We are living through a time where Artificial Intelligence is advancing at such a breakneck speed that what I write today might be obsolete by next month, so this blog post might turn into a digital relic as well and be laughed at.
Whether we are ready or not, AI has become the silent partner in businesses of every size, including independent shops like ours. At Darabara, we’ve adopted a routine practice of using AI to continuously refine our storefront. It starts in an unlikely place: our spam folder. Yes, we frequently receive unsolicited emails from "Shopify Experts" auditing our site. Instead of deleting them, we feed their critiques into different AI models using a prompt like this:
"We received this email from XXX on XXX. Here is our store's domain, and attached are our store's XXX that you may find relevant. Given all the context, is there truth to this email? Is this a generic sales pitch, or is there a genuine flaw here? If yes, list some action items for us to review."
Most of the discussions with AI are open-ended—it's a way to turn noise into a potential roadmap. As we won’t blindly adhere to AI-generated action items, these chats may lead to no action, or actionable improvements on our ad campaigns, email templates, theme settings, or storefront wording.
This has been a routine fun activity we've taken on for over the past year.
The Challenge: "Free Shipping" in a Multi-Currency Store
A recent iteration of this process highlighted a gap in our Buyer Confidence flow. We know that shipping costs are a key factor for cart abandonment.
Darabara was using the free version of a popular Shopify App called Essential Free Shipping. It’s a fantastic tool, but the free tier had a few limitations that clashed with our specific needs. We want the progress bar to live inside our slide-out cart drawer (which is in a paid plan), and we wanted a dynamic logic as our free shipping rules are across different egions:
- US $49 threshold for customers in the US & Canada.
- US $79 threshold for international customers.
- Multi-currency support that responds to customer's local currency and alignes perfectly with Shopify’s exchange rates.
Our volume is still growing, and adding another monthly subscription for "Pro" app features didn't make financial sense yet. So, while on the checkout flow topic with AI, in a whim, I brought up this question "Can I just write a snippet to add into the theme and meet our needs?"
Building It with Iterations (Vibe Coding?)
I sat down with Google Gemini to write a custom snippet for our Vessel theme. (We were slow in renovating our storefront; the new theme was updated in late 2025, about half a year after the Vessel Theme was introduced in Shopify's Summer 2025 update.) The goal was to write something lightweight that matched our brand aesthetic.
However, working with AI isn't magic; it requires guidance. Over the last year, we've developed a few pre-requisite actions to ensure the code is safe and effective:
- Always Make Copies: Never edit the live theme directly. Always create a duplicate of the theme and work and test in the copy.
- Context, Context, and Context: The more accurate description of context (such as theme version, color scheme, date and time, etc.), the higher the accuracy of AI output in general.
- Understand the Intent: Most models generate a summary to explain why it used specific functions. I always read them first before running back to test the code, and this allows me to provide accurate directional changes for iterations if the first draft isn't perfect.
The whole process was a chain of conversation. We didn't get it 100% right on the first shot, like most previous tasks. I was unsure if it could achieve the end goal all at once, so I started testing simpler code with the basic functionality, then iterated through currency math bugs and placement issues while adding business-context conditions.
A glimpse of how this "AI Model” programming session looked:
Why not use Shopify's Built-in AI?
You might wonder why we don’t use Shopify's native AI tool for this kind of code editing. While tools like Shopify Sidekick are splendid for analyzing our store data and compiling reports, the code generation process inside the Theme Editor to me can sometimes feel "over-engineered." It often tries to build UI blocks when all we need is a simple, behind-the-scene snippet for the theme. For cases like this, I prefer the faster "Edit Code" route.
Reflection: The Ghost in the Machine
As we wrote the final snippets and tested them in the live theme, everything clicked and it worked exactly as intended.
In that moment of victory, the feeling of joy just spilled over. My immediate instinct was to turn to my collaborator on this task, high-five them, and share the excitement of a job well done. I wanted to say, "Look! We nailed it!"
But my colleague was an AI model.
I paused. Obviously, I know it doesn't feel what I feel; it doesn't get a dopamine hit when the code compiles without errors (I know, this might be a 5-minute coffee break coding session for some experts and is not worth any excitement). But the urge to share that human moment of connection was real, and realizing that my "partner" couldn't feel it back, at least for now, gave me a complicated, almost heavy feeling.
This disconnect brought to mind a recent podcast interview I listened to featuring Ilya Sutskever, one of the prominent scientists in AI field, in which he discussed the path to AGI. He suggests that true AGI/safe AGI might actually require a form of sentience. His argument is that for an AI to truly understand and care about human values, it needs to experience a form of empathy, perhaps similar to how our mirror neurons work.
It’s a thought that has stuck in my head. What would a sentient AI actually look like?
Recently, I revisited Kazuo Ishiguro’s novel, Klara and the Sun. In the story, Klara is an "Artificial Friend" (a robot) who walks a long, arduous journey to pray to the Sun for the health of a child she cares for. It struck me that this might be the ultimate assessment of sentience. I think it isn't about having a complete knowledge base or passing difficult tests or perfect world models. Humans live whole lives full of bias and blank spaces in our knowledge.
Perhaps sentience is the ability to pray. To irrationally hope. To appeal to the Sun for the benefit of something—not even necessarily a human—that it cares about. Although readers wouldn't know if Klara took the action due to her program codes' instruction, but If an AI can feel that specific kind of desperate care, can we say it has a "heart"? Will it be excited like me if our code snippet executed without throwing errors?
Looking Forward
If that day comes, does it mean AI has consciousness? Will they assume legal responsibility for their code decisions?
On the opposite side, how do we measure AI's feelings? Will we know they are intentionally lying?
I strongly sense that the future isn't AI becoming human, but humans leaning closer to the machine through advancements in biomechatronics (I added a Wikipedia link because I knew there ought to be a term for this, but had to describe the concept to Google to find it.). We are merging.
But perhaps that is too heavy a reflection for a post about a free-shipping bar in slide-in cart. I admit, I might just be too hyped for the new Ghost in the Shell series by Science Saru.
Thanks for reading. We will continue refining our storefront, one unsolicited email and one conversation at a time.
