Every few weeks someone posts a grid of mental models and the internet nods along. First Principles. Opportunity Cost. Bottleneck. Sunk Cost. They feel timeless, like physics. They are not. Most of them are economics, and economics has assumptions baked in. The biggest assumption running through the canon of decision frameworks is that building is the expensive, scarce, slow part of any system. That was safe for forty years. It is no longer safe. When agents collapse the cost of building, some models get sharper and a few invert completely and quietly start giving the wrong answer. The dangerous part is they still feel right while they do it.
The four that break
Opportunity Cost. The classic version says every yes is a no to something else. That only holds when building is scarce. When you can stand up five agent-built prototypes over a weekend, the cost of a fifth yes approaches zero. The opportunity cost has not disappeared, it moved, from engineering hours to attention and judgment. Teams still treating what do we build next as the hard question are optimizing a constraint that left the building.
Bottleneck. The model is correct, the slowest part sets the pace. The error is where everyone thinks it is. For two decades the bottleneck was engineering throughput, so we hired and sprinted and measured velocity. That is dissolving. The new slowest part is taste, problem selection, and the judgment to know what is worth shipping at all. Orgs pouring AI into the old constraint cannot understand why nothing got faster. They relieved a bottleneck that already moved downstream.
Sunk Cost Fallacy. This one does not break, it gets stronger. The model says past spend should not bear on the next decision. Fine. But when code is nearly free to regenerate, throwing work away stops being painful and becomes the cheap option. Clinging to last quarter's build now costs you twice, because you could regenerate something better in an afternoon. The fallacy is the same. The penalty went up.
Local vs Global Optimum. The frame warns that incrementalism can trap you in a good-enough version while the radically better one sits unexplored, because exploring costs too much. Read that last clause again. Agents make exploration affordable for the first time. The model survives intact. The excuse for staying local does not.
The ones that get sharper
First Principles Thinking is the whole game now. When building is cheap, the differentiated act is knowing what is actually true before you generate anything. The 5 Whys gets sharper because agents give a confident answer to the question you literally asked and are indifferent to whether it was the right one, so root-cause discipline stays the human's job. Incentives stays exactly as true, because it was never about technology.
The takeaway
A framework is inherited infrastructure, and infrastructure carries the assumptions of the era that built it. Run a Chesterton's Fence check on your own head: find which models still price building as scarce, and retire the ones that do. This week, take the last big what should we build next debate your team had, and ask: would this still be a hard decision if building it cost an afternoon instead of a quarter? If not, you were not making a product call. You were rationing a resource that is no longer scarce.