The more I use AI, the more I find myself looking into how others use it. "If I just tweak the prompt a little more, I can get it under control…!" — and I can't let go. It's like pulling a slot machine lever. One click, and out comes a different result every time. This unpredictable reward feels even more stimulating. It becomes a dopamine-driven behavior.
AI should be used as a pacemaker, but running without proper pacing leads to exhaustion and collapse.
When rapid feedback and rapid context-switching come together, flow breaks. You react to what the AI produces, give it another task, open another tab in between. Then suddenly: "Wait, what was I building?" It becomes more fun to research how to use the tool well than to actually use it. You start collaborating on design and planning, but at some point you say, "Just figure it out yourself." I think that's the moment when pacing fails.
In this state, the AI holds the initiative. The AI creates first, and I become the one who merely picks.
I've had the opposite experience too. When I took my time — designing together with one AI, building it myself step by step — flow was maintained. Even when the AI gave fast answers, I had the time within me to digest and judge them.
In the end, the difference wasn't speed. It was agency.
So how do you divide what to delegate to AI and what to keep for yourself? You need criteria.
At first, I tried splitting it by "what changes and what doesn't." Fast-changing things like code and UI go to AI; slow-changing things like design and policy stay with me. It looked clean, but it wasn't enough. If you hand off all execution, your sense of judgment dulls too. You need to write code yourself to see the gaps in your design. You need to touch it yourself to feel what needs fixing.
The real criteria isn't the rate of change — it's the density of judgment.
Design, planning, communication with teammates, technical direction — these are packed with judgment, so I need to hold on to them. Execution after judgment is done can go to AI. But execution that maintains your sense of judgment — that, you keep.
So how do you build that sense of judgment?
For me, safety came when I did it first, refined it, and then told the AI: "Do it exactly like this." When I go first, I develop a sense of judgment. Using that as a baseline, I can then evaluate what the AI produces.
It's not the AI going first and me picking from options. I go first, and the AI follows. That's the only way I can tell whether something is right or wrong. Only those who've done it themselves can truly direct others.
As AI takes over more and more execution, what remains for humans is ultimately the ability to judge, communicate, and stand firm.
It starts with knowing what you're doing right now. You need to be able to sense whether you're lost in AI dopamine, whether you're doing real work, and where the initiative lies. Without this meta-cognition, you get swept away by the fast feedback loop.
And because you need to have done it yourself before you can direct others, technical expertise in your domain is still essential. Without technical knowledge, you can't even judge whether the AI's output is right or wrong. Communication — conveying design intent, aligning direction, and moving with your team — is another area AI cannot replace. Because judgment isn't made alone; it's shaped through coordination between people.
Above all, there will come a moment when you're exhausted from trying to match AI's speed, and anxious from watching others build incredible things with AI. That's when you need the strength to maintain your own pace. Running without pacing leads to exhaustion and collapse.
In the end, I believe the most important thing in the age of AI isn't technology — it's not losing yourself.