There is something almost reassuring about the way NYT Cooking has chosen to wade into the strange, swirling waters of artificial intelligence. In a moment when so many tech companies are racing to bolt chatbots onto every possible surface, the team behind one of the internet’s most beloved recipe collections seems to have taken a deep breath and decided that the most important thing an AI assistant can do is sometimes tell you no. “We put a lot of effort into trying to tell you when not to do something if it’s no longer in the spirit of the recipe,” says Camilla Velasquez, the general manager of NYT Cooking. That single sentence captures a philosophy that feels almost radical in its restraint. Rather than positioning the AI as a brilliant culinary oracle that can magically transform any dish into something entirely different, the Times has built a tool that is meant to be a quiet, sensible companion in the kitchen—one that understands that a recipe is not just a list of ingredients, but a carefully balanced set of intentions. It knows, for example, that swapping out one kind of dairy for another might work in a sauce but completely ruin a delicate custard. It understands that a recipe’s character, its texture, its very soul, can be lost with a careless substitution. And so the AI has been trained not only to suggest but to warn, to pause, to say, “Maybe not this time.” That kind of measured, thoughtful approach feels particularly welcome in a culinary landscape where online recipe comments are often filled with people proudly announcing that they replaced half the ingredients and were shocked when the dish didn’t turn out. The Times is essentially trying to build a digital kitchen assistant that respects the craft of cooking, and that starts with knowing when a clever idea is actually a bad idea.
The mechanics of how this AI actually works are just as deliberate as its philosophy. According to Velasquez, when someone asks the tool for a recommendation, it doesn’t just pull an answer out of thin air. Instead, it follows a careful hierarchy of sources, looking first at the recipe itself, then at the Times’ own ingredient substitution guide, then at the thousands of user comments that have accumulated beneath the recipe over the years, and finally—only after all of those human and editorial sources have been considered—it draws on what the team describes as AI “drawing on our expertise.” That ordering matters. It means the AI is not inventing its own culinary truths; it is synthesizing the accumulated wisdom of the Times’ test kitchen, its recipe writers, and its home cooks. It is, in a sense, a librarian rather than an author. And Velasquez is emphatic about the AI’s proper place in the cooking process. “You shouldn’t be chatting with it,” she says. It’s not a dinner party guest or a cooking show host. It’s a tool, like a good chef’s knife or a reliable oven thermometer—something you use to get a job done, not something you build a relationship with. She also reveals that the development process was deeply collaborative: recipe editors helped create the original rule set that governs the AI’s behavior, and then, at the end of the process, they evaluated how well it actually worked in practice. That kind of editorial oversight is crucial. It means the people who understand cooking at a professional level were in the room from the very beginning, shaping not just the data the AI learns from, but the very rules by which it operates. They were the ones who taught it that a recipe is more than the sum of its parts, and that a substitution isn’t just a swap—it’s a ripple effect that touches everything from cooking time to final texture to flavor balance.
Of course, whenever a major publication introduces AI into its workflow, there is an immediate and understandable fear about what it means for the humans who have been doing that work all along. The New York Times has more than sixty editors, writers, video journalists, and other employees dedicated to NYT Cooking, and the idea that a chatbot might start replacing them is enough to make anyone in the food media world nervous. Velasquez, however, is quick to push back on that anxiety. She stresses that NYT Cooking has been “hiring more than ever” and that the organization has “zero intention of publishing fewer recipes.” That is a significant reassurance, especially in an industry where AI has already cost people their jobs and where many publications are quietly shrinking their food sections. The Times seems to be taking a different path, one where AI is not a replacement for human creativity but a supplement to it. The recipes themselves—the ones that people clip, bookmark, cook for their families, and pass down to friends—are still being written, tested, and refined by real people with real palates. The video journalists are still filming, the writers are still telling stories, and the editors are still curating that particular vibe that makes NYT Cooking feel less like a database and more like a trusted friend who happens to know a lot about dinner. The AI, in this vision, is not there to generate new recipes from scratch or to take over the creative process. It is there to handle the more mundane, practical questions that arise after you’ve chosen a recipe: Can I use this instead of that? What if I only have one cup of oil? How do I make this vegetarian? These are the questions that home cooks have always asked in the comments section, hoping that someone—anyone—might have an answer. Now, the AI can offer one instantly, drawing on the collective expertise of the Times’ staff and its community of readers. It’s a way of making the recipe more useful, not making the people behind it obsolete.
So far, the early usage patterns suggest that the tool is indeed being used for exactly the kind of practical, everyday problems it was designed to solve. Velasquez says that people are primarily using it for ingredient and equipment substitutions, as well as for adjusting recipes to meet dietary needs. That makes perfect sense. A home cook might find a beautiful recipe for a braised lamb shank, only to realize they don’t have a Dutch oven. Or they might be cooking for a friend who is lactose intolerant and needs to figure out which cheese can be swapped without destroying the dish. These are the small but crucial decisions that can make or break a meal, and they are exactly the kind of thing that a well-trained AI can help with. Velasquez also mentions that some people use the chatbot to scale recipes, either up or down, to match the ingredients they have on hand. She gives the example of someone who had only one cup of oil for a recipe that required more. That is such a wonderfully specific, relatable problem. It’s not about culinary ambition or artistic expression; it’s about looking at your pantry, realizing you don’t have enough of something, and needing to know whether you can proceed anyway or whether you should make a trip to the store. The AI can answer that question in seconds, and in doing so, it can save a meal from disaster. It can also, crucially, tell you when you really do need to go to the store—when that missing ingredient is so essential that no substitution will work. That kind of honest guidance is rare in the world of online cooking advice, where so many sources are either overly permissive (“sure, just use whatever you have!”) or overly rigid (“no substitutions, ever!”). The Times’ AI, with its careful rule set and its grounding in editorial expertise, aims to be something in between: a knowledgeable friend who tells you the truth, even when the truth is inconvenient.
As a food writer whose own career has taken some substantial and unexpected turns because of AI in recent years, I admit I was not exactly clamoring for this development. There is a part of me that views any new AI tool with a certain amount of wariness, a sense that the technology is advancing faster than our ability to understand its consequences. But I also have to admit that the Times’ measured approach works pretty well. There is something deeply comforting about the way the AI is designed to be a helper rather than the main act. It doesn’t try to be charming or witty or creative. It doesn’t try to replace the experience of reading a recipe written by a human being who loves food. It simply sits there, quietly, waiting to be useful. And that, I think, is exactly the right role for AI in cooking. Cooking is one of the most human activities there is. It is about memory, culture, love, and the simple pleasure of making something with your own hands. No chatbot can replicate that, and no chatbot should try. But a chatbot can help you figure out whether you can use buttermilk instead of yogurt, or how to halve a recipe that serves eight, or whether you can get away with using a regular skillet when the recipe calls for cast iron. Those are not creative questions; they are practical ones. And by focusing on those practical questions, the Times is making it clear that the AI is not a rival to human cooks, but a servant to them. It is there to make cooking easier, not to make it meaningless. It is there to help you get dinner on the table, not to take over your kitchen.
I decided to put the tool to the test myself, and I started with something I know well: chicken and dumplings. I chose a five-star Alison Roman recipe with more than six thousand ratings, a beloved classic that has been cooked and loved by thousands of home cooks. I asked the AI to swap out bone-in, skin-on chicken thighs for boneless, skinless breasts. It thought about it for a moment—the process is described, in the charming patois of AI cooking, as “simmering”—and then it offered up a new version of the recipe. The title was now “Chicken Breast and Dumplings,” with the word “breast” rendered in a red-orange color to indicate that it had been modified. Just below the title, in a small but significant line of text, it said: “Joseph’s version made with AI. Not tested or endorsed by the original recipe’s creators.” That disclaimer is so important. It is an honest acknowledgment that this is not the real recipe, not the one that Alison Roman tested and perfected and put her name on. It is a variation, a what-if, a digital experiment. And alongside those visible changes, I could also toggle back and forth between my version and the original, seeing exactly what had been altered and what had stayed the same. That transparency is what makes the whole thing feel trustworthy. The AI is not trying to pass off its work as something it isn’t. It is not pretending to be Alison Roman. It is simply offering a possibility, a path forward for a cook who might be staring at a package of chicken breasts and wondering if they can still make something delicious. In the end, that is the best thing an AI cooking assistant can do: not replace the human experience of cooking, but make it more accessible, more flexible, and more forgiving. It is a tool for real life, with all its missing ingredients and imperfect substitutions and last-minute changes. And that, I think, is something worth simmering on.
