The New York Times Cooking announced on October 23, 2026 the arrival of an artificial intelligence assistant built into its iOS app — a tool that promises to transform how home cooks interact with thousands of tested and editorially approved recipes. Since then, the feature has undergone a gradual rollout, and with the beta now available to all subscribers, the question driving the debate is no longer whether AI can help in the kitchen, but how far it should go.
The assistant works as a chat embedded in each recipe page, accessible via a star icon positioned just above the ingredient list. Users ask questions or give instructions in natural language: substitute basil for parsley, scale this recipe to feed eight, make this recipe vegan and gluten-free. The model processes the request, generates a customized version of the recipe, and highlights in orange every change made — ingredient substitutions, quantity adjustments, altered cooking times, or adapted cookware.
But what sets the New York Times Cooking apart from most previous attempts at AI cooking is not the technology itself, but how it is built. Instead of letting the model act freely, the Times engineering team fed the system a set of rules created by the recipe editors themselves, complemented by an editorially maintained ingredient substitution knowledge base and, crucially, user comments. Camilla Velasquez, general manager of NYT Cooking, told WIRED that the AI draws on our expertise — meaning the model does not invent rules from scratch, but operates within a perimeter defined by professionals who have tested each recipe hundreds of times.
The practical result is mixed, which may surprise those expecting a mature product or, conversely, those fearing a culinary catastrophe. Across tests conducted by multiple publications, the assistant excelled with savory recipes — that flexible kind of dish where adjustments are the norm. In one case, when a chicken dinner was undercooked at 7 PM, the assistant suggested temperature and timing adjustments that saved the meal. When scaling a halibut fillet recipe from two to four servings, the model calculated the quantities and cooking time correctly.
Where the system fails is where it is least forgivable: in baking. Tests revealed that when halving a cinnamon roll Bundt cake recipe, the assistant changed five eggs to three — when the correct amount would be two and a half. This is the kind of error a home baker only discovers once the cake is already in the oven. Additionally, when asked to adapt the Bundt cake to a regular loaf pan, the assistant failed to account for the fact that changing the pan alters baking time and final texture, resulting in a cake that stuck to the pan and crumbled.
There is an interesting aspect to the assistant's safety architecture: when the user asks something absurd — like turning a gochujang buttered noodles recipe into a cookie — the system does not try to work magic. Instead, it displays a hard redirect and points to an alternative recipe in the Times' archive that fits the request better. We put a lot of effort into trying to tell you when not to do something, Velasquez told WIRED.
Public reception, however, was far from enthusiastic. On the comments section of the original announcement in the NYT Insider, readers threatened to cancel subscriptions, questioned who wanted this technology, and demanded to know how to turn it off. The phenomenon echoes earlier attempts at the smart kitchen market that always sounded like solutions in search of a problem — toasters with optical sensors and touchscreens that still could not toast bread particularly well.
The New York Times Cooking and America's Test Kitchen are, in fact, more than publishers — they are factories of tested culinary content. Good recipes, expensive to produce and demanding to create, are what they do best. Placing AI in front of that content represents a calculated risk: the cost of not innovating may be higher than the risk of occasional errors, especially when the product preserves the original recipe intact and allows the cook to switch between the modified and original versions at any moment.
The question remains whether this assistant represents the future of home cooking or merely an experiment the New York Times needed to run to understand how far artificial intelligence can go without destroying what makes it special — the human, unpredictable, and often improbable connection between the cook and the creator of the recipe.
Sources: The Verge, New York Times, ATLABYTE
✓ Independent sources cross-checked and verified before publishing
