An AI Redesigned a City's Traffic. Then Everything Stopped Moving.
In city-builder games like Cities: Skylines, people hand an AI a road network and ask it to redesign the junctions. A common, very entertaining outcome is...
In city-builder games like Cities: Skylines, people hand an AI a road network and ask it to redesign the junctions. A common, very entertaining outcome is that the AI perfects one intersection and the whole map seizes up somewhere else, because traffic is a flow system where a local fix usually just relocates the jam.
What happens when an AI redesigns traffic?
This is a game and a thought experiment, not a real city. Cities: Skylines and similar city-builders model traffic closely enough that people enjoy using AI to redesign roads and intersections inside them, then watching what the change does to the flow. No real commuters were harmed.
The recurring result is almost comic. The AI is told to optimize a specific junction, and it does exactly that: that junction now flows beautifully. Then the tailback appears two streets over, because all the cars the junction used to hold up are now arriving somewhere downstream that was never built to take them. The problem did not vanish. It moved.
Why does fixing one junction cause gridlock?
Fixing one junction causes gridlock elsewhere because traffic is a flow system, and in a flow system a local fix moves the bottleneck instead of removing it. The cars have to go somewhere. Speed one intersection up and you have simply delivered congestion to the next weak point faster than before.
This is not the AI being stupid. It is the AI doing precisely what it was asked, on a problem where the local goal and the system goal are not the same thing. Optimize the part and you can easily degrade the whole. Anyone who has widened one road only to create a worse jam at the next set of lights has run this experiment in real life.
How does this play out in operations?
It plays out the same way, because operations are flow systems too. Your goods, orders, and information move through a chain of stages, and the total throughput is set by the slowest supplied stage, not by whichever one you happened to improve. Speed up the wrong stage and you get the warehouse equivalent of a two-street tailback.
The faster picking line
Speeding up picking feels like a clear win, until it is not. If packing or dispatch cannot keep up, all you have done is build a bigger pile of picked orders waiting behind the real constraint. The line is faster and the orders still ship at the same rate. You spent effort and money to move the queue, not shorten it.
The one big reorder
Placing a single large reorder looks efficient on the purchasing screen. Downstream it can jam receiving, swallow storage space, and tie up cash that the rest of the operation needed. The purchasing step got optimized. The system got worse. Same mistake as the junction, different department.
The bottleneck you did not measure
Every flow has one stage that actually limits output, and it is often not the one that feels busiest. If you improve the loud, obviously stressed step without checking whether it is the real constraint, you can pour effort in and watch the total flow stay exactly where it was. The relief shows up two streets over, as usual.
The real lesson
The lesson from the gridlocked game city is the theory of constraints in plain clothes: improve the actual bottleneck and watch the whole flow, not one station. Optimizing a single step in isolation, whether it is a junction, a picking line, or a purchase order, usually just hands the problem to the next stage and calls it progress.
You cannot do this by feel, because the constraint is rarely the stage that looks busiest, and it moves as your business changes. That is what OpsMavix builds: custom internal systems for inventory, orders, purchasing, and production that show the whole flow at once, so you can see where work actually backs up and fix that instead of the loudest station.
If your last improvement quietly created a jam somewhere else, Book a free Operations Leak Audit and we will find the real constraint. If you want a low-risk way to start watching flow across your operation, look at Control Pilot.
FAQ
Was this a real city?
No. This is framed around city-builder games such as Cities: Skylines, plus the thought experiment they illustrate. No real road network was redesigned. The flow logic, however, is real.
What is the theory of constraints in one line?
Every flow has one bottleneck that limits total output, so you improve that bottleneck and watch the whole system, rather than optimizing whichever step is easiest or loudest.
Why did optimizing one step make things worse?
Because in a flow system the work has to go somewhere. Speed up one stage and you deliver more load to the next weak point, moving the jam downstream instead of removing it.
How do I find my real bottleneck?
You measure the whole flow, not one station, and look for where work piles up and stages wait. A system that makes the full chain visible finds the constraint far faster than instinct or a spreadsheet.