A writer used Google's Gemini AI to build a yard-care app, discovering both the promise and the friction of AI-assisted coding — and that her landscaper's rock-and-fabric solution was slowly suffocating her plants.

What happened

The author prompted Gemini to create an Android app to manage yard chores, get plant-care recommendations, consider weather, and diagnose plant problems via image recognition. The AI generated a working prototype in minutes, but the author spent hours iterating on bugs, design flaws, and missing features like the inability to edit chores once created or pick specific dates for tasks.

Why it matters

The story illustrates the real-world gap between AI-generated code and usable software — a working prototype is not a finished product. The author's most useful feature turned out to be the AI plant doctor, which could analyze photos and recommend action items, showing where AI excels (diagnosis) versus where it struggles (building coherent, editable task management). This reflects broader questions about how much iteration users must still do when building with AI.

What to watch

The author's yard diagnosis — that landscape fabric and river rock were clogging the soil and cooking roots — led her to spend a full afternoon manually removing the rocks and fabric. The experience reveals the irony that while she burned computing resources (described as equivalent to a microwave dinner's worth of electricity) refining an app, the AI's most valuable contribution was identifying a real, fixable problem in her physical yard.