Working with AI — Thoughts & Reflections

What Makes Someone Good at Working with AI?

What makes AI outputs good? What does a person need in order to achieve great outputs? Is this something anyone can do, or is there an underlying skill set that separates people who get mediocre results from those who get exceptional ones?

The Human in the Loop

When I get really good results from AI, I feel more like a human in the loop than a passenger. I've explained everything to the best of my understanding and knowledge. I've asked it to research what I don't know. I've been clear with my intent and trusted that someone with the knowledge to execute could complete the task I've given them — the way you'd brief a capable colleague and then get out of their way.

Practically, this means breaking work down into manageable pieces. Applying software engineering methodology — sprints, incremental delivery — rather than trying to conjure something complex in a single pass. Then manually testing and refining. Tweaking a thing or two where necessary. The human stays in the loop at every stage.

Conversely, I've had projects where I placed too much reliance on AI, and the results spoke for themselves. Things became overcomplicated, over-engineered, and I didn't understand what was going on. The work suffered because the human thought was absent from the process.

Good Things Take Time

You cannot create something complex in one day. It requires human thought to iterate on an idea, and sometimes the inspiration for that next iteration comes from strange, non-work related places. Maybe it's playing golf, walking in nature, going for a surf. I bet many people have had the "aha" shower moment.

This is a critical insight: the bottleneck is not AI's speed — it's the pace of human understanding and creative thought. AI can generate fast, but the quality of what it generates is bounded by the quality of the thinking that directs it. Speed is the tool's strength. Knowing when to slow down is yours.

Open-Endedness: The Ironic Key

So what actually makes someone good at working with AI? I think it comes down to a topic I dearly love from Ken Stanley: Open-Endedness.

Ken speaks about open-ended systems — systems that achieve outstanding results with no clearly defined objectives. To understand what he means, consider the two most powerful examples we already have.

Evolution is the ultimate open-ended system. It has no goal. No objective function. No grand plan. It doesn't "try" to create eyes or intelligence or consciousness. It simply explores, iterates, and selects — and from that process, over billions of years, the most staggering complexity emerges. No designer sat down and said "make a human." Humans happened because the system was open enough to wander there. Evolution's genius is that it never needed to know where it was going.

Then consider a baby. A baby has no goals, no KPIs, no five-year plan. They are pure open-endedness in action. They reach for things, taste things, fall over, try again. They learn to walk not because they set an objective to walk, but because they are endlessly, genuinely curious about what happens next. A baby doesn't optimise — they explore. And from that exploration, something as extraordinary as language, movement, and understanding of the world emerges. Every single one of us started this way.

This is what Ken Stanley is pointing at. The most remarkable outcomes in nature don't come from rigid goal-chasing. They come from systems that are free to explore without a fixed destination.

This rang true for me, and a certain peace settled in. I'm 27 years old. It is very normal to not know what is going on. You don't need lofty objectives that you chase down. You just need an open mind and an open heart. Be receptive to the things that life blows in your path.

At the time of listening to Ken, he was working at OpenAI in the Open-Endedness lab. And here's the irony: I believe that the very thing researchers think will take AI to the next level — incorporating that natural human open-endedness, the same quality that drives evolution and that every baby is born with — is actually the same thing that makes a person incredibly good at working with AI right now.

There is no playbook for this. We are in uncharted territory. The people who thrive may be the ones most comfortable with not knowing the destination — the ones who can work the way evolution works, the way a baby learns. Not by optimising toward a fixed target, but by staying open to what emerges.

Can Open-Endedness Be Taught?

Of all the unanswered questions, this one has a funny answer: no. Open-endedness cannot be taught. But it must be learned. And there is a world of difference between the two.

Teaching is external. Someone hands you a framework, a set of rules, a methodology. Learning is internal. It's what happens when something settles not just in your mind, but in your bones. There is a disparity between the mind and the heart — to know something is not the same as to feel it, to have internalised it.

Patrick Rothfuss captures this beautifully in The Name of the Wind. When Elodin teaches Kvothe about Naming, the lesson is not one that can be delivered in a lecture. You cannot Name the wind by memorising its properties. You have to know it — not with the waking, analytical mind, but with something deeper. The sleeping mind. The part of you that understands without being told. That's the gap between teaching and learning. A professor can point you toward the wind, but only you can learn to hear it.

Open-endedness has a similar paradox at its core. You cannot pursue it. The moment you set out to "become open-ended," you've defined an objective — and in doing so, you've contradicted the very thing you're chasing. It's like trying to fall asleep: the harder you try, the further it retreats. Open-endedness actually requires that you don't pursue it. You simply go along a journey where you remain open to things.

This is where Brandon Sanderson's The Way of Kings speaks directly to the point. The first ideal of the Knights Radiant: "Journey before destination." It is crucial. The path you walk matters more than where you end up. The moment you fixate on the destination, you close yourself off to everything the journey is trying to show you. You stop noticing. You stop being open.

And so it all ties together. Evolution doesn't have a destination — it journeys. A baby doesn't have a destination — they journey. The best work with AI doesn't come from knowing exactly where you're headed — it comes from walking the path with enough openness to recognise what emerges along the way. Open-endedness can't be handed to you in a lesson. It can only be learned by living it — by choosing the journey over the destination, by letting the sleeping mind do what the waking mind cannot, and by trusting that the uncertainty is not something to be solved but something to be inhabited.

Failure, Sensitivity, and the Open-Ended Life

Failure is not a detour from the open-ended process. It is the process.

Evolution runs on failure. The vast majority of mutations lead nowhere. Most species that ever existed are extinct. But evolution doesn't treat failure as a problem to be avoided — it treats it as information. Every dead end narrows the space of what remains, and from that narrowing, something extraordinary eventually emerges. A baby learning to walk falls down hundreds of times. They don't experience it as failure. They experience it as next.

Uncle Iroh from Avatar: The Last Airbender understood this deeply. "Failure is only the opportunity to begin again," he says. "Only this time, more wisely." Iroh doesn't romanticise failure or dismiss it. He simply sees it clearly: it's a thing that happened, and now you know something you didn't before. That's it. No drama. No shame. Just the next step on the journey.

But Iroh teaches something even more important about what makes open-endedness possible: the relationship between pride and humility. "Pride is not the opposite of shame," he tells Zuko, "but its source. True humility is the only antidote to shame." This matters enormously. Pride makes you rigid. It makes you cling to your plan, your identity, your idea of how things should go. It closes you off. Humility opens you back up. It lets you say "I was wrong" or "I don't know" without it costing you anything. And that openness — that willingness to not know, to begin again — is the very foundation of open-endedness.

In many ways, Uncle Iroh embodies what it means to be an open-ended person. He is deeply sensitive — but not sensitive as in fragile. Sensitive in energy. He feels the things around him because he is open to them. He reads the room. He meets people where they are. He knows when to speak and when to listen, when to push and when to let go. He moves through the world with a kind of quiet awareness that most people are too closed off to access.

I think this is a lot of what we refer to as social skills. The ability to talk to people, to be likeable, to connect. We treat these as soft skills — secondary, nice-to-have. But they may actually be open-endedness in its most human form. A person with strong social skills is someone who doesn't walk into a conversation with a rigid script. They listen. They adapt. They respond to what's actually happening, not what they planned to say. They are, in the truest sense, open-ended.

And this is where the idea starts to spill beyond AI and into everything. Open-endedness isn't just a strategy for working with technology. It manifests across life. In relationships — the best ones aren't optimised toward a goal, they're explored with curiosity and care. In careers — the most interesting paths are rarely the ones that followed a five-year plan. In creativity — the best ideas come when you stop trying to have them. Even in something as simple as a conversation — the ones that stay with you are the ones that wandered somewhere neither person expected.

Open-endedness is not a technique. It's a way of being in the world.

The Role of Patience in a Tool That Rewards Speed

AI is fast. Dizzyingly fast. You can generate a thousand lines of code in seconds, draft a document in moments, scaffold an entire project before your coffee goes cold. The tool rewards speed at every turn — more prompts, more outputs, more iterations per hour than any human could produce alone.

And yet.

The best results don't come from going fast. They come from knowing when to stop.

There's a temptation, when the tool is this responsive, to keep pushing. To generate one more version, try one more approach, add one more feature. The speed becomes a current, and before you know it, you've been carried somewhere you didn't intend to go. You've built something complex that you don't fully understand. You've outrun your own thinking.

Patience, in this context, is the act of deliberately stepping out of the current. It's closing the laptop and going for a walk. It's sitting with a half-formed idea overnight instead of prompting your way to a finished version by midnight. It's trusting that your subconscious is working on the problem even when you're not — especially when you're not.

The paradox is that slowing down makes you faster. Not in the trivial sense of "measure twice, cut once," but in a deeper sense: the quality of your thinking determines the ceiling of what AI can produce for you. If you haven't given yourself time to think clearly, no amount of speed will compensate. You'll just arrive at the wrong destination more efficiently.

This connects directly to open-endedness. An open-ended process is not in a hurry. Evolution doesn't rush. A baby doesn't set deadlines for learning to walk. The journey unfolds at the pace it needs to unfold. Patience is what allows you to stay in the open-ended space long enough for something genuinely good to emerge, rather than grabbing the first adequate output and moving on.

In a world that's about to get much, much faster, the people who know when to be slow may have the greatest advantage of all.

Unanswered Questions

• What is the relationship between domain expertise and AI collaboration skill?
• Are "taste" and "judgment" the real underlying skills?
• How does open-endedness relate to leadership? To parenting? To love?