A binding constraint is the step that determines the output of a whole system. Relieve it and the system moves; improve anything else and nothing happens. In theory of constraints terms: the bottleneck sets the pace of the chain.
Until recently the binding constraint on creative and knowledge work was execution: someone had to actually make the thing, and making was slow, expensive and skill-gated. Model capability is no longer that constraint. Every team building seriously with these tools reports the same finding — output is now limited by the quality of the human direction, not by what the machine can do.
Why this distinction carries the whole argument
Saying imagination is more valuable invites agreement and changes nothing. Saying imagination is the constraint makes claims that can fail: that returns will concentrate among people who specify well, that the disciplines of specification can be taught, and that organisations optimising for execution capacity are optimising for something that is no longer scarce.
Those claims are the load-bearing spine of the imagination economy. Without a moved constraint, the phrase is just another cheer for creativity.
Specification as the new bottleneck
Specification here means more than a prompt. It is the whole cluster of acts that turn intent into something a system can produce and a human can judge: articulating the brief, choosing references, setting constraints, comparing candidates, and knowing when to stop. That cluster is what director mode names as a posture, and what the five capacities of direction decompose into trainable parts.
When specification is weak, abundant generation produces noise: more drafts, same confusion. When specification is strong, abundant generation produces leverage: the same human attention covers a wider field of options and converges faster on the right one.
How organisations misread the constraint
Common failure modes look productive from the outside. Hire more makers. Buy more seats. Mandate that everyone “use AI.” Measure success by artifacts shipped. Each of these can raise throughput while leaving the real bottleneck untouched — or while making it worse, by flooding decision-makers with options they cannot evaluate.
The diagnostic question is simple: if we doubled execution capacity tomorrow, would output of recognised-as-right work roughly double? If not, execution is not binding. Something upstream is — usually clarity of intent, comparative judgement, or permission to decide.
What relieving it looks like
Relieving a specification bottleneck does not mean writing longer briefs. It means investing in the skills and rituals that make intent legible: shared reference libraries, disciplined comparison, explicit stop rules, and roles that treat direction as real work rather than a side effect of making. The cost collapse made that investment rational. The binding constraint makes it urgent.