AI compresses the distance between idea and execution but not between good and bad judgment

For two decades, the binding constraints in software were hiring engineers, writing code, and shipping products. AI is dissolving all three in real time — code is generated, prototypes are instant, iteration is nearly free. But dissolving production constraints doesn’t eliminate all constraints. It reveals the one that was always there, hidden behind execution bottlenecks: judgment.

The top 5-10% of builders are now 3-5x more productive than a year ago. The median builder is up perhaps 10-20%. Both groups use the same tools and have access to the same models. The difference is what they choose to build and why. Judgment — the ability to distinguish signal from noise, to say no to good ideas in favor of great ones, to hold conviction when data is ambiguous — is the one capability AI does not compress.

This connects directly to The intelligence-to-judgement ratio determines which professions AI automates first: AI automates the intelligence layer (complex but rule-based) while the judgment layer (experience, taste, intuition) remains stubbornly human. It also extends AI automation amplifies demand for expert human judgment rather than replacing it — as execution becomes free, demand for people who make consistently better decisions intensifies rather than declining. The practical implication: when you can build anything, the question “what should we build?” becomes the only question that matters.

This is constraint theory applied to the AI transition: removing the binding constraint doesn’t eliminate scarcity, it moves scarcity to the next bottleneck in the chain. The practical test: if you removed all execution constraints from your current project and could build anything instantly, would you know exactly what to build next? If the answer is uncertain, execution speed is not your bottleneck — judgment is. At the firm level the institutional form of judgment is shape — When products and models converge, the moat is the institution — how a company attracts people, distributes authority and compounds judgment.

Thariq Shihipar makes the same point to AI engineers (“A Field Guide to Fable”, sources/thariq-field-guide-to-fable.md): “building is easier, but generating value is still hard. I think this is something we run into as AI engineers sometimes, where we think so much about the process of building and our setups… but the point is to generate value… it takes a lot of swings, it takes a lot of tries, to find the valuable stuff.” The failure mode is mistaking a shrinking build-time cost for evidence of progress. Verification is the single highest-leverage practice for agent-assisted coding tells you the code works, but nothing in that loop tells you the thing you built was worth building.