El nuevo asistente de IA de NYT Cooking ayuda a cocineros caseros a ajustar recetas, pero aún necesita supervisión humana

A sous‑chef in the palm of your hand

La app New York Times Cooking, un elemento básico para cocineros caseros desde hace tiempo gracias a sus recetas curadas y su comunidad activa de comentarios, lanzó este mes un asistente de IA para suscriptores de iOS. Ubicado en un pequeño cuadro de texto en cada página de receta, el mensaje “Customize this recipe…” permite a los usuarios escribir solicitudes en lenguaje natural —desde “What can I substitute for this ingredient?” hasta “Scale this recipe to feed seven people, and make it vegan and gluten‑free.”

The assistant is trained on the app’s extensive archive of user comments, giving it a practical edge over generic large‑language‑model tools that lack cooking‑specific context. According to the NYT, the system also refuses “crazy requests” that would likely produce an unpalatable dish.

Weeknight chicken, saved in seconds

El revisor, un periodista experimentado en hogares inteligentes, probó la función con un dilema típico de día laborable: una receta de One‑Pot Chicken and Rice With Ginger que pedía chicken thighs, chard, cilantro y scallions, ninguno de los cuales tenía a mano. Al escribir:

“I have chicken tenders, fresh ginger, garlic, rice, broth, limes, but no chard or cilantro or scallions. I do have green beans and red onions.”

el asistente generó una nueva versión titulada One‑Pot Chicken and Rice With Ginger and Green Beans. Sugirió un tiempo de dorado más corto para las tenders, cocinar la red onion alongside the chicken, and steaming the green beans on top of the rice for the final eight minutes. The only unchanged element was the optional cilantro, which the user later swapped for Thai basil after a quick follow‑up query. The resulting dish cooked in the same total time as the original and earned the reviewer’s husband’s “seal of approval.”

Ingredient swaps and measurement conversions

Beyond meat cuts, the AI proved useful for a series of substitution tests:

• Thai red curry: Replaced makrut lime leaf with grated lime zest, noting the loss of the “deeper floral, herbal aroma.”
• Caramel apple Dutch baby: Swapped cream with half‑and‑half, with the assistant handling the metric‑to‑imperial conversion seamlessly.
• Roasted halibut: Suggested Castelvetrano olives and smoked paprika in place of kalamata olives and chile pepper, while warning that fresh cayenne peppers would add “significant, sharp heat.”
• Dessert trifle: Correctly rejected a request to replace ladyfingers with minced beef, stating the substitution was impossible.

These examples illustrate the assistant’s ability to recognize when a swap would compromise flavor or texture, a common pitfall of generic AI cooking advice.

Baking test: from Bundt cake to muffins

Baking, traditionally the reviewer’s weak spot, offered a tougher challenge. Using Genevieve Ko’s Cinnamon Roll Bundt Cake, the assistant was asked to convert the recipe into muffins—a logical request when a Bundt pan is unavailable. The AI:

1. Reduced the total bake time from three hours to 50 minutes.
2. Split the batter across two 12‑muffin tins (24 muffins total).
3. Halved the recipe to produce 12 muffins, adjusting the egg count to three (instead of an ambiguous 2½).

Both the original Bundt cake (made with a purchased pan for comparison) and the AI‑generated muffins turned out well. However, the reviewer ran into a usability snag: the app does not let users save the modified version separately. When asked later, “How much butter do I need?” the assistant quoted the modified recipe’s 6 tablespoons rather than the original’s three‑quarter cup, without indicating which version it was referencing. The reviewer recommends deleting the AI‑generated version after use to avoid such confusion, as the excess butter would have produced a dry cake.

Voice assistants vs. in‑app AI

For comparison, the reviewer asked Amazon’s Alexa Plus on an Echo Show 11 to handle the same chicken‑and‑rice swap. Alexa supplied similar ingredient suggestions but could not display a revised recipe; it narrated changes over the original screen. When prompted to walk through the steps, Alexa improvised, telling the reviewer to cut the chicken into bite‑sized pieces—a step not present in the NYT‑generated version. The in‑app AI proved more reliable for visual guidance, though the reviewer notes that voice control would be a welcome addition to the NYT Cooking interface.

Practical takeaways

• Speed and convenience: The assistant eliminates the need to sift through dozens of comment threads at 7 PM, offering instant, context‑aware suggestions.
• Safety net: It refuses obviously bad requests (e.g., substituting minced beef for ladyfingers) and flags potentially risky flavor changes (e.g., swapping cayenne for chile).
• Human verification still required: The butter‑measurement mishap and occasional timing tweaks remind users to double‑check quantities and cooking times against the original recipe.
• Limited persistence: The inability to save a customized recipe as a separate entry can cause confusion during cooking.

Outlook for home‑cooking AI

The NYT Cooking AI assistant demonstrates a practical, niche‑focused deployment of generative AI in everyday life. By leveraging the platform’s rich comment database, it offers more reliable culinary guidance than generic chatbots while still demanding human oversight. As more recipe apps experiment with similar features, the balance between convenience and control will likely shape user adoption, especially among purists who prefer to keep the original author’s instructions untouched.

For now, the tool appears to be a helpful sous‑chef for the nightly “what’s for dinner?” battle, saving trips to the grocery store and reducing the anxiety of missing a single exotic ingredient.