My Dying Yard: The App I Made to Save It

What began as a simple list of yard work tasks unexpectedly evolved into an intricate "vibe-coding" project.

Upon returning to my computer just five minutes after providing Gemini with an extensive prompt, I discovered a functional application in a preview window, alongside an alert indicating a bug.

The message, “~ Channel is unrecoverably broken and will be disposed!” was concerning. Yet, directly beneath it, a button offered a solution to the bug. It struck me as peculiar that, after instructing an AI to construct an entire application from a single prompt, it still required my manual intervention to resolve an issue. Nevertheless, I clicked, and within 233 seconds, Gemini confirmed its success, employing terms such as “blockages” and “race conditions,” none of which I comprehended. The experience was exhilarating.

This marked either my second or third foray into "vibe-coding" an application, depending on whether a previous project, which never advanced beyond the preview stage, is included. That unlaunched web app was designed solely to ascertain if a local upscale grocery store was hosting its annual Peach-o-Rama event (to date, no peaches). Regardless of the count, the current endeavor is considerably more ambitious: an application intended to assist me in managing my challenging yard.

Eight years ago, when my husband and I purchased our home, we largely overlooked the demands of yard maintenance. Our assumption was that beyond basic lawn mowing, shrubs and trees were largely self-sufficient. This neglect persisted until weeds aggressively encroached, rapidly infesting the flower beds lining the house and the property boundaries to a prodigious extent. It became clear that the scope of "yard care" far exceeded our initial expectations.

While we initially achieved some minor victories against the weeds, we ultimately lost the broader struggle, necessitating the hiring of a landscaper. His single visit allowed us to maintain the yard largely on autopilot for several years. This solution proved effective for a time, but eventually, weeds began to resurface, and our shrubs exhibited clear signs of distress. As the weather transitioned to spring this year, I committed myself to understanding and resolving the issues plaguing our yard.

Although I had a preliminary understanding of how to begin, I desired ongoing assistance and a structured approach to organizing the necessary chores. This led me to consider: why not develop an application for this very purpose?

I endeavored to make my prompt as detailed as possible, essentially presenting a list of requirements: manage extensive yard care tasks, provide recommendations, incorporate weather conditions, and utilize image recognition for plant problem diagnosis. I input these specifications into Google’s AI Studio, aiming to produce an Android application that I could install on my phone and use outdoors – precisely where the plants reside. My expectation was that the process would take about an hour, allowing me to dedicate the remainder of the day to documenting my yard's condition and executing the app's instructions.

My initial time estimate proved somewhat inaccurate. While a functional app appeared in the preview window within minutes, logically structured with sections for various plant zones and an AI "plant doctor" for image uploads, it presented a significant aesthetic flaw: its color scheme.

Inexplicably, Gemini had opted for a dark mode interface, featuring dark purple and brick red accent colors. The text was not only illegible but also visually unappealing. I then proposed a white background with light green, pink, and blue hues, emphasizing the importance of human readability. The AI subsequently delivered a more aesthetically pleasing version, complete with an enthusiastic greeting on the app's homescreen: “Welcome Back, Gardeneer!” I found the inventive term "gardeneer" quite appealing, so I decided to retain it.

I largely retained the foundational structure Gemini had generated. However, I did implement some modifications, such as integrating live weather data instead of the somewhat peculiar climate presets the AI initially provided. Gemini seemingly anticipated that I would manually select a "profile" corresponding to the day's weather, which would then inform its watering recommendations. This struck me as an unusual design choice, given the straightforward availability of live weather information via API, and it underscored a recurring need to clarify the distinction between the physical world and theoretical constructs for Gemini. Despite these minor issues, I promptly transferred the app to my phone, eager to begin using it, too thrilled about launching my first application to concern myself with further iterations at that moment.

Nevertheless, I had overlooked several critical deficiencies during my initial review of the app on my laptop. I discovered that I was unable to edit chores once created, nor could I schedule them for specific dates. While I could establish profiles for individual plants and group them by zone, there was no functionality to link them to particular tasks or, indeed, to utilize them effectively in any meaningful way. Furthermore, despite the presence of separate tabs for one-off and recurring tasks, every chore I entered into the app consistently defaulted to the recurring tab, ignoring the intended categorization.

This process devolved into a tedious cycle of iterative adjustments. I would request an update, await Gemini's implementation, then delete the existing app version from my phone and install the new one. Invariably, I would uncover another malfunction, such as a date picker that failed to allow date selection, necessitating another return to the chatbot. Consequently, beyond an unruly yard, I now contended with an equally unruly application. A valuable lesson undoubtedly lies within this experience.

Conversely, the AI plant doctor functionality proved highly effective from the outset. Operating as a simple "Hey Gemini, figure out what’s wrong with this plant" button, I uploaded an image of a struggling rhododendron. Within approximately a minute, it generated a comprehensive health report (diagnosing its condition as critically bad!), outlining probable contributing factors and suggesting actionable steps that I could add to my planner with a tap. This was precisely the kind of assistance I required for my yard.

Our landscaper's "set-it-and-forget-it" solution involved covering the flower beds with landscape fabric and river rock. He assured us this would effectively manage the weed problem for an extended period and that existing plants would remain healthy, further sweetening the deal with a cash discount.

Years later, however, clear issues began to manifest. A bush near our front door developed yellowing leaves and attracted a constant swarm of flies, while the rose bushes became leggy with diminished blooms.

Gemini swiftly attributed these problems to the landscaper-recommended fabric and rocks. It explained that the root system was being suffocated and drying out, likely due to the landscape fabric becoming clogged with dirt over time. Additionally, the sun-baked rocks were effectively cooking the roots from above on hot days. It was no surprise then that our yard appeared so distressed; in fact, it was remarkable that anything had survived at all.

By then, it was too late in the day to commence "Operation: Rhododendron Rescue." After the extensive back-and-forth involved in developing my app, I had inadvertently spent a beautiful afternoon typing prompts into a chat window. Each time I sent Gemini on a new coding task, I was acutely aware that I was likely consuming the energy equivalent of a microwave dinner at some distant data center, perhaps near Spokane. The irony of this situation was not lost on me.

Despite the app's imperfections, the following day I set aside my feature requests and opted to immediately implement Dr. Gemini's urgent recommendations for the rhododendron. I dedicated a strenuous afternoon, accompanied by a podcast, to raking back the river rock, cutting away the landscape fabric, and pruning some of the shrub's more tangled branches. Subsequently, I shifted my focus to another rock bed, which was already overrun with weeds growing atop the fabric itself. A crucial piece of advice: refrain from using extensive landscape fabric in your yard.

Here lies a revelation about yard work that eluded me eight years ago: it is profoundly satisfying.

Despite the exhausting labor under the full sun, and the inevitable expletives that arose when confronting thorny Himalayan blackberry vines, the underlying truth about yard work, unknown to me eight years prior, is its immense satisfaction. The sensation of skillfully extracting a large weed, roots and all, or leveraging a shovel to tear out a blackberry bush, is uniquely gratifying. While weeding can be arduous, it possesses an addictive quality; once engaged, I consistently find it easy to persuade myself to continue for "just another 20 minutes," even when it's time to conclude.

I eventually concluded my day, then opened my app to mark off several completed yard chores. Having spent hours immersed in the literal weeds of my property, a new list of feature requests formed in my mind. I desired continuous assistance from Gemini throughout the plant revival process, rather than merely a singular diagnosis. Furthermore, while the concept of organizing my yard by zones appealed to my Type A personality, I questioned its practical utility for a small urban-suburban backyard, as opposed to, say, Central Park. It prompted me to consider if a simpler approach, perhaps just a Gemini chat and a Google Keep to-do list, might have been equally effective.

While I doubt my "Gardeneering" app will ever reach the Play Store, its creation has proven remarkably instructive. Articulating the profound experience of watching a computer transform a text prompt into functional software is challenging, akin to recounting a vivid dream. However, it underscores the necessity of approaching such projects with a crystal-clear vision of the problem the app is intended to solve. I realize now that a more thorough upfront effort to define my needs would have significantly reduced the iterative back-and-forth.

My "vibe-coding" journey also illuminated a truth I understood intellectually but hadn't fully internalized: AI lacks genuine comprehension of the real world. It unhesitating