AI-built yard app shows promise and limits of vibe coding

Google Gemini helped a tech reviewer create a working Android yard-care app from a text prompt, showing how quickly AI tools can turn an idea into software. The project also exposed a familiar problem with AI-assisted coding: the first version worked, but it missed practical details that mattered in actual use, according to The Verge’s Allison Johnson.

Johnson wrote for The Verge that she used Google’s AI Studio to build an app for managing an overgrown home yard. She wanted a tool that could organize yard chores, suggest next steps, account for weather and use image recognition to assess sick plants.

The app appeared in a preview window within minutes, Johnson reported. Gemini also flagged a bug and offered a button to fix it, later saying it had resolved the issue after 233 seconds.

A fast build with rough edges

Johnson said the first version included sections for plant zones and an AI plant-doctor feature that accepted photos from a phone. But she found the interface hard to read because Gemini chose a dark purple and red color scheme with illegible text.

After she asked for a brighter design, Gemini changed the colors and added a greeting that called the user a “Gardeneer,” Johnson wrote. She kept the term and the basic structure, while asking for changes such as live weather data instead of preset climate profiles.

Johnson said the app became more frustrating after she loaded it onto her phone. It lacked basic task-management functions, including editing chores and scheduling them for specific days. She also found that one-time and recurring tasks did not sort properly, and that a date picker failed to let her choose a date.

The process required repeated exchanges with Gemini, Johnson wrote. Each fix meant asking for a change, waiting for the AI tool to implement it, deleting the old app from her phone and installing a new version.

The plant diagnosis worked better

The most useful part of the project was the image-based plant assessment, according to Johnson. She uploaded a photo of a struggling rhododendron, and Gemini returned a detailed assessment, possible causes and actions that could be added to her planner.

Johnson wrote that Gemini blamed the plant’s condition on landscape fabric and river rock that a landscaper had previously installed to control weeds. According to Johnson, Gemini said the fabric could have clogged over time, dried the roots and trapped heat from sun-warmed rocks above them.

Johnson then followed the app’s recommendations by pulling back river rock, cutting away landscape fabric and pruning parts of the shrub. She also worked on another rock bed where weeds had started growing on top of the fabric, she wrote.

After a few days, Johnson reported seeing new leaves on one branch of the rhododendron. She said that result suggested Gemini’s gardening advice may have been useful, even if the app itself remained unfinished.

A narrower app may be the better app

Johnson wrote that the project taught her to define an app’s purpose more clearly before asking AI to build it. She concluded that her yard tool may not need plant zones and many custom features for a small urban-suburban backyard.

She also said the experience showed a gap between AI-generated software and the physical world. Johnson pointed to Gemini’s unreadable color choices, its preference for generalized weather settings and an earlier test app that she said tried to simulate checking a grocery chain’s seasonal promotion rather than actually verifying it online.

Johnson said she does not expect her “Gardeneering” app to reach the Play Store. For now, she wrote, the better version may be a simpler mix of Gemini chat and a basic to-do list.

This story draws on original reporting from The Verge.