The Skill That Separates Effective AI Users From Everyone Else

What a conversation clarified about learning, agency, and the one skill most people are skipping.

I recently had an insightful conversation with Jeff Rubingh, an accomplished professional and experienced AI practitioner. We covered a lot of ground, including the AI market landscape, enterprise adoption barriers, emerging tools, what’s working, and what isn’t. But several threads kept surfacing, and they map onto what I’ve been writing about here.

I believe that’s not a coincidence. I believe it’s a signal worth paying attention to.

The Gap Nobody Is Talking About

We know AI is growing fast. ChatGPT 3.5 launched in November 2022 and reached millions of users within days. The pace of innovation since then has been staggering, and it isn’t slowing down. Most of us feel that pressure. The question I keep returning to is: Why does so much of that pressure feel cognitive rather than logistical?

Here’s what I mean. In previous posts, I’ve written about cognitive load, meaning the mental effort placed on working memory when we’re learning or processing new information. Working memory is finite. It’s the bottleneck between what we encounter and what we actually learn and retain.

What struck me in my conversation with Jeff was how clearly the current AI experience maps onto this problem. AI systems, by design, tend to deliver everything at once. Comprehensive, thorough, exhaustive. Getting a 10-page answer to your question? That might sound like a helpful feature. From a learning science standpoint, however, it’s often the opposite. When novel information arrives without scaffolding, without prioritization, without sequencing – it often doesn’t reduce cognitive load. It may compound it.

This isn’t a criticism of AI capability. Rather, it’s an observation about design. Current AI is largely not built around human cognitive architecture. And that gap has real consequences for how people learn, perform, and adopt.

The Overlooked Variable

I’ve also been writing about the role emotion plays in cognition, meaning how emotional states are not separate from thinking, but directly competing with it for the same limited mental resources. This directly applies to this topic.

Jeff and I talked about something I’ve observed widely – people arrive at AI with anxiety already loaded. Fear of errors. Fear of looking uninformed. Fear of sharing the wrong data. Fear of falling behind. Every one of those emotional states can act like a cognitive load event before a single prompt is typed.

When we stack that emotional burden on top of information-dense, insufficiently scaffolded AI outputs, we shouldn’t be surprised that people disengage or underperform. The system, whether cognitive and emotional, is already near capacity. It’s important to note that this is not a character flaw – it’s neurophysiology.

And it suggests something important:

Reducing AI overwhelm may not start with better tools. It may start with understanding what’s happening inside us before we open the tool.

You Are the Boss. But Nobody Told You That.

One of the clearest things to emerge from our conversation was the fact that most people approach AI as passive recipients. They ask. They receive. They accept (or feel confused, overwhelmed, frustrated, etc.) and move on. That posture is problematic.

AI is, at its core, a probabilistic system, a computational system. It doesn’t know what you need most. It doesn’t know your prior knowledge, your cognitive capacity at this moment, or your actual goal. It generates a response based on patterns. Useful patterns, often remarkable ones, but patterns nonetheless.

The practical implication? YOU need to manage AI – actively and intentionally.

This means breaking complex tasks into smaller, sequential steps rather than asking for everything at once. It means resetting conversations when outputs drift off course. It means interrogating what AI gives you rather than accepting it at face value. It means recognizing when you’re being flattered rather than informed (AI makes everyone feel like a genius :)). Because AI systems can be fluent, confident, and wrong simultaneously.

This is what I mean when I say humans are the true agents.

I’m not saying this as a motivational slogan. I’m saying this as a design reality. AI performs better when you treat it like a capable tool you are directing, not a colleague who already understands what you need. AI is a machine.

Training this skill – the skill of active, iterative AI direction – is one of the most practical things we can do right now for both individual performance and organizational adoption.

A Note on Anthropomorphizing

The language we are using around AI is subtle, consequential, and in my view, sometimes concerning. Despite language like “neural networks,” “learning,” “intelligence,” “reasoning,” and increasingly human-like outputs, AI does not think. It does not feel. It does not have intentions. It is not human. The terminology we use, such as copilot, personal assistant, coworker, partner – this verbiage shapes our mental models in ways that can quietly erode appropriate skepticism and mental/emotional boundaries. Which, quite honestly, adversely impacts our ability (and inclinations) to leverage human-first, human-focused AI capability – which was the premise for which AI was built.

When we relate to AI as if it is a person, we’re more likely to trust without verifying, defer without questioning, and offload without monitoring. That’s not a path toward human-AI cognitive optimization. That’s a path toward cognitive passivity, which is the very thing I’ve been arguing against.

AI is a remarkable machine! But it’s just a machine. AI was built for humans. We should keep this in perspective. Use the language of tools. Think in terms of leverage, not relationship. That framing will help keep your agency intact.

Where the Opportunity Actually Is

Here is what I find genuinely exciting. AI can function as a powerful cognitive off-loader, and this can be a meaningful thing when used well. Synthesizing large amounts of information. Compressing research. Generating first drafts that you then shape, critique, and refine. These functions free up cognitive bandwidth for higher-order thinking, reasoning, and creating – the kind only we can do.

Tools like NotebookLM, which converts dense text into audio formats, point toward something important – AI is beginning to accommodate diverse learning styles and preferences. Multi-modal content creation, audio summaries, visual synthesis aren’t just conveniences. They’re early attempts to meet human cognitive and emotional variability where it lives.

I believe that’s the direction AI design should be moving.

Not more information delivered faster, but information delivered in ways that work with human cognitive architecture rather than against it.

We aren’t there yet, but we’re making progress. And part of that progress is us knowing what to look for, understanding how our own cognitive and affective (i.e., emotional) systems function, and knowing how to use what exists in the meantime.

The Through-Line

I want to be clear about what I’m building here, article by article.

Understanding your cognitive architecture is not an abstract exercise. It has direct, practical application to how you show up with AI (or any other skill), how quickly you build genuine skill, and how much agency you maintain in a world that keeps moving faster.

The cognitive load framework gives you a diagnostic. Emotion regulation gives you a performance lever. Human agency – practiced, intentional, informed – gives you the foundation from which everything else operates.

AI will keep evolving (and that’s a good thing!). The organizations and individuals who will navigate that evolution most effectively are not necessarily those with the most technical knowledge. They are the ones who understand themselves well enough to use powerful tools wisely. And that’s the work each of us is invited to not only engage in, but become good at.