The analogy
In fitness, there's a concept called an anabolic hyperresponder. Two people follow the same program, same diet, same recovery. One gets normal results. The other puts on muscle at twice the rate. Same inputs. Different wiring.
Same thing happens with AI. Everyone has access to the same models, the same APIs, the same documentation. Most people get modest productivity gains. Some people catch fire.
I'm trying to figure out why. Not whether it's good or bad — that argument bores me. I want to know what actually changes when the collaboration starts working.
Reckless artist, ruthless editor
The dynamic that works for me: I throw out half-formed ideas, associations, weird connections between unrelated things. The AI catches them, structures them, finds the thread I was reaching for. I break rules. It follows rules. Together we cover ground neither of us could alone.
I'll say something like "this monitoring tool should work like a dog that barks at strangers but knows the mailman." That's useless as a spec. But Claude turns it into a concrete architecture where known patterns get filtered and anomalies get flagged. The metaphor carried more information than a requirements document would have.
I provide the creative leaps and the half-baked intuitions. The AI provides disciplined execution and the patience to actually implement them properly. Most people try to use AI as the creative partner. I think that's backwards — for me, anyway. The AI is the contractor. I'm the weird client with strong opinions and bad drawings.
The fit
My brain doesn't do linear. Never has. Ideas arrive sideways. I'll be thinking about database schemas and suddenly connect it to something I read about ant colonies six months ago — and that connection is actually useful, but only if someone can catch it and do the disciplined work of turning it into code.
That's the ADHD pattern-matching thing. You see connections between distant concepts faster than most people. The trade-off is you can't always execute on them because your attention is already somewhere else. AI fills that gap. I spot the pattern. The model builds the bridge. Before AI, half these connections evaporated because I'd move on before finishing the implementation.
And when the dynamic is clicking — when the collaboration is producing enough novelty and feedback — I can go six hours without noticing. A solo coding session might lose me in 45 minutes. An AI-assisted session where we're building something new? I forget to eat. Each interaction is a coin in the machine. Prompt, response, evaluate, adjust. The cycle is fast enough to keep the attention system engaged and slow enough to require actual thinking.
I'm not claiming ADHD is a superpower. It's mostly a pain in the ass. But the specific deficits — linear execution, sustained attention on routine tasks, finishing things — map almost perfectly onto what AI is good at. The fit is complementary dysfunction.
I have a 600+ game Steam library. Haven't touched it in months.
What shifted
The usual worry at this point is dependency. What happens when the API goes down? Can you still write a for loop?
I find this boring. We have clothes, shoes, central heating. Every tool we adopt trades some capability for another. The question was never "are you dependent?" — of course you are. The question is whether what shifted was worth shifting.
Building software is largely syntax and convention. We reuse existing techniques and standardize. Most of what a developer does on any given day is respecting the grammar of systems other people designed. That's the part AI is eating, and I don't think it's the part worth protecting.
The hard part — the part that was always hard — is knowing what people want when they don't know what they want. Translating a vague business need into a system that actually solves it. Reading a room. Hearing the thing behind the request. That work didn't move to the AI. If anything, it moved closer to the center, because when the implementation grunt work is handled, the only thing left is the thinking.
So the redistribution looks like this: I used to spend 70% of my time on syntax and convention and 30% on the actual problem. Now those numbers are closer to reversed. The AI handles the boilerplate, the formatting, the "how do you configure nginx to do X" lookups. What's left is deciding what to build, why, and what to throw away. Design. Intent. Judgment.
That's not dependency. That's leverage. But — and this is the part that keeps it interesting — I can't always tell the difference from the inside.
The interesting question
The hyperresponder framing isn't really about productivity. It's about what happens to thinking when you redistribute cognitive labor. Some brains adapt to the new arrangement faster than others. The ones that were already wired for pattern-matching and weak on linear execution — the ADHD brains, the lateral thinkers, the people who see the architecture but can't sit through the implementation — those are the ones where the redistribution hits hardest. Not because they're smarter. Because the thing that was bottlenecking them just moved.
Same tools. Different results. The difference is which cognitive bottleneck the tools happen to relieve.
That's not a hierarchy. It's a variable. And the people it helps most are the ones who need to watch most carefully — not for dependency, but for the quieter risk: mistaking the tool's competence for your own understanding. Not "can you still code without it?" but "do you still know why the system works the way it does?"
That's a different question. And it has a different answer depending on the day.