Standardisation beats optimisation
Why an agreed interface is worth more than a better component.
Standardisation beats optimisation because a system built from many parts gains far more, and far more reliably, from every part sharing a common, agreed interface than from any single part being made as good as it could possibly be in isolation. A beautifully optimised part that fits only its own maker's system captures its advantage once, at that one join, while an ordinary part built to a shared standard captures a smaller advantage at every join it ever makes.
A charger that needs an adapter at every border
A phone charger works perfectly with the wall socket at home, built to whatever shape the local standard specifies, and none of the socket shapes used around the world is meaningfully worse engineering than the others. Each is a perfectly serviceable way of getting mains power into a device. That local adequacy is exactly why the charger needs an adapter the moment it crosses into a country using a different shape: two sockets can each be excellent and still be unable to serve each other without an extra piece of hardware between them.
The same small mismatch recurs at every border a traveller crosses, which turns one manageable inconvenience into a recurring cost over a lifetime of travel. The whole cost comes from the two designs never having agreed to be the same thing, and that is standardisation's advantage in miniature. Being the same as everything around it is worth more to a socket than being locally superior and incompatible, whichever design would have won a contest on its own merits.
Two questions that pull in opposite directions
Optimising a part means asking how good that one part can be, judged against its own job. Standardising it means asking whether it can connect, without negotiation, to anything else built to the same rule, made by someone else, possibly years later and on the other side of the world. The two questions often pull apart, because the changes that make a part locally better (a shape trimmed exactly to its one intended load, a size chosen for a maker's own convenience, a fitting made slightly non-standard to save a sliver of material) are frequently the changes that stop it connecting with anything it was not designed alongside.
A system depends on its connections at least as much as on the quality of what is being connected. A part that is excellent on its own bench but compatible with nothing else contributes that excellence at a single boundary. The shared interface, meanwhile, is agreed once, by whoever writes the standard, and every connection made to it afterwards, the tenth part or the ten millionth, inherits that agreement for free. An optimisation to a single component has to be re-earned on every unit built to it and delivers nothing to anything made independently by somebody else, so a standard's payoff grows with every future connection built against it while a local optimisation's payoff stays inside the one system it was made for.
As Levinson's The Box keeps returning to, a shipping container is the ideal shape for almost none of the cargo loaded into it. It wastes a little volume against nearly every load and is never the cheapest box that could be built for any one use. That small inefficiency is paid once per box, while the benefit of matching every crane, ship and lorry chassis in the world is collected again at every junction the box passes through over its working life, which is why the shared standard wins long before anyone adds up the totals.
Supply chains, buffers and a shared view of demand
The bullwhip effect running back through a supply chain is, underneath its specific mechanics, a symptom of the same problem. Each link optimises its own order in isolation, with no common, unfiltered view of actual demand, and that missing shared interface lets each link's reasonable local adjustment compound into a swing nobody downstream caused.
A buffer between two stages of a process does the opposite job for the same underlying reason. It stands in for the perfectly synchronised interface that would let two stages run at exactly matched speeds, and it earns its keep because building that interface is usually far harder than simply absorbing the mismatch. A supply chain, a production line and a global shipping network are all the same kind of problem, many independently made parts trying to work as one system, and in each of them the standard connecting the parts matters more than how good any single part is on its own.
Bespoke parts with one counterpart
Local optimisation is the right choice when a part will only ever meet one specific counterpart that nobody else will build against, such as a bespoke component inside a single tightly integrated machine that no one intends to service with another maker's part or connect to any wider system. Forcing that part to conform to a shared standard buys no future flexibility and still costs some of the performance a bespoke design could have delivered.
Standardisation earns its advantage from the possibility that a part might one day connect to something built independently, by someone else, at another time. The deciding question is how many future, unknown connections a part is likely to make. A component with exactly one destined counterpart gains nothing from a standard built for counterparts it will never meet, while a component that might end up joined to something nobody involved in its design has yet imagined has almost everything to gain from one.