craft as ai moat/spring series 2 of 5
fluency is cheap. divergence is the moat.
koivisto and grassini's 2023 creativity study found the chatbots beat the typical person and never beat the best one. volume got cheap; the top of the range did not.
koivisto and grassini1, published in nature scientific reports in 2023, tested 256 people against three ai chatbots on the alternate uses task, the standard test for divergent thinking. product teams still lean on the finding.
on average, the machines won. the chatbots scored higher than the typical participant, and produced their ideas instantly. the best human ideas still matched or beat anything the chatbots produced.
the average moved. the ceiling did not. volume is what got cheap. the top of the range, which is where products actually get made, stayed put.
a breakthrough product rarely comes from idea #47 in a list of 50 variations on the same concept. it comes from a direction the rest of the list never considered.
that is why "generate 100 options and pick the best one" hits diminishing returns fast. if all 100 cluster around the same center, the best one is still the best version of the average.
raw model output has a second problem. without a named reviewer in the loop, it tends to sound reasonable while staying thin: coherent on the surface, hollow underneath. and the model almost never refuses. it does not push back on a bad direction, say an idea is weak, or admit uncertainty. it keeps producing, confidently, in whatever direction you pointed it. fluency without judgment.
a senior designer's taste is less about generating more options and more about knowing which option is actually different, and which is just a rearrangement of the same thing.
the builders winning right now have the taste to spot divergence when it appears, and the craft to make it real. ai gives you volume. you still bring the judgment about which direction is worth obsessing over.