parkinsons-law ADG-0020

Parkinson's law

“Work expands so as to fill the time available for its completion.”

Origin

C. Northcote Parkinson, 1955, The Economist, on the growth of the British civil service. About people and bureaucracies, where the mechanism is that effort has no natural stopping point short of the deadline.

Why it should, or should not, apply to models

The model's analogue of time is the token budget. If the law transfers, a model given a larger budget for reasoning or output will use it whether or not the task needs it, and the extra tokens will not improve the answer and may degrade it. The mechanism would be different from Parkinson's — not self-interest but the absence of any internal signal that the task is done — which makes the transfer worth testing rather than assuming.

Standing: Untested

Standing counts claims a person has reviewed. Claims reviewed by AI alone are listed and marked, and do not move it.

Where it holds only under a condition

Claims that keep the heuristic but bound it.

Across six models and six prompting interventions on Boolean, multiple-choice and open-ended questions, explicit controls to stop reasoning — a disabled thinking mode or an instruction to answer only — do not reliably eliminate visible inference before the answer: reasoning persists even under strict controls ("thinking inertia"), and it becomes more prevalent as the answer space opens, least on Boolean questions and most on open-ended ones; so a direct answer cannot be assumed from model settings or instructions alone.
Cutting to the chaseReviewed by AI
Tests the low end of the transfer, not the high end: offered no reasoning budget, models kept reasoning visibly, more so as the answer space opened. That supports the mechanism the transfer names, effort set by something other than the budget offered, with no internal signal of when to stop. It does not test whether a larger budget is filled, which is the law as stated.

Add a case where it holds or breaks

A break is worth more than a hold. If you have seen this fail for a model, with the setup written down, that is the most useful thing you can add here.