I Don't Need Meal Planning, I Just Need Help

My grocery cart defaults to food that's easy enough for lunch but good grief do I struggle come dinner time.

So by 4pm on a Tuesday, I'm staring at the fridge thinking "we have stuff but nothing to eat." By 8pm someone in the house has ordered DoorDash. I looked it up, as a household we spent an embarrasingly large amount of money on DoorDash.

The tempting fix is to build a meal planner. An app that asks me my preferences, suggests recipes, maybe syncs with Instacart. However, those apps obviously always exist and I've never been successful with them and also that's not my entire problrem.

The actual problem is I'm making dinner decisions at the worst possible moment.

Decisions made in low-load moments are ten times cheaper than decisions made in high-load moments. Friday afternoon when the work week is winding down, relaxed, talking to Claude? I can make ten good decisions about dinner in the time it takes to finish a cup of tea. Tuesday at 4pm after a day of meetings? I can barely decide whether to open the fridge.

So I shouldn't use or build a thing that asks me to decide at 4pm. I need to do this when decision making is free.

I used Claude Code to dump my last 30 Instacart and DoorDash orders into files. Then I triaged them. What's mine versus what's for the kids or my husband. What I'd actually eat for dinner versus what gets ordered when I've given up. Side nerd note, I used claude --chrome so Claude could access my apps as there are no direct APIs or MCP options. That also reduces friction. I don't have to screenshot or copy things over.

Patterns fell out fast. I was able to identify dinner archetypes. Five archetypes total, all built from things I actually eat, all with shelf-stable proteins, all with pre-cut ingredients because chopping veggies makes me want to lie on the floor and cry after a full day of back to back Zoom calls and decisions.

I wrote those five archetypes into a file. I wrote a second file listing everything that should be filtered out.

Then I wrote a Claude Code skill that reads both files, asks me two questions (how many dinners this week do I want to make? anything sounding gross?), and produces a shopping list. I also built it to do the inverse. I can screenshot my pantry and it'll tell me which archetypes I already have and what I need to add.

Two scheduled interventions do the work:

Saturday at 8am, Telegram DMs me my shopping list. Not a notification. An actual text message that is the list. I can forward it to Instacart or just paste it in. I did the thinking Friday night. Saturday me doesn't have to think.

Monday through Thursday at 3:30pm, the same bot DMs me tonight's dinner options from the five I unlocked this week. Not a suggestion engine. A list. I pick one. Sometimes I order DoorDash anyway. It's fine because my goal isn't to eliminate takeout, it's to eliminate the 4pm decision.

The pings are both launchd jobs on a Mac Mini, each one about eighty lines of Python that do nothing but read a markdown file and POST to the Telegram API. There's no app, no UI, no cloud service. There's nothing to maintain. If I want to change something I edit a markdown file.

what this isn't

This isn't meal planning. I don't have a recipe library. I don't have a weekly menu. I don't care what day I eat which thing.

This isn't a grocery tracker. I'm not pretending to know what's in my fridge. I just screenshot the fridge when it's time to build a list.

This isn't an AI doing my thinking for me. I did all the thinking. The AI just holds my own decisions where I can see them at the right moment.

why this works for me

Apps break because they require me to open them at the exact moment I least want to open an app. The whole point of the 4pm rescue is that 4pm-me has no capacity for "opening a thing." A text message that arrives in the same notification pane as everything else meets me where I am.

The whole system runs on files I control. If my AI tools break tomorrow, I still have my five archetypes in a markdown file. The pattern of "write your decisions into files, have a simple scheduled job surface them at the right time" isn't specific to any particular AI stack. It's just what happens when you stop trying to build "The Solution "and start removing friction from the one moment where friction actually hurts.

If you take one thing from this, don't take the infrastructure. Take the reframe.

Look at a problem you keep solving badly. Ask when in your week you have the most mental capacity to think about it. Pre-make the decisions then. Find the dumbest possible way to surface those decisions back to you at the moment you need them. This could be a text, an email, a sticky note on the fridge.

Your calm-whatever-the-best-day-of-the-week-night self is smarter than your 4pm-Tuesday self. Let her do the work.