My yard is dying, so I made an app for that
A Verge writer uses Google’s AI Studio to build a yard-care app, then runs into the limits of vibe-coding and app debugging.
Intelligence analysis by GPT-5.4 Mini
The article is a hands-on experiment in building a mobile yard-maintenance app with Gemini. It starts with a working prototype in minutes, then turns into a lesson about awkward design choices, missing features, and the gap between a flashy AI demo and a usable tool.
The writer asked an AI to build a phone app to help with yard chores, like a helper that writes a checklist for a messy garden. It made something quickly, but the first version had bad colors and missing parts, so a human still had to fix it.
Analysis
From weeds to workflow
The piece starts with a simple problem: the author’s yard has become a mess, and the usual approach to keeping track of chores is not cutting it. Instead of using a generic to-do app, she decides to build a custom Android app with Google’s AI Studio. The prompt asks for help organizing many yard tasks, making recommendations, using weather, and analyzing plant photos.
A fast demo, then reality
Gemini produces a working app preview in minutes. It includes separate sections for plant zones and a plant-doctor feature for image-based diagnosis. But the first version is not really usable: the default dark color scheme makes the text hard to read, and the app makes odd choices such as relying on weather “profiles” instead of live weather data.
The hard part is the details
After changing the design and bringing the app onto her phone, the author finds more serious problems. She cannot edit chores after creating them, cannot schedule tasks properly, and tasks do not always land in the right category. The article frames this as the real cost of vibe-coding: a prompt can generate a convincing shell, but a functional app still needs repeated correction, testing, and judgment from a human.
What the experiment reveals
The most useful part of the app turns out to be the plant-doctor feature, but the overall story is less about gardening than about the state of AI-assisted app building. The tools are fast enough to create something tangible, yet they still need a person to notice what is broken, explain what should be different, and keep pushing until the result matches the real-world task.
Key points
- The author uses Google’s AI Studio and Gemini to build a custom Android app for yard care.
- The first version appears quickly, but it has readability and design problems.
- The app needs several fixes before it becomes even partly useful on a phone.
- Missing basics like editing chores and correct task sorting show the limits of vibe-coding.
- The plant-doctor image feature emerges as the most useful part of the app.
If this kind of AI-assisted building keeps improving, people could make small custom apps for everyday problems without hiring a developer. The article suggests that even a messy first draft can become a useful tool when the human keeps refining it.
The piece also shows that a fast AI demo can hide major flaws, like broken editing, bad sorting, and weak defaults. If users trust the first version too much, they may end up with software that looks finished but fails at the job it was meant to do.
