AI services we were glad to introduce for Shopify and ecommerce development — Spring 2025 edition
Hello. I'm Nosta Representative and CTO Takasaki.
There’s a lot to see in the recent evolution of AI.
We are also actively utilizing AI and are researching more resources to interact with our customers and develop new features.
In this article, we introduce AI-related services that were good for us to introduce.
Contents [ hidden display ]
Coding Support (editor)
Until now, it was an image that AI supported humans but this year the function of writing code in agent mode is incorporated into a code editor and human confirmation of code written by AI.
This really has the impact of changing our senses so far, or hiring one person.
Among them, many of our team members use the following editors:
Cursor
VSCode + Github Copilot
URL:https://code.visualstudio.com/
I've tried both of them, but if they are paid versions, it feels like there is no big difference.
Cursor feels a little quicker to release new features.
Also, if you ask an AI to do the whole thing in a rounded way, you have to split your project into tasks and specify the order of execution, which is similar to when you interact with human staff.
terminal
Warp
When an AI-loaded terminal loads the error message that comes out when a local server is launched, it automatically eliminates the cause of the error and thus it becomes much faster to solve errors caused by the server.
In addition, since it automatically eliminates errors when building an environment, the stress of constructing an environment is greatly reduced when challenging new technologies.
MCP server
MCP server is a tool that allows AI to perform various functions (browser operation, file creation, search etc. )
I use it from the chat column in AI code editor.
Among them, the following MCP server was convenient.
Shopify Dev MCP Server
URL:https://github.com/Shopify/dev-mcp?tab=readme-ov-file#setup
When you put this tool into your AI code editor, it is possible for the AI to read Shopify’s official technical documentation and generate code from official information.
As a result, the code generated by AI is more accurate to the code that Shopify recommends.
It’s especially useful when writing queries and answers questions like “How do I get the metafield of product information?” that you can only answer if you understand Shopify API data structures.
The effort to write GraphQL for pulling data from Shopify has been greatly reduced.
Microsoft Playwright MCP Server
URL：https://github.com/microsoft/playwright-mcp
By introducing this, you can direct E2E test (browser test) to AI in natural language by chatting.
EC site creation can cause fatal errors such as not displaying a cart button due to CSS interference, but it solves the problem that it is difficult to check all operation by human every time.
You can access this URL and check if a product is added to your cart, or whatever else you want to test in the chat section of the editor.
If you want to test it in the chat section, an AI will launch and test your browser for you.
code review
Code Rabit
URL:https://www.coderabbit.ai/ja
When a pull request is sent to GitHub, it will automatically provide you with a summary and code review of the pull reek.
It’s useful because it can quickly check the misspelling of a variable and other areas that are difficult to notice by humans.
Example of AI Code Review
what someone tried and gave up on
Dify
With this service, you can easily create AI-related apps.
In our company, Notion has been able to communicate with clients and have technical knowledge so that AI can learn from the in-house Notion.We tried to help make recognition common among members and to desubscribe from correspondence, but perhaps there is also a problem of data behavior of our Notion, and we overlooked the introduction as an internal RAG with lower accuracy than expected values.
I think that it is evolving again over time, so I am looking forward to the future!
Figma Automatically Converts To Code
Since there is no qualia (sensory quality experienced by consciousness, subjective sense and emotion) in AI, I think that the work that requires a qualia such as converting visual information into code tends to be bad.
There are a lot of services where AI automatically codes the designs created by Figma, and if you want to fix this generated code, I thought it would be faster to write your own code completion function from scratch.
This is also expected to power up in the future.
service someone tried
untested drop-off
There is an Agent mode in the code editor itself, and if we introduce other services similar to AI Agents, I think it will be a source of confusion.
Above all: a good book to read.
In thinking about AI-driven development, we asked “What kind of system is AI and how it can provide good performance?”. Code × AI – Introduction to Generated Artificial Intelligence Practice for Software Developers This book is very helpful, and I recommend it.
That’s all we’ve found so far, and if there is anything good about the service that isn’t mentioned here, please let us know by using our feedback form or SNS comments.
If we have new learnings, I would like to share them with you from time to time.