Why My AI Meal Planner Failed
1. Context
Imagine being a busy family of five trying to eat clean after being spoiled in China with fresh-made everything. For two years, we had a house helper who made meals feel effortless.
I know, I know… we were incredibly privileged. But stick with me here.
Coming back to the US, meal planning became this relentless hamster wheel. It felt super repetitive. My wife and I were burned out trying to come up with something new each week. We were in pure survival mode: just getting by and keeping routine, rather than paying attention to what we’d actually like, what season it is, or remembering the many recipes we’d always thought would be fun to make. We didn’t want to trade family health and variety for convenience, but it was just mentally draining to plan on a tired Friday night before our Saturday morning Costco runs.
And no, YOLO-ing it without a plan in Costco feels great, but there’s SO MUCH WASTE. It seemed pretty straightforward, and then of course AI comes to the rescue. Knowing me, naturally, I built a robust database in Notion of all our family recipes and liked dishes. I started using AI to scan grocery receipts (because who doesn’t love a robust spreadsheet?!), and create this whole system to help us with meal planning. Spoiler alert: I was solving the wrong problem.
2. What I Tried
I set up Claude (yes, I’m playing with different tools!) to generate comprehensive meal plans that would remove all the cognitive load. Each week, we’d get Monday through Sunday mapped out with specific recipes, full ingredient lists, and step-by-step instructions.
Maximum automation. Minimum thinking. Problem solved, right?
Narrator voice: It was not solved. After a few weeks of testing, I noticed something during our Saturday morning planning sessions over coffee: we kept ignoring the detailed recipes. Like, every single week.
3. What Happened
We weren’t following the plans as written. Instead, we were using them as decision-making springboards. Seeing “One-Pot Creamy Tuscan Chicken Pasta” reminded us we had chicken thighs in the freezer, but then we’d just… search for a different creamy chicken pasta we actually wanted to make. “Italian soup with ground beef” sparked us to look for Italian-style soups that sounded way better than what the AI suggested. Turns out the meal plan’s real value wasn’t telling us what to do. It was removing the blank page problem and sparking the intuitive decisions we actually wanted to make. Who knew? (Spoiler: not me when I built the system.)
4. The Real Insight
Families don’t want scripts. They want just enough structure to make good decisions without the cognitive load. I shifted from exhaustive documentation to lightweight decision support: Before: Detailed recipes with full ingredients and instructions After: Quick inventory snapshot (“you have tilapia and chicken thighs”), theme reminders (“Monday fish, Tuesday tacos”), and category suggestions (“search for a creamy chicken pasta you like”)
The new format prevents decision paralysis and Costco overbuying while respecting our autonomy to choose what actually sounds good that day.
Because here’s the thing: it has to feel like something you’d actually want to eat, not just something that checks the “healthy family dinner” box.
5. Try This If…
You’re building AI systems for your team or family: Give frameworks, not rigid processes. When you over-prescribe, people disengage or work around the system. (Guilty even as a PM leader - it’s more art than science!) When you provide just enough structure, they engage more deeply because the tool enhances their judgment instead of replacing it. Start by asking: Does this person have expertise and intuition in this domain? If yes, shift toward frameworks. If they lack context or knowledge, prescribe more. If engagement and buy-in matter, lean into frameworks. If consistency or safety matter, prescribe more.
6. Systems Lens
This maps directly to project management and organizational design, which is probably why I got so nerdy about meal planning in the first place. The best systems aren’t exhaustive playbooks that no one follows. They’re “just enough” frameworks that remove cognitive load while preserving the autonomy people need to do their best work.
I see this pattern everywhere now: career planning (directional guidance over exact steps), home projects (decision checkpoints over detailed plans), team workflows (principle-based guardrails over rigid processes).
The goal isn’t to eliminate human judgment. It’s to enhance it by removing the friction that prevents good decisions from happening.
Where in your work are you over-prescribing when frameworks would create better engagement and outcomes?