Optimization Is Not Wisdom

The optimization instinct#

Modern institutions possess a strong instinct toward optimization. Reduce friction. Increase efficiency. Compress timelines. Remove redundancy. Measure performance. Eliminate waste. The assumption beneath these efforts rarely requires defense because optimization appears self-evidently rational. If something can become more efficient, it should. If throughput can increase, increase it. If a system can move faster with fewer resources, redesign it.

The machinery is understandable. Optimization frequently produces visible gains. Costs decline. Outputs increase. Performance metrics improve. The evidence appears persuasive.

But beneath the logic sits a question institutions often discover later, and usually too late: what exactly is being optimized? Because optimization and wisdom have never been identical capacities. Systems repeatedly become fragile by improving the wrong thing.

Efficiency creates evidence. Coherence creates resilience. The distinction initially appears small. Its consequences rarely are.

What optimization rewards#

Optimization rewards local improvement. A standup gets compressed from fifteen minutes to seven, and the information that used to surface in the last eight minutes simply stops surfacing. A procurement process gets streamlined from four approvals to one, and the institutional memory that lived in the three removed conversations quietly disappears with them. A hiring loop gets shortened, and the candidates who would have stood out only on the longer signal stop being detectable.

Each of these is a real efficiency gain. Each of them measurably succeeds at what it was designed to do. And each of them weakens something that was never on the dashboard.

Systems rarely fail at the point being optimized. They fail somewhere adjacent. And the adjacent costs frequently remain invisible until much later, by which time the optimization itself has become culturally untouchable.

Meadows and systems blindness#

The clearest intellectual work on this problem belongs to Donella Meadows. Her work returned repeatedly to a deceptively difficult observation: systems behave according to relationships, not isolated parts. The issue is not merely what exists inside a system. The issue is how the elements influence one another. Optimization frequently misunderstands this. It improves components while weakening relationships.

Meadows's canonical illustration is induced demand. A city observes that its highways are congested. The optimization is obvious: add lanes, increase throughput, reduce travel time. The lanes are added. For a brief period the optimization appears to work. Then a second dynamic activates. Lower travel time changes where people choose to live, where firms choose to locate, and how often people choose to drive. Demand rises to meet the new capacity. Within a few years, congestion returns — often worse than before, now distributed across a larger and more dependent system.

The optimization succeeded at the component level and failed at the relational level. The lane count improved. The system degraded.

Meadows's point was not that highways are bad. It was that interventions inside systems produce outcomes opposite to those intended whenever the intervener optimizes for what can be measured rather than for the relationships that sustain the system itself. The problem is not intelligence. The problem is narrowness.

Boeing and metric gravity#

Metrics have gravity. Behavior orbits what is measured. Over time, the orbit becomes the system.

The clearest institutional embodiment of this dynamic in recent memory is Boeing after the McDonnell Douglas merger, culminating in the 737 MAX. Each individual optimization was defensible. Move corporate headquarters from Seattle to Chicago, separating executive judgment from the engineering floor — a defensible governance optimization. Outsource larger portions of design and manufacture to a global supplier network — a defensible cost and capital optimization. Extend the 737 airframe rather than design a clean-sheet successor — a defensible certification and capital optimization. Add MCAS as a software correction for the aerodynamic consequences of the extension — a defensible engineering optimization within the constraints the prior optimizations had created. Train pilots on the MAX as a minor variant rather than a new type — a defensible training and commercial optimization.

No single decision in this chain was obviously wrong at the time it was made. Each was the locally rational answer to a locally measured pressure. The cumulative effect was something different. The relational substrate that had historically carried Boeing — the intimacy between engineering judgment and executive authority, the trust between Boeing and its regulator, the cultural assumption that the aircraft itself was the unit of optimization rather than its components — was thinned by each defensible decision until there was not enough of it left to catch what the metrics could not see.

The MAX failures were not the result of optimization failing. They were the result of optimization succeeding, at the wrong layer, for long enough.

The same pattern recurs across domains. NHS bed-throughput targets steadily eroded the slack that absorbed seasonal surges, and the system that looked most efficient on a normal Tuesday was the system least able to survive an abnormal one.

The danger of optimization is rarely a single decision. The danger is gravitational.

Efficiency still matters#

None of this is an argument against efficiency. Efficiency matters. Measurement matters. Performance matters. Organizations cannot function through perpetual ambiguity.

The opposite failure exists too. Some institutions confuse complexity with sophistication. Others avoid measurement entirely. Constraint can become vagueness performing wisdom.

The argument is narrower. Optimization becomes dangerous when optimization begins replacing understanding. Efficiency answers: how can this improve? Coherence asks: what changes elsewhere if it does? The two questions are partners. The failure emerges when one is permitted to eliminate the other.

The constraint review#

Complex environments increasingly require a discipline institutions rarely perform: constraint review. Not what can we optimize, but what invisible function are we protecting. Not what friction can disappear, but what purpose does the friction currently serve. Not why does this step exist, but what would we discover only after removing it.

The difficulty is emotional. Optimization feels responsible. Constraint frequently feels inefficient. But systems often survive because invisible limitations quietly protect them.

The leaders who maintain this discipline tend to ask, repeatedly and without embarrassment: which inefficiencies are carrying intelligence? Not a performance review. A constraint review. The board cycle measures whether the plan is being executed. Almost no institution has a comparable cycle for asking which of the institution's frictions are load-bearing.

The quieter distinction#

The institutions of the future will continue rewarding optimization. They should. Efficiency remains valuable. Measurement remains valuable. Performance remains valuable.

But increasingly complex environments require a quieter capability: the ability to distinguish efficiency from coherence. Some leaders will continue improving visible components. Others will continue strengthening underlying systems. The distinction initially appears small. Over time it becomes enormous.

Optimization creates efficiency. Coherence creates resilience.

And the most fragile systems are rarely inefficient. They are often systems optimized beyond the point their underlying relationships could survive.