Start with the problem. Add AI where it helps

What building a meal planner taught me about where AI actually creates value — and where it simply gets in the way.

What I wanted to solve for

When I started building my MealMate app, my main question was simple:

Could I use AI to make weekly meal planning easier?

The obvious answer was to use AI to generate meals, recipes and ingredients.

But as I started using the application, I realized that the real problem wasn't simply creating meals with AI.

Planning meals involves many small human decisions: add something, change it, remove it, find an alternative, explore an idea and choose what works.

I didn't want AI to get in the way of those decisions.

My first versions were too eager to use AI.

Changing a meal could trigger recipe and ingredient generation. The application was trying to do too much, too early.

I realized that planning and enrichment are two different problems.

Planning should be fast and lightweight:

Once the plan is ready, AI can do the heavier work: generating recipes and ingredients.

This led to a much simpler flow:

PLAN → decide
GENERATE → enrich
SHOP → buy
COOK → execute

AI became part of the workflow — not the workflow itself.

It's easy to ask: “Where can I add AI?”

A better question is: “Where does AI actually remove friction or create value?”

Sometimes the answer is AI. Sometimes the best solution is simply a fast, deterministic interaction.

One of my biggest technical learnings was the importance of context.

Before asking AI to act, MealMate builds the relevant context: the current request, meal and slot, conversation, preferences and dietary constraints.

The AI model is therefore only one component of a larger context pipeline.

AI is not the product.
The user's problem is the product.

AI should step in where it improves the experience — and stay out of the way when it doesn't.

Start with the problem. Add AI where it helps.