Catching the underlying current
When you stand in the ocean, the surface draws your attention. The chop, the spray, and the waves knock you sideways. But any surfer will say the real force is below. The underlying current, mostly invisible, is what truly moves everything.
I keep coming back to that image when I think about how most people are trying to get value from AI right now.
There is a version of AI productivity that looks very busy. One agent is ordering your groceries. Another is monitoring your inbox. A third is half-building your website. You are toggling between them, checking in, correcting, redirecting. You are not doing the work anymore, technically, but you are not free from it either. You are a supervisor in a very chaotic open floor plan office, and the context switching is exhausting you. That is the surface chop. It feels like progress. It often is not.
The current is different. It is closer to what a long-tenured executive assistant does. Not someone you brief every morning, but someone who has worked closely with you for years and absorbed, over time, how you think, what you prioritize, which relationships you protect, and what trade-offs you are never willing to make. They act on your behalf, not because you issued an instruction, but because they know you well enough to infer the right one.
AI is not there yet. But the groundwork for it is something you can start building today.
The concept I am borrowing here is what Nate B Jones calls the semantic work primitive. For an AI system to act usefully in your life, it needs to understand not just the task but the intent behind the task. Consider a scheduling agent that rearranges your calendar to resolve a conflict. Can it tell the difference between bumping a casual coffee and cancelling on your direct manager two hours before a meeting? The mechanics of both actions are identical. The consequences are not. A human assistant with context knows the difference intuitively. An AI without it is just pattern-matching on calendar data.
That is the gap. And closing it, gradually and practically, means building a knowledge base that belongs to you.
This is not about picking the right app. It is about developing a portable, auditable record of how you work: what you are optimizing for, which relationships carry the most weight, and what kinds of decisions you make the same way every time. Notion and Obsidian are both good tools for this. I am leaning toward formats that stay portable (Obsidian). Plain markdown files live on your machine and travel with you if a platform gets worse or goes away. That portability matters more than it might seem, because the AI tools that are best at different tasks are not all from the same provider, and you want your context to move with you.
A practical starting point is to go back through the AI tools you already use and ask each one to summarize what it knows about you — your role, your projects, your working style — and export that into a simple document. It will be incomplete. That is fine. You are not building a biography. You are building a starting corpus that can grow.
Actual productivity gains, from AI, won’t come from simply multiplying the number of tasks agents do. Real value will come when AI systems understand your priorities, constraints, and working style at a deep level. Although those AI agents are not fully here yet, you can start preparing by building a knowledge base that captures the essential context of your work (the current beneath the surface). This foundation will enable future AI to make decisions you can trust.
What would it look like to give an AI system enough context about your work that you could actually trust it to make a decision without asking you first?
Take care,
Emanuel
P.S. The concept of the semantic work primitive comes from Nate B Jones. His YouTube channel is worth your time: https://www.youtube.com/@NateBJones