how it happened
- whoAI orderinghundreds of locations
- the gapno human fallback18,000 water cups accepted
- the resultrollout rethoughtbusiest stores excluded
what everyone assumed
By 2025 the technology had a track record. Voice AI was live in more than 500 Taco Bell locations and had taken more than two million orders, numbers that read like a system past its experimental phase. Scale is usually the proof that something works: if it handles millions of orders, it has earned trust. The reasonable belief was that the hard part was behind them, and that a tool running in 500 restaurants was a tool that ran itself.
what happened
The orders kept coming and the system kept taking them, which is what made the volume convincing. Most interactions went fine. The ones that did not were scattered across hundreds of locations and thousands of shifts, each one small on its own, none of them visible from a dashboard counting completed orders. Scale measured how often the system worked. It did not measure what happened in a busy restaurant when it did not, and those moments were spread too thin to register until people started filming them.
step by step
- automated drive-thrus take two million ordersMost go fine.
- failures are scattered and smallNo dashboard counting completed orders shows them.the assumption breaks here
- clips go viralAn 18,000-cup order, loops on simple requests.
- August 2025Taco Bell reconsiders its busiest stores
how it surfaced
The edge cases became the public record. Clips of the drive-thru accepting an 18,000-cup order, and looping on simple requests, circulated widely. In August 2025 Taco Bell’s chief digital and technology officer told the Wall Street Journal the company was reconsidering where and how to rely on the technology, acknowledging it might not be the right fit for the busiest restaurants. The system that had handled two million orders was being scaled back at exactly the locations where the pressure was highest.
Volume proves a system can repeat what it already does well.
It does not prove the system holds up at the edges, and the edges are where the busiest moments live.
picture it at a company like yours
The company here is made up. The other two paths are not: they happened, and each links to the story.
- in this storythe day it mattersCustomers film the system getting stuck or accepting absurd orders.what it costs youYou scale back at your busiest sites, where it mattered most.
- or it goes like thisit happenedMoffatt v. Air Canada, 2024howA customer-facing bot promises something your policy does not offer.what leavesmoney you now owewho ends up with itthe customer, backed by a tribunal
- or it goes like thisit happenedDPD chatbot, 2024howA bot is talked into mocking your company and the chat is shared.what leavesyour reputation, in screenshotswho ends up with ita million strangers in a day
sources
your turn
what you can do about it
this week
Take an automated process you rely on and ask where its edge cases go: who catches them, and whether anyone would know if they stopped being caught.
Somewhere in your company a tool has earned trust by handling a high volume of routine cases, and that trust has quietly extended to cases it was never tested against. The number of successful runs is not the same as coverage of the hard ones. Scale measures the ordinary case. The exceptions are where it breaks.
go deeper
if you're not sure where to start
Run the offboarding checklist on one leaver. It covers every account and token to close when someone leaves.
open the checklist →