Helping people rewild their yards: The AI journey continues

Helping people rewild their yards: The AI journey continues

I started on the AI journey with a few goals in mind: to figure out where the edges are for people like me, and to build two climate-resilient apps.

I grew up gardening, and I love everything about it. My yard growing up was a food forest, and our street had a massive community garden. All summer we’d snack on raspberries, blueberries, cherries, Santa Rosa plums, apples, pears, and when the fall was settled in and winter was on the horizon, we’d get to tackle the corn stalks.

We gardened without pesticides. Instead, we had flowers growing everywhere, planted companion plants, repellents (marigolds), and traps (calendula and nasturtiums) to keep the insects away.

I grew up understanding that sometimes that mess we are so eager to clean up is actually our pollinators' winter home.

As a gardener today, I’ve realised I can’t garden the same way; in fact, the growing zones were recently adjusted. I live in Portland, and my tiny part of the world is a whole growing zone different from the outlying areas because of the heat island effect. Portland, Oregon is the same growing zone as San Antonio, Texas.

Climate Resilient Practices Are the Best Way Forward

So I asked myself, “How do we continue to grow and thrive in a climate resilient way?” I wanted a way for new and experienced gardeners to be able to adapt to the changing environment. Food in our yards is food for the critters, birds, bees, and other insects too. We are part of the ecosystem.

Learning to code using AI meant I had to learn to be comfortable with being uncomfortable. In fact, I’d imagine this how it feels for new gardeners. I was terrified of deleting code, going in and changing the code myself and I was worried I’d break it.

Just like gardening combinations to get new gardeners started, I figured out how to get started coding with assistance. I wouldn't call it vibe-coding. I'd call it product with a systematic approach to solving problems. Hmm, doesn't have the same ring to it. Anyway, I tinkered a bit until I figured out a requirements doc format that works for me. I answer the following questions:

• How will it be used?
• Who’s it for?
• Multiple users with auth? OR no users, all public
• Using external data through APIs?
• Storing user data in your own database?

It seems simple, but any seasoned Product Manager will tell you the devil is in fact in the details. So, for a simple chore app, you'd do the following:

Create a Chore app for people with ADHD. It should include the following:
1. User selects what category of chores they wish to do daily, weekly, monthly, or seasonally. The list can be found in this PDF (https://i.pinimg.com/originals/01/46/10/01461068831a5193bb0ed353e00bedfe.jpg). 
2. Provide 2 choices in a choose-your-own-adventure experience with a fun fact surfaced after each selected chore is completed. Please use [fun facts found in this PDF](https://factrepublic.com/wp-content/uploads/2017/03/1000-Interesting-Facts-to-Blow-Your-Mind.pdf) 
3. Connect to the music streaming app of your choice to help keep the user engaged and motivated with their chores.

This app should be clean, crisp, and engaging.

Systematically building an app

All I wanted was the right framework. I tried a few tools until I got a starting point I liked. With the right framework, I systematically go through section by section and build out the details. Doing it this way helped me track how the code was laid out, and I could start to pattern-match how the code worked together. It also reduces the context window, which makes token usage more efficient. All around win!

I still don’t know what I don’t know

In many respects, this was the fun and easy part. Aside from tabs. For the love, I could not figure out the right prompt to get Lovable, Windsurf, Co-Pilot, or Cursor to do tabs.

I started to get comfortable checking logs, learning to read the code (I’m still terrible at it, but I am making progress), and identifying the cause of an issue, even if I don't understand how to prompt a solution.

Then I got overconfident and decided to start pulling data. Oof, some of the biggest challenges came from trying to pull from government sites. The way Cursor struggled with Celsius and Fahrenheit was entertaining. This turned into a phone a friend scenario. With help, I sorted out how to use Bruno, so I can prompt AI to pull from the API correctly.

All of that got me garden spaces with hover icons and a set of containers. Think of it as a framework for gardening.

I have a context aware chatbot to help users create a container space or to ask it garden questions. My test case was making a salsa garden because I wanted to see if it could put together something specific. As an apartment and small garden space dweller, themed containers were perfection! Salads and salsas taste better when you grow the ingredients. I don't make the rules.

The context aware piece was another opportunity to learn. AI has no concept of size or gardening, so I taught a GPT. I seeded the model with permaculture and indigenous practices for gardening, added reference links to programs, and created a rule that the Sprout bot needs to link to references so users can fact-check if needed.

By the end of it, I built an alpha/beta garden app Root & Stem and started on my climate resilient home app. I thought the garden app was a learning curve. I was not prepared for the shenanigans that happened with my next app! A tale for another time.

You can try out Root & Stem here. If something seems off, assume it is the app. Let me know about any bugs, features, or questions at [e-mail].

If you are interested in being a Beta tester for the home app, where you’ll get insights on climate resilient updates, both big and small, sign up here.

I'm hoping I can make your gardening journey an adventure.