Everyone Finds the Shortcut

Nobody on that side street did anything wrong. Each of you was offered a faster way home and took it. That is the unsettling part. The jam was not caused by a bad decision. It was caused by everyone making a good one at the same time, on the same piece of shared ground.

We usually picture inefficiency as somebody’s fault: a lazy clerk, a dumb rule, a driver who does not know the way. There is a kind of waste that needs no fault at all, and it is about to become far more common, because we are about to give every person a tireless assistant whose whole job is to find them the shortcut.

A road that makes everyone late

In 1968 Dietrich Braess, a German mathematician, published a short paper on traffic planning with a result that still sounds like a trick. Picture two routes from home to work. Each has one road that slows down as more cars use it and one wide road that always takes an hour. Drivers spread out until neither route is faster than the other, which is the settling point a British transport researcher named John Wardrop had described in 1952: nobody can save time by switching.

Now build a free shortcut connecting the middle of the two routes. It lets a driver use both of the roads that can be fast and skip both of the wide ones. Every driver who takes it saves time, so every driver takes it. Then both of the roads that can be fast are full, and they are not fast anymore.

That last line is the trap. Close the shortcut and every driver gets half an hour back. But no individual driver can get that half hour by choosing differently. The waste belongs to everyone and to no one. Computer scientists later named it the price of anarchy, and Roughgarden and Tardos showed that on roads like these it can reach a third, never more.

It is not only a thought experiment. The same shape has turned up wherever selfish choices share something scarce.

The first personal agents were in your car

Wardrop’s settling point assumes that every driver knows every road’s travel time and always picks the fastest. For most of history that was a convenient fiction. Real drivers had habits, did not know the back streets, and could not be bothered. Then navigation apps arrived and made the fiction true. Leonia is what it looks like when a theorem becomes literal: a town with a police department enforcing a road closure by hand against software.

Here is the turn. AI agents do not change the game. They perfect the players. An assistant that books, buys, schedules and reserves on your behalf is informed, tireless, and loyal to exactly one person, which is what you want from it. It is also precisely the player Braess’s theorem is about.

The friction we complain about, the not knowing and the not bothering, was quietly keeping a great many shared systems away from their worst settling point. Agents remove that friction everywhere at once. And the non-obvious consequence is this: making your agent smarter does not help, because the shortcut it finds faster, every other agent finds faster too. The fix stops living in the player. It moves into the rules of the game.

1. First come, first served dies wherever agents arrive.

   When everyone can be instant, a race is just a coin toss that costs the system a flood of requests. Drawings, booking windows and prices will replace the line for reservations, appointments, permits and launches. Being first will stop meaning anything.

2. Holding a place will cost something.

   An agent can hold four tables and release three at no cost, and for each person that is the smart move. Multiplied across a city it empties dining rooms. Deposits, firm holds and records of who shows up will spread to everything an agent can reserve.

3. The best upgrades will be removals.

   Like Stuttgart and Seoul, businesses and towns will improve by taking a road away: the line that agents cannot call, the booking that opens first to returning guests, the hour kept for people who walk in. Closing a shortcut will become an ordinary design decision instead of a strange one.

4. Coordination becomes the product.

   The gap between everyone choosing alone and one planner choosing for all is real time and real money, and someone will sell it back. Expect booking layers for a whole district’s restaurants and shared schedulers for a whole medical campus, with an honest pitch: you will sometimes get your second choice, and on average you will get there sooner.

5. Agents will carry a reputation for manners.

   Systems will learn to tell a cooperative agent from a greedy one and serve the first one sooner. How well an assistant plays with others will become a setting people choose, and a reason to choose one assistant over another.

I see this from the restaurant’s side. At Charleston AI I have put a phone agent on a restaurant’s reservation line, and once you have done that, the next question arrives on its own: what happens when the callers stop being people? One guest’s agent asking politely is a customer. Ten thousand asking in the same second is weather. Whatever gets built on the restaurant’s side of that line is, whether anyone calls it this or not, a planner for a road network. I wrote about where waiting goes once servers can copy themselves in The Last Line Is You, and about another place where everyone’s sensible move adds up to a worse map in The Middle of the Beach. This is the traffic between them.

How this could be wrong

Already true

Navigation apps already route millions of drivers by individual advantage, and towns have already closed streets in response. Sellers of scarce concert tickets and sneakers already use waiting rooms and drawings because automated buyers made the race meaningless.

What has to happen

Agents have to act, not just advise: book, hold, cancel and pay on their own, at scale, through the same doors people use. And enough people have to hand over the same kinds of errands at the same moments for the collisions to show.

Where I am probably wrong

Braess’s paradox needs a particular shape. Most of the time, adding capacity helps, which is why we keep building roads, and in many real systems the price of anarchy is small. Agents may also smooth demand instead of piling it up, nudging a flexible diner to Wednesday because they can see what a person cannot. If that is how it plays out, I will have mistaken a worst case for a forecast.

Go back to the side street at 5:10. In the world coming, your assistant finds that street in a millisecond, and so does everyone else’s. Being faster stops being an advantage the moment everyone has it. The only thing that will get you home sooner is a rule somebody wrote for all of you at once.

Somebody has to close the road.