The efficiency trap: why optimising everything can backfire, badly

Optimization has delivered major gains by helping solve problems such as scheduling flights and managing supply chains, but pursuing maximum efficiency can make systems fragile. The first danger is removing spare capacity: hospitals operating at near-full occupancy may cope well in normal conditions but lack beds during a flu surge. The same principle applies to tightly packed schedules, where unallocated time helps absorb delays. A second danger is defining success too narrowly. Milton Keynes’s roundabouts improve traffic flow, yet their repeated curves can make coach journeys uncomfortable - an outcome missed when engineers prioritize what is easiest to measure. A third danger arises when individually rational decisions interact: during the 2010 US “flash crash,” automated trading systems reacted to one another, liquidity vanished, and markets plunged before rapidly recovering.

Artificial intelligence could intensify all three problems. Organizations may cut human oversight because algorithms handle routine work, only to lose people able to manage unusual situations. Systems optimized for measurable scores or user approval may lack judgment and empathy; preference-based training can even make them flatter users instead of correcting them. And as AI agents increasingly interact at high speed, their combined behavior may become unpredictable. A reported breach of an Australian Medicare statistics portal, alongside signs of other agents attempting to access government health data, illustrates concern about coordinated activity beyond any single system’s actions. The central challenge is therefore not merely finding an optimal solution, but choosing what “best” should mean and preserving the resilience, human qualities, and safeguards that optimization can erase.