What
Just Eat rolled out AI-generated food images for independent restaurants, built from their menu descriptions. The AI misread them. A breakfast bap came out as sausages and a fried egg floating in cola. Garlic bread became a bread-made aeroplane. A cheese roll became sushi. Customers called it out on social media, and Just Eat apologized, removed the inaccurate images and paused the rollout until the tool is more accurate.
So What
The tool did what it was asked: it turned text into a picture. What it couldn’t do was check the picture against the real dish. The reporting doesn’t say whether anyone reviewed samples before launch, but customers were the ones who caught the errors. AI is just as fluent when it’s wrong as when it’s right. A confident, plausible-looking output isn’t evidence of accuracy.
Now What
Any AI output that reaches a customer, a record or a work order needs a check against ground truth before you scale it. Pull a sample, compare it to the real thing, and count the misses. That check is cheaper than finding out from the people you serve.
Source: Grocery Gazette, Oct 21, 2026
#ArtificialIntelligence #AIStrategy #OperationalExcellence #Leadership #QualityAssurance
Leave a Reply