(👉Watch the video version of this article!)
Why Bother Trying This?
Has there ever been a meal you really wanted to recreate but you just couldn’t get right? Maybe you even have the recipe but it lacks the specificity to get it to taste just the way you remember it. Today I’ll show you how you can use AI to reverse engineer your favorite dishes, honing in on the exact ingredients and steps you need to recreate the flavors you love.
You can use this process to recover a lost family recipe, recreate your favorite dish from a now-closed restaurant, or redesign a meal that’s too expensive or too cumbersome to make. Eating at a restaurant puts someone else in control of what goes into your body. This process allows you to control the cost, the quality, the ingredients – all without sacrificing the taste.
There are many uses for AI that rob us of our creativity, but this process invites AI to be our collaborator. Ultimately, AI is just a brain in a jar – it can’t experience the world. It can explain all the complex chemical interactions and masterful techniques, but it has no tongue. That’s where you come in! Odds are you haven’t spent your life training to be a master chef, let alone a master chemist, but by partnering with an AI “Test Kitchen Director”, you gain access to its wealth of knowledge like Neo in The Matrix “downloading” the training to pilot a helicopter.
The Process: The Test Kitchen Engine
🔍Step 1: The Gathering (Of Data)
Collect whatever details you can about the dish. A link to the website, photos of the specific dish in question but also photos of the restaurant to give valuable context, contextual info, and of course your raw sense memories, even how it made you feel if you like. Each detail helps flesh out a part of the bigger picture – sharing that the chicken taco you’re recreating costs $2 helps to narrow down ingredients. A restaurant in a specific neighborhood, with a specific cultural background can suggest alterations from traditional method. You don’t need to worry about overwhelming the AI with too much information – if recreating a family recipe, you can collect a subjective account from each member of your family that wants to participate.
NOTE: You don’t need too go overboard in this first data collection phase, you’d be surprised how much it can glean from very little information an the AI will discover any additional data it needs in the next step.
🗨️ Step 2: The Interview
After you give the AI all the data you feel like, have the AI interview you. Direct it to ask you a series of questions to zero in on all the nuances of the recipe. Did you see little pools of orange oil floating on top, or was the sauce completely uniform and creamy? Was the spice level mild, or did it have a distinct smoky heat, like BBQ? You can decline any of these questions or ask it to rephrase them, if they aren’t clear enough. AI is smart enough to be able to convert a complex technical question – “What type of fat solids are present?” – into something anyone can understand – “Did it feel slimy on your tongue?”
📜Step 3: The Hypothesis
Whenever you get tired of answering questions, ask the AI to output its best guess at the recipe. This might not be the correct and final recipe – it’s an experiment. Review the recipe and see if it seems reasonable and feasible. There might be ingredients you won’t be able to find, or cannot eat due to dietary restrictions or preferences. There might be too many steps or techniques that you don’t feel comfortable with – the entire recipe is up for negotiation and I encourage you to do so! You’d be surpised by the workarounds you might find (brilliant chefs like Lan Lam and Dan Souza at America’s Test Kitchen fry capers and shallots in the microwave). Just know that some things can’t be rushed or substituted. You might need to accept a degradation in quality in order to facilitate the likelihood of actually cooking the meal (a bird in the hand). Remember, this is just version 1 and when you cook it, you’re testing a hypothesis. Even if the recipe fails, your complaints about the failure are essential feedback for the AI to improve the recipe in the next round!
🍳Step 4: The Cook
You’ve got the recipe, so now you cook it, you taste it, and you go back to step 1, gathering data! Take note of the results of your experiment, both successes and failures – and not just about the taste but about the experience of cooking it too. Though it might be more efficient or better tasting to take certain steps, if they cause you so much friction that you won’t actually follow these steps, then tell that to the AI and let it offer alternate suggestions. Your displeasure won’t hurt its feelings, so feel free to be brutally honest. You’re a team working towards a common goal, after all. From this point you simply rinse and repeat steps 1-4 to your hearts content.
Just in case you’re concerned about the amount of work it seems like you have to do here: my experience has been that the 2nd version is good enough. Plus the output of each test is some hopefully delicious food! The only real work is taking the time to feed the data back into the AI so it can update the recipe.
The “Test Kitchen Director” AI Profile
So I’ve outlined the process but how do you get an AI to do this for you? Do you have to build this AI persona from scratch? Though you’d likely have some success without much prompt engineering skill, I’ve already designed the “Test Kitchen Director” AI profile for you and I offer it for FREE to anyone that subscribes to my Andrew James Creative newsletter (if you’re already subscribed you should have already received it!). Click the link below, sign up, and you’ll receive my tailored Test Kitchen Director profile immediately – ready to be copied and pasted into the LLM of your choice!
👉I want The “Test Kitchen Director” AI Profile!
If we can use a workflow like this to shrink the distance between you and your next home cooked, Michelin-star meal, imagine what we can do with your business operations or your creative blocks. At Andrew James Creative, I offer creativity coaching, consulting, and process design to help you turn your painful chores into frictionless, systems tailored to your needs. Let’s make tools and processes work for you, not the other way around.