
Last week, I wrote, tongue-in-cheek, that I should vibe code an app for managing my bean recipes and ingredients. Well, that’s exactly what I did this morning. Ryan, another coffee shop regular, encouraged me to grab Codex from OpenAI or Claude Code and give it a try. I scoffed at the silly idea, thinking, I don’t have time for that.”\ After he left, that’s exactly what I did though: I downloaded and installed Codex. Ten minutes later, I was giving it permission to read a directory with my recipes and ingredient lists. I identified the layout structure of the text and the spreadsheets to Codex. Minutes later, I had this rudimentary app running locally on my computer. See screencap above.
I see the potential utility here plain as day. This idea of having my recipes, inventory, and shopping list living on my phone is an obvious no-brainer. Now that I have to give consideration to exactly what I’d want from such an app before I throw it up on one of my unused domain names to be available all the time, no matter where I’m at, as one never knows when the need arises to check on their bean recipes.
It’s mid-afternoon as I return to this post, and while I thought this app might be a great assistant for tackling bean management issues regarding the hundreds of recipes split between salad or crockpot variants, I had another thought: Why don’t I create a complete inventory of all foodstuffs we have on hand, all recipes we already enjoy, and those we are looking forward to trying?
Agents start to enter the picture and make sense. I buy beans from twelve different sources, they are often out of stock, awaiting a new harvest, and while a couple have notification processes, most do not. Or maybe they have brought a new bean into inventory that is not in our list of 124 varieties we own or have tried out of the 399 estimated to exist. Utilizing AI agents, it could have them make daily sweeps of those resellers and alert me when something interesting shows up. Come to think of it, if I understand things correctly, I can also give it permission to purchase me a pre-agreed-upon amount, simply by asking my permission to buy them by text.
This brings me to shopping agents. Recently, I was looking for the lowest price I could find for Fishwife Albacore Tuna in Spicy Olive Oil. Target at $6.99, while not as cheap as Costco (they only had it for a short time though), was the best price I could find. So, on their site, I ordered ten cans to pick up during the day. When I got to the store, I visited the shelf where tinned fish is kept, and checked what else was on offer. On sale for 50% off was more Fishwife, Albacore Tuna in Olive Oil for only $3.49 a can, far cheaper than Amazon. I bought a case of twelve. Writing this, I’m wondering who else in Phoenix might be selling Fishwife product and is running a sale that I don’t know about, if I had an agent working on my behalf scouring these company websites, I could keep up on where I can save us money without the effort of me spending time in a browser searching daily or worse, visiting each store to try and keep up.
Writing “out loud,” so to speak, I see that my helper app should also be able to negotiate the creation of our meal plans. If I’m looking at a recipe, why should I be able to assign it to a particular day? Once I have two weeks of meals mapped out, which is typically as far as I try to see forward, I could have the agent notify me in the morning what our meal plans are for the day. With that, it could also let me know if I have a particular bean to soak or ingredients I should be buying later in the day. From this logic, shouldn’t I be able to have it reason that if I bought something, say red cabbage with an average weight of 1.5 pounds (680 grams), and my recipe calls for only a quarter pound (113 grams), it could consider other recipes in our repertoire that use it and suggest subsequent meal ideas that will use the remaining ingredient if it sees I’ll likely have leftovers.
Uh-oh, this is a rabbit hole. I already know from experience that AI works quickly to dissect a recipe and its ingredients to provide me with a nutrition profile. This just had me ask AI (DeepSeek) to consider my activity, current weight, and age; it said I should be eating about 2,250 calories per day to maintain my weight. Great, DeepSeek, now consider that my triglycerides run on the high side, look at my meal plan, and offer me red flags where I’m crossing healthy levels from particular foods, such as those unhealthy fats, and offer green flags where other foods are offering healthy alternatives.
By having an overview of my diet and the impacts it can have on my diabetes, weight, and cholesterol, I think AI can build me a personal assistant that provides daily oversight on things I’ve never had an easy means to monitor. After bringing this all together, I’ll need a robot in the kitchen to cook on our behalf.
Just as I think I’m finished, I realize it will learn our favorite flavor profiles and then start searching for recipes that are complementary to our tastes. Then, I don’t need to know the names of Burmese salads that I don’t already have the name of, which stops me from asking for their recipe, or Korean stews we should try that aren’t available in Phoenix, Arizona. What about Indonesian recipes that are healthy and complement our interest in flavors from other lands? My agent could constantly be on the lookout for those treasures, even making suggestions that would alter a recipe to fit my diet restrictions.
All I needed to do was take my eye off the novel that has consumed me for more than two years to open the headspace to think a bit differently. Although I should add, this is my way of lashing out at my doctor who can recommend meds, but can’t give me real guidelines that are compatible with my likes, lifestyle, and culinary curiosities.