AI as Clarity

You want to stay inside the loop for every decision that actually matters, especially the decisions that require human intuition or taste.

For example, instead of asking AI to design an entire game system for me, I might first ask it to help me figure out what questions we should discuss about the gameplay design.

This helps because when a problem is too broad, both you and the AI start making a bunch of assumptions at once. And once those assumptions pile up, your judgment gets clouded. You don’t really know which parts of the answer you should trust anymore.

That kind of leads back to the bigger question of how you actually trust AI with the decisions it makes, which is something Lauren Tan emphasized during her SpaceX talk.Left openHow do you actually trust AI with the decisions it makes?

I think this matters even more when you’re building something greenfield. When there aren’t a bunch of existing requirements telling you what the product has to become, you have a lot more freedom to choose its direction yourself.

But that also means directionality becomes taste.

There isn’t some requirement you can point to and say, “well, this is why we built it this way.” You actually have to decide what feels right, what belongs, and what kind of product you want to exist.

So in a weird way, having fewer constraints means you should put even more effort into those decisions, not less. AI can help you explore the space, break down tradeoffs, and surface possibilities you wouldn’t have thought of. But the direction itself is still the most human part of the product.

Directing the Loop

This is why I don’t really think of what I’m doing as just prompt engineering anymore.

You’re not trying to find the perfect prompt that gives you the answer in one shot. You’re designing the conversation itself: breaking a vague problem into smaller decisions and using each answer to figure out what the next question should be.

This connects pretty naturally to human-in-the-loop and human-on-the-loop ideas. Sometimes you’re directly inside the loop, making decisions with AI. Other times you let it work more autonomously and supervise where it’s going.

There are other ways to structure this too, like dual-agent setups, reviewer agents, multi-agent workflows, and probably a bunch of frameworks that will keep emerging. But in my own work, I’ve had the most success thinking in terms of human-in-the-loop and human-on-the-loop.

A big reason is that, at least for game design, I’ve found AI is still pretty weak at identifying the very specific things that actually make something entertaining.

It can give you ten mechanics that make logical sense. It can analyze balance, find patterns in other games, or decompose a system. But deciding that one tiny interaction feels satisfying, that an evolution feels worth chasing, or that a mechanic is technically good but somehow just isn’t fun? That’s where I still find human taste doing most of the important work.Left openWhat makes one tiny interaction feel satisfying?

What matters is that you’re still the one directing the loop.

I like to think of these loops as little meetings with AI. You don’t need to attend every meeting, but you probably want to show up for the ones where the product’s direction gets decided. The more of those you stay involved in, the more the product reflects your own taste.Left openWhich meetings with AI are worth showing up for?

And as AI makes implementation cheaper and faster, I think those directional decisions become even more valuable.

From Intuition to Decision

At its best, working with AI is like good teamwork. One person says something, someone else pushes it further, and eventually you reach a level of thinking none of you would have started with on your own.

I might have an intuition about where I want something to go, but AI helps me turn that intuition into concrete requirements. In doing that, it sharpens my logical thinking and gives me more clarity in my decisions.

That’s what I mean by augmenting my thinking. AI isn’t making the decision for me. It’s helping me get from intuition to a decision I can actually explain.