Plenty of people believe that no matter how good AI gets, going deep into a subject and building real judgment will stay the thing that makes you good at using AI well. I mostly agree. But there’s a question that keeps nagging at me: could AI eventually get good enough to outstrip most people’s judgment outright, to the point where judgment itself stops being what separates one person from another? I don’t have an answer. I don’t think it deserves to be waved away, either.

Before the internet, how much you personally knew mattered enormously for how competitive you were. Once the internet arrived, getting information got a lot cheaper — and yet people who already knew more tended to be better at finding and filtering what was actually useful online. That’s hard to deny. What turned out to matter most, though, wasn’t old knowledge at all. It was a completely new kind of knowledge the internet itself gave rise to — SEO, the whole logic of how platforms and algorithms actually work. None of it existed before the internet. Nobody needed it before the internet.

This looks like a general pattern once you zoom out. Every paradigm shift seems to split into two kinds of ability: leverage on the old ability, where people who already had more knowledge or experience get better mileage out of the new tool, and a genuinely native ability that the tool creates from scratch, with no prior equivalent. How much these two overlap is impossible to calculate ahead of time — you only find out after the fact. SEO is a clean example. It has almost nothing to do with how much you already knew. It runs on an entirely new intuition for algorithms and platform ecosystems.

We’re already seeing early signals of the same pattern with AI. Prompt engineering got treated as a major new skill for a while, then lost most of its value fast as the models themselves improved. If the half-life of these “AI-native” skills is shrinking faster than SEO’s did — SEO stayed stable for well over a decade before search algorithms slowly rewrote it — that would suggest the half-life of new knowledge is compressing hard this time around.

Here’s where I think the comparison might break down, though. The internet never actually threatened judgment. It just made getting information cheap. Filtering, synthesizing, deciding — that stayed human work the whole time. AI is different, because it’s pressing on both the information layer and the judgment layer at once. It’s not just helping you find answers faster. It’s generating judgment itself: opinions, recommendations, arguments. Which means judgment might be, for the first time in a shift like this, the thing that actually ends up on the test table.

To figure out whether that’s likely, I’d rather turn to history than to current research, and the cleanest precedent I know of is chess. Chess ran the full arc, not just a snapshot: pure human judgment first, then a stretch of human-AI collaboration where, for a while, centaur chess — human plus computer — actually outperformed AI alone, and eventually pure AI overtook the collaboration entirely, squeezing the space for human differentiation down toward zero. If AI judgment ends up on the same arc, something like an oral defense — proving in the moment that you can reconstruct your own reasoning — might turn out to be a waystation, not a final destination.

Current research — productivity data from call centers, scored writing samples, whatever it is — can only measure AI as it exists today. It can’t tell you what happens once AI’s judgment keeps improving past that point. At best it’s a single point on a trajectory, and a single point can’t carry the weight of a claim about a long-term trend. Chess is different: it’s a trajectory that’s already run its full course, which puts it on entirely different evidentiary footing than research still mid-arc. That’s why, on a question like this, I trust historical precedent over contemporary studies.

I genuinely don’t know whether this will happen. But I don’t think it can be ruled out, and I think it deserves to be taken seriously rather than dismissed.

If judgment really does get flattened someday, I don’t think that’s where the story ends. I think it just pushes differentiation down to another layer. There’s a consistent pattern here: every time some ability gets commoditized at scale, competition shifts to whatever layer hasn’t been flattened yet. Once literacy became widespread, competition moved to credentials. Once credentials became widespread, it moved to prestigious schools, connections, internships. Motivation, initiative, interpersonal skill, values — even social capital and looks — were always part of the competition. Their weight is likely to keep rising.

Literacy, credentials, connections were all things you could, to some degree, still work for or strategize your way toward. Looks and social capital are different. They’re largely unearnable, often decided at birth. If competition really does sink down to that layer, it isn’t just that the axis of differentiation changed — meritocracy itself starts to come loose at the joints. Effort and judgment stop being the decisive variables, and where you started and how lucky you got carry more of the weight. Push that one step further, and what AI flattens might not be just knowledge and judgment. It might quietly erode how credible the whole idea of working your way up actually is.

This is close to the part I don’t feel entirely comfortable putting into writing. I’m not certain of it, but I genuinely think it can’t be ruled out, and part of that uncertainty is, I suspect, a fear I’m not fully willing to name. I don’t know whether there’s any real answer to it either, personal or societal, if it does turn out to be true. But avoiding the question is probably less responsible than getting it wrong. So I’m leaving it here, without a conclusion — just leaving the question on the table.