By Vekt AI | May 2026
A Pizza Hut franchisee is suing its parent company for $100 million.
Not because the AI hallucinated. Not because a chatbot said something offensive. The AI worked exactly as designed, and that was the problem.
Chaac Pizza Northeast, which operates more than 110 Pizza Hut locations across New York, New Jersey, Maryland, Washington D.C., and Pennsylvania, alleges that the chain’s Dragontail AI system created “cascading operational breakdowns” that gutted delivery times, cratered customer satisfaction, and erased sales growth that had previously outpaced the national system. In New York City alone, a market that had seen 10% year-over-year sales growth flipped to -9.78% in the same quarter Dragontail was fully deployed.
The culprit wasn’t a bug. It was a system that gave DoorDash delivery drivers greater visibility into kitchen operations — and those drivers, rationally and predictably, used that visibility to batch orders, wait up to 15 minutes in-store, and cherry-pick orders with higher tips. The Dragontail system had been built for in-house delivery drivers. Chaac ran almost entirely on DoorDash.
Nobody apparently asked whether those two facts were compatible before going live.
Fast food has been a $28 billion AI experiment, and the results are uneven
Restaurant automation is now a $28 billion market globally. Labor shortages and thin margins drove fast-food operators to adopt almost anything that promised to cut costs or speed up service. A lot of it hasn’t worked the way the pitch decks said it would.
McDonald’s spent years and serious money building out AI-powered drive-thru ordering with IBM. After a two-year live deployment, customers started posting videos; one order came out as nine sweet teas and 260 McNuggets. McDonald’s ended the program in 2024.
Wendy’s announced AI menu boards and immediately got hammered in the press for what looked like surge pricing. They said that wasn’t the plan, but the story had already written itself.
Yum! Brands, which owns Pizza Hut and Taco Bell, is now partnering with Nvidia to put AI in drive-thrus and call centers across 500 locations. That’s live right now, while the Chaac lawsuit is moving through a Texas court. It’s worth paying attention to how that goes.
The technology in all of these cases was real. The failure wasn’t that the AI didn’t work. The failure was that nobody thought hard enough about what happens when you drop it into the actual messy, human, politically complicated system where it has to operate.
The thing nobody talks about in the AI sales cycle
When you’re being sold an AI tool — any AI tool — the demo is controlled. The environment is clean. The edge cases don’t show up. What the salesperson isn’t walking you through is what happens to your existing power dynamics when the thing goes live.
That’s what happened with Dragontail. The system gave DoorDash drivers visibility into kitchen timing. Drivers immediately used that information in their own interest — batching orders, waiting out slow periods, passing on low-tip jobs. Totally rational behavior. Nobody at Pizza Hut corporate apparently thought through the fact that these were independent contractors with zero obligation to optimize for franchisee outcomes.
Ajay Agrawal, who studies this stuff at the University of Toronto, made the point bluntly: to actually get the benefit out of AI or robotics, you have to redesign the whole system around it — you can’t just slot one tool into an existing process and expect the math to improve. The Chaac situation is what happens when you ignore that.
The other thing the lawsuit mentions — quietly, but it’s in there — is that Pizza Hut didn’t properly train operators on Dragontail and didn’t respond adequately when they asked for support. This part tends to get treated as a footnote in these stories. It shouldn’t. You can deploy a tool perfectly and still have it fail in the field if the people running it every day don’t know what it’s actually doing or how to step in when it goes sideways.
What works
Domino’s is a good counter-example in this space. They’ve outperformed the pizza category consistently for years, and a big part of that is AI-assisted ordering and logistics that they built and iterated on incrementally, over a long time, integrated with how their operations actually work — not parachuted in from a vendor contract. The technology evolved with the system.
The businesses getting real returns from AI right now almost always share a few things: they started with a problem they understood well, not a technology they wanted to use; they adjusted the surrounding workflow, not just added a tool on top of it; and their teams knew what the AI was doing well enough to notice when it wasn’t.
AI surfaces information and automates decisions, but humans still make the call when something breaks. If your team doesn’t understand the tool, they won’t catch the failures until they’re expensive.
Before you sign anything
If you’re evaluating AI right now: automation, agents, customer-facing tools, back-office workflows, the Pizza Hut situation is worth sitting with for a minute before you move forward.
The question isn’t whether the tool works in a demo. It’s who in your operation gains new information or capability when this goes live, and how will they use it? Is the tool built for your actual workflow, or a generic version of it? What does your team do when the output is wrong?