Dinner from what's in your fridge: how AI meal planning helps
What to cook when the answer is 'whatever's already in there.' How AI meal planning thinks about cooking from what's on hand, and where Aioli fits today.
It’s 5:47 PM on a Wednesday. The fridge has half an onion, the wilted-but-not-yet-dead remainder of a bag of spinach, two eggs, a piece of feta you forgot about, half a sleeve of tortillas, and a chicken breast that’s about to commit a real act of crime if it doesn’t get cooked tonight.
You did not plan for this. You did not shop for this. The question — what’s for dinner — is not a recipe-search question. It’s a constraint-solving question. The constraint is: what can I make from what’s already in there, in thirty minutes, without anyone going to the store?
This is one of the most universal cooking problems and one of the least well-served by traditional meal-planning apps. Most plans assume you start from a clean slate, generate a recipe, then send you to buy the ingredients. But the fridge already has stuff. The fridge always has stuff.
This post is about how to think about cooking from what you have, what AI meal planning currently does and doesn’t help with, and where Aioli fits today.
The honest problem with “what’s in my fridge”
Searching “recipes with chicken and spinach” has been possible since the early 2000s. It produces an avalanche of pages — recipes that almost match, recipes that match if you have eight other things, recipes optimized for ad inventory rather than for you actually cooking dinner.
The real problem isn’t “find me a recipe with these ingredients.” It’s:
- I have these things, in these rough quantities
- I have a few staples — pasta, rice, garlic, olive oil — that I haven’t counted as ingredients but I do have
- I have thirty minutes
- I have one or two people who will eat this and possibly a kid who won’t
- I do not want to go to the store
That’s a generation problem, not a search problem. You’re not retrieving an existing recipe from a database. You’re asking something to compose a meal that respects all those constraints, including ones you didn’t articulate.
This is exactly the kind of thing language models are good at, in principle. The trick is the input.
Three things that actually work, with or without an app
Audit before you generate. Take sixty seconds and look in the fridge and pantry. Not a deep inventory — just “what’s here that needs to get used soon, and what staples do I always have.” Most of what you cook in a fridge-clearing meal pivots around two or three perishables — the chicken, the spinach, the half-block of feta — plus pantry staples like olive oil, garlic, pasta or rice, salt, and lemon.
Build around the protein that’s about to expire. Most weeknight from-the-fridge meals are organized around the one ingredient that won’t make it to Saturday. Vegetables go around it. The starch is whatever’s already open in the cupboard. This isn’t an aesthetic choice — it’s triage. The chicken from three days ago is the deadline.
Accept that it doesn’t have to be a “recipe.” A pan with olive oil and garlic, the wilting spinach cooked until it’s not wilting, two eggs cracked in to scramble, feta on top, served over the tortillas you warmed up — that’s dinner. That’s not a recipe with a name. It’s a plate of food that used what was there. Mealtime is a result, not a performance.
How AI meal planning approaches “use what you have”
The general approach — across any AI planner that handles this well — looks roughly like this:
The user provides a constraint set: ingredients on hand, dietary preferences, time budget, family situation. The model composes options that respect all the constraints. The user picks one and cooks it, or asks for alternatives.
Where AI specifically beats search is in the composition step. Search retrieves recipes that already exist. Generation can produce a coherent meal idea from any input, including weird inputs — half an onion, three eggs, soy sauce, no rice, vegetarian, a kid who’ll actually eat it.
The catch: this only works if the planner actually accepts an inventory as input. Most current meal-planning apps — Aioli included, in v1.1 — are oriented around generating a fresh weekly plan and then producing the shopping list. The flow assumes the fridge will be filled, not consulted.
That’s a real product gap, and worth being honest about.
What Aioli does today, plainly
v1.1 is what’s currently on the App Store.
The plan creation flow takes an eight-step form: family size, days to plan, meals per day, diet type, allergens to avoid, location, whether to lean local or international cuisine, and an “adventurousness” dial from classic to bold. From that, Aioli generates a custom plan with recipes and a shopping list.
For someone who wants to plan a week in advance and shop fresh, that flow is the right one. For someone standing in front of an open fridge at 5:47 PM, it’s not. The current Aioli plan generation does not take inventory as input. There’s no place in v1.1 where you tell Aioli “I have chicken, spinach, feta, two eggs — what do I make.”
Inventory mode — generating meals from what you already have — is the next major feature direction on Aioli’s roadmap. There’s no timeline I’d commit to. It’s coming, and it’ll change the shape of the app when it lands. Until then, the honest answer is: Aioli does the weekly-plan job today, not the empty-fridge job.
What you can do with v1.1 if “use what’s there” is your problem
Two practical patterns that fit how Aioli currently works.
Plan around what’s already in the freezer or pantry. When you generate a weekly plan, you can implicitly steer it. Pick “classic” on the adventurousness dial — more familiar territory, more likely to use staples you already have. Pick a smaller plan length. Review the shopping list before you go. If half the items are already in your kitchen, mentally cross them off. The plan still works — you just buy less.
Use Aioli for the planned half of the week, leave the rest unscheduled. A four-person family doesn’t need seven fully-planned dinners. They might need three — the ones where decision fatigue would otherwise win. The other four nights run on improvisation, leftovers, or fridge-clearing meals. Aioli is good at the planned half. The improvised half is a human-and-fridge problem until inventory mode ships.
This isn’t a workaround in a defensive sense. It’s how a lot of households actually eat. Some nights are planned. Some nights are “what’s in there.” Both are valid.
Closing
The fridge-clearing meal is a kitchen skill more than it is a meal-planning feature. The principle — audit, build around what’s expiring, accept that it doesn’t need a name — works whether you have an app or not.
What AI meal planning will eventually do well is take an inventory as input and generate options that respect it. Aioli isn’t there in v1.1. It will be. In the meantime, the app handles the weekly-plan side of household cooking, and the Wednesday-night fridge-clearing remains a human craft.
If you’re hoping for “snap a photo of your fridge and get dinner ideas” — that’s not what v1.1 does, and I don’t want to pretend otherwise. When inventory mode is real, it’ll be on this site, plainly described.
Frequently asked questions
Can I tell an AI meal planner what's already in my fridge?
Some can, most can't yet. The general approach — called inventory mode in product land — takes a list of ingredients you already have and generates meal options that respect that inventory. It's a different shape from the standard 'plan a week, generate a shopping list' flow most apps default to. Inventory mode is on Aioli's roadmap but not in v1.1.
Does Aioli have an inventory mode?
Not in v1.1. The current plan creation flow is an eight-step form covering family size, days, meals per day, diet, allergens, location, cuisine direction, and how adventurous the recipes should be. It generates a fresh plan and a shopping list. There's no place in the current flow to tell Aioli 'I already have chicken, spinach, and feta, what do I make.' Inventory mode is the next major feature direction. There's no timeline we'd commit to.
How is 'what's in my fridge' different from a recipe search?
Recipe search retrieves existing recipes that match keywords. Cooking from what you have is a composition problem — you have specific ingredients in rough quantities, you have time and energy constraints, you have someone to feed who has opinions, and you want a coherent dinner that respects all of that. Search returns a list. AI generation can compose a meal that fits the actual constraint set, including the constraints you didn't articulate. The catch is the planner has to accept inventory as input, which most don't yet.
What's the fastest way to figure out dinner from leftover ingredients?
Audit the fridge for sixty seconds before you do anything else. Identify the protein or perishable that's about to expire — that's your anchor. Identify two or three vegetables or aromatics that pair with it. Pull a starch from the pantry. Cook it as a one-pan thing rather than a recipe with a name. Most weeknight fridge-clearing meals are protein plus vegetable plus pantry staple, treated as a plate of food rather than a performance.
Are AI meal plans worth it if I mostly cook from what I have?
It depends on which part of the week you mostly improvise. A lot of households plan two or three dinners and improvise the rest. For the planned dinners, an AI planner saves real decision energy and produces a shopping list. For the improvised dinners, you're on your own until inventory mode is widely available. Aioli today fits the planned-half use case well; the improvised-half is a human-and-fridge problem for now.
What pantry staples make fridge-clearing meals easier?
Olive oil, garlic, salt, an acid (lemon, vinegar), pasta or rice, eggs, and one or two long-shelf-life flavor anchors like soy sauce, parmesan, or a jar of harissa. With those, almost any combination of perishable protein plus vegetable becomes a plausible dinner. The pantry is what turns 'random fridge contents' into 'a meal' — it does more of the work than recipes do.