I Built an End to End Meal Prep App & I Hated It

I've been doing basically the same meal-planning routine for at least the last decade. Every week I'd open a spreadsheet full of recipes I'd collected over the years and stare at it for far longer than I'd like to admit.

I had about four recipes I knew I'd actually make because they were my go-to meals. Then there were another twenty that sounded good at some point, so I added them to the spreadsheet...and never cooked them once.

So I'd inevitably end up searching recipe sites looking for something different. The problem was that "different" usually wasn't what I wanted. I've accepted that I have a type. I like meals that are high protein, simple enough for a weeknight, and don't require buying five specialty ingredients I'll never use again.

My first attempt automated the wrong thing

My first version made the same mistake I see a lot of AI products making today. I tried to automate the entire workflow before I really understood which parts I actually wanted help with.

So I built an app that planned my week, generated a grocery list, pushed everything to Instacart, and left me with one button: "Place Order."

It worked.

I just didn't enjoy using it.

I didn't always want to eat what the app picked for me. Maybe we'd gone out for tacos the night before. Maybe I was craving pasta. Maybe it was ninety-five degrees outside and soup suddenly sounded terrible. Those are little decisions that are surprisingly important, and they're hard for software to get right.

I also discovered something much less philosophical: every recipe needs salt.

Which meant every grocery order contained...another container of salt.

The app had no idea there was already one sitting in my pantry.

Finding the right boundary

That forced me to ask a better question:

What part of meal planning actually feels like work, and what part do I enjoy?

What I really needed was a better set of options to choose from.

So every week the app now does two things:

• It selects a portion of the options from recipes I already know I love.
• It searches for new recipes that match my preferences: high protein, no red meat, simple weeknight meals, and generally the same style of cooking I already enjoy.

Every week I end up with a mix of recipes I already trust and recipes that feel like something I probably would have found myself if I'd spent an hour searching.

That's a much better starting point than staring at a spreadsheet or endlessly scrolling recipe websites.

When I find something I love, I can save it back into my permanent recipe library. Over time, the app gradually gets better because I'm the one deciding what belongs in my collection.

The grocery list was almost an afterthought

Interestingly, this was already enough to make the app useful. Once I'd picked my meals, generating a grocery list was easy.

Part of that was practical. We've been trying to go to the grocery store more instead of defaulting to Instacart (even though we both genuinely hate grocery shopping).

Once the planning experience felt right, I went back and added the Instacart integration. It turned out that wasn't the part that made the app valuable in the first place.

I also don't think I'll ever let it complete the process without me.

Recipes don't know what's already in my pantry. They don't know I bought soy sauce yesterday or that I still have half a bag of spinach sitting in the fridge. Having one final review before checkout still matters.

What building this reminded me

This little side project ended up reminding me of something I think applies to AI products far beyond meal planning.

Good AI products aren't necessarily the ones that automate the most.

They're the ones that automate the parts people don't enjoy while leaving the decisions that benefit from context, taste, and judgment to the person using them.

For this app, AI is great at:

• Searching thousands of recipes.
• Finding new ideas that fit my taste.
• Organizing grocery lists.
• Remembering recipes I've approved.

It can search thousands of recipes in seconds.

It can organize a grocery list better than I can.

It can even build an Instacart cart.

But it still doesn't know that after a long Tuesday I just really want pasta.

Code here: github.com/cfederer/mealprep