We can do better than filesystem as memory.

It works today for agents and personal intelligence because we had NOTHING before.

There’s a little field called data engineering that has been doing context engineering for decades.

I’ve built a lightweight ETL pipeline to process, structure, and clean all my personal data. Then I expose this to an agent with tools that allow it to search this enrich dataset.

Now I can say things like “based on my last week, what is the highest leverage task I should do this morning that over laps with both my job as a research engineer and my augmented intelligence system”.

And it said today “Ship more, you’ve built this for yourself, but you don’t know if others want to use it.”