Pizza Hut in 2024 deployed something called Dragontail. It was billed as AI, meant to give DoorDash drivers real-time visibility into kitchen workflow, oven timing, tip amounts, and cash status on every order.

Unfortunately, it lacked sufficient intelligence to be made artificial.

Here’s the simple math. Drivers work for DoorDash. They are paid per delivery (although some still expect the slavery-era concept of tipping). Pizza Hut wants pizza to go immediately out the door hot, while Drivers want maximum deliveries per trip. Before the rollout, drivers had no way to game the ordering system. After the rollout, they had full visibility into variables that they could selfishly optimize.

So they heavily optimized towards themselves at direct cost to Pizza Hut.

A new lawsuit documents that drivers started waiting up to fifteen minutes inside stores, to batch multiple orders that came out of the oven. Seems logical. They studied who would pay high slavery-era tips and oriented wait-times around these slavery-era inequality markers. The kitchen became an interface for driver market manipulations, driven by slavery-era tipping culture, rather than a production line for equitable customer delivery.

The “tipping” point was Pizza Hut delivery collapse. Chaac Pizza Northeast went from over ninety percent of deliveries under thirty minutes to alleging they suffered a hundred-million-dollar disaster across 111 stores. New York year-over-year sales swung from positive 10.19 percent to negative 9.78 percent. Pizza Hut watched all these metrics, adhering to a futuristic belief in the AI, and kept the toxic “tipping” driven system in place.

This is textbook design failure, perhaps even a 101 for future political science and economics students. Hand someone with misaligned incentives the data they can use to game you, hook it up to an inequality engine like tipping, and don’t act surprised when it collapses the market. Claims about the data pipe being AI are somewhat important. While information sharing is indeed the underlying issue, AI was the power move.

It’s like shifting from bolt-action to repeating rifle, you can’t just talk about gunpowder. The NRA runs this narrative constantly. They say the gun is incidental because the person aiming into public places is the cause. They won’t let the mechanism be the mechanism. We saw this clearly in Washington recently. An officer at the White House Correspondents’ Dinner fired at least four rounds toward his own colleagues at a checkpoint that was already dismantled, while a man sprinted through it. The administration converted their man’s failure into claims of an assassination attempt and demanded a billion dollars for ballroom security infrastructure. The biggest threat was their own officer, their own design failures. The shots came from inside the perimeter the new spending was supposed to harden.

The Pizza Hut AI infrastructure expense was supposed to improve delivery, when instead it turned into a threat to delivery.

Scientists say they are surprised, while also saying it’s simply logical that ravens remember where to find food.

Researchers found that wolf kills often clustered in certain parts of the landscape, especially flat valley bottoms where wolves hunt more successfully. Ravens visited these areas much more often than places where kills rarely happened. This suggests the birds learn and remember long-term feeding patterns across the environment.

“We already knew that ravens can remember stable food sources, like landfills,” says Loretto. “What surprised us is that they also seem to learn in which areas wolf kills are more common. A single kill is unpredictable, but over time some parts of the landscape are more productive than others — and ravens appear to use that pattern to their advantage.”