Chefs at food service company Sodexo halve recipe time with AI

At Sodexo, the food service company, every new dish starts with a chef's first draft, called the bench recipe. Before the dish goes into development in the kitchen, the chef asks an AI tool to make a picture of the finished plate. The AI also helps write the culinary brief the team takes into research and development, Sodexo's culinary leaders told the trade publication FoodService Director for a story published Aug. 3.

Sodexo runs operations at about 27,000 locations around the world, from hospitals and college dining halls to schools and workplaces. "We have over 70,000 recipes in North America alone," said Richard Arakelian, vice president of culinary and service development for Sodexo North America. With team members moving between parts of the business, he said, quality and consistency are critical, and AI agents now help hold them steady. For a kitchen of any size, the lesson comes later: how specific the chefs get with what they ask.

"We are utilizing AI almost like a sous chef or chef's companion," Arakelian said. A sous chef is the second in command in a kitchen. He said AI condenses the company's development timelines by 50 to 60 percent.

How the chefs use it to cut waste

Recipe work topped the list of wins. Lloyd Mann, Sodexo's vice president of culinary, said the team "did a trial across 25 to 30 recipes to see how we could minimize food waste," and the AI came back with suggestions such as ingredient swaps. The chefs also asked it how to reduce a recipe's carbon footprint.

The swaps can get specific. When the team wants to replace white rice with something more drought tolerant, or grown with regenerative agriculture, the AI returns a list of options, Arakelian said, and with the right prompt it can narrow the list to the country or region where a crop is grown. In health care, where one kitchen manages many diet types and patient needs in a day, the AI swaps ingredients and runs the nutrient analysis quickly, and it does the same for allergens, food sensitivities and sodium.

Sodexo has changed what's on the plate at scale before, when it made plant-based meals the standard option at nearly 400 US college campuses in 2024. And the same tools are reaching people still learning to cook: four Northern Arizona University culinary students and graduates described using AI to count stockroom supplies from a photo and to scale recipes.

Why the prompt matters

Sodexo looked at building its own AI platform and chose mainstream tools instead. "The tools out there were better than what we could build," Mann said. What makes them useful is what the chefs feed in. "The more information you put in, the quicker it learns," he said.

A kitchen that wants to try this can start small. Take the dish that sends the most food to the bin, give an AI tool the full recipe, the portion and what usually comes back on the plate, and ask what you could swap. We'll be watching whether Sodexo takes the waste trial past its first 25 to 30 recipes.