Watch a company tighten. One quarter's caution and every team, sensibly, protects itself: estimates gain buffer, hiring requests grow justifications, nobody volunteers people to the cross-team effort, every group holds a little capacity back for its own emergencies. Each decision, examined alone, is responsible management. Add them up and the company has quietly voted to slow itself down — and each team's caution now looks more justified than before, because everything genuinely is slipping.
John Maynard Keynes built a large part of his economics on the pattern hiding in that scene: the fallacy of composition, the assumption that what is true for one part must be true for the whole. His version concerned saving. For a single household in hard times, spending less is plainly prudent. But one household's spending is another's income, so when everyone cuts at once, total income falls, and the community can end up saving no more than before while being poorer — the thrift that would rescue any individual, generalised, deepens the slump. The argument runs through the General Theory of 1936, written to explain a depression that individual prudence was visibly failing to end.
The mechanism, stripped of the macroeconomics, is portable to any system whose parts transact with each other: the move from one actor to all actors changes the situation itself, because the actors are each other's environment. Reasoning that skips this — proving a behaviour sensible for one team and concluding it is sensible for every team — has not shown what it thinks it has.
Find where the totals stop adding
The discipline is a checking step for any policy or norm about to be generalised: rerun the argument at full adoption, tracking whose income each actor's spending was. Padding one team's estimate buys that team safety. Padding every estimate builds a roadmap of stacked buffers that commits the company to a fraction of its capacity — while the safety each buffer promised evaporates, because the dates everyone plans against are now fictions all the way down. One team declining to lend engineers to the shared platform keeps its velocity. Every team declining starves the platform all of them stand on. Attending the meeting to stay informed is cheap for each attendee; generalised, it fills the building's calendar with rooms of people each present because the others are.
Notice the cruel property these traps share: inside one, the individually rational move remains individually rational. The team that unilaterally stops padding gets punished by everyone else's padding. That is why exhortation fails — no part can fix a composition problem from where it stands. The fix has to change what is rational locally, which is a design task for whoever can see the whole: settle estimates centrally and spend one shared buffer, fund the platform off the top before team allocations, put a price on meeting hours. Keynes's policy conclusion, translated inward, is exactly that some level above the parts must act when the parts are trapped.
Watch out for
The tool cuts the other way too, and honesty requires saying so. Some behaviours do compose: if every team writes down its interfaces, everyone benefits more, not less. Waving the composition flag at any local practice you dislike is the mirror error — the test is always in the transactions, whether one actor's gain is another's loss at scale, not in the vibe of collectivism. And Keynes's remedy imports Keynes's risk: the level above the parts can be wrong at scale, with nobody left to offset it. Central buffers get raided, top-slice funding protects pet platforms past their usefulness. The composition argument earns intervention only where the trap is demonstrated, not wherever the centre would enjoy deciding.
Answer this next
Which behaviour is your organisation currently praising team by team — padding, protecting, attending — that nobody has re-argued at the level where the totals have to add up?
