drowns in it. Older documents get the same weight as yesterday's ICP decision. That's broken.

What actually works:

1. Freshness hierarchy. Not all context is equal. Yesterday's call notes are worth 10x a wiki page from last year.

2. Raw sources stay immutable. Drop notes and transcripts as append-only files. Never edit them. Synthesis gets written on top.

3. Human curates, agent synthesizes. I decide what goes into the brain. The agent turns it into usable knowledge. Different jobs. Don't mix them.

4. Queryable beats comprehensive. The goal isn't to put everything in. It's to build something the agent can search and return the 3 most relevant facts out of 500 pages.

The prompts you write today are temporary. The context you engineer is the durable asset.

That's what the second brain is actually for.