If you are just typing questions into an LLM and waiting for answers, you are playing at the lowest possible level of leverage. You are treating a reasoning engine like a glorified Google search.

When you scale up to 100 or 200 AI interactions a day—like the 221 Cursor requests I logged in a single month recently—you realize that prompting is a dead end. High-leverage AI work is about system management, context alignment, and forcing the machine to uncover your own cognitive limits.

Here are five systems I use daily to shift from asking questions to directing outcomes:

1. Expanding the Johari window

The Johari Window is a psychological framework dividing knowledge into four quadrants: the Public Space, the Secret Space, Blind Spots, and the Void. Never ask the AI to simply agree with your logic. Feed it your context and ask: “Based on our discussion, what are my blind spots? What should you ask me to lift my secrets so we can uncover the void?” I use this for both therapy and programming. It forces the machine to challenge your perspective instead of mirroring it.

2. Start with Paraphrasing

Before starting any complex task, I like to get to terms with my AI system. This means I don’t just start typing. I first take the opportunity to pour out my brain in a long voice message and let the AI paraphrase it. That serves two powerful purposes. For one, I get to terms with my own thoughts and can reflect. And two, I get feedback from the AI- making sure it got the context right. Now, we can proceed from this foundation. Burning tokens on misunderstood context is a rookie mistake.

3. Persona shifts

To break a logic deadlock, shift the perspective. I constantly force the AI into roles. I will tell it: “Act as a Staff Engineer. Review this architecture.” Or, “If you were Naval Ravikant, how would you approach this risk?” I even ask it to impersonate me, based on my chat history, to see how my own biases are framing a situation.

4. Keeping the Cutoff in check

LLMs suffer from knowledge cutoff dates. To mitigate staleness, you must force the machine to prove it is online. I start research sessions by asking: “Paraphrase what date it is today, and find proof of it online.” Once established, I command it to find the newest, most current answer to my topic via direct internet research.

5. Expert feedback wrapped in an AI

When programming, I separate the “knowing” from the “doing.” Before writing any code, I ask the AI to adopt the persona of an external expert. I tell the expert to build an investigation plan for the codebase. We run the investigation to map out blind spots before a single line of implementation code is written.

The human provides the judgment. The AI provides the labor. Use these systems to scale your judgment.