A 20-Year Return: From CEO to Developer, Starting a New Journey with an AI Chatbot

In October 2005, I left my life in Frankfurt, Germany, and founded a company called M-BIZ Global Company Limited in Kingston-Upon-Thames, a beautiful town on the outskirts of London. Everything started with a single contract signed with Samsung Mobile (HQ in Korea).

Our business model was an innovative mobile game sales method called 'Try and Buy.' Users could try popular games like Tetris for about 60 seconds (Try), and after the game automatically locked, they could purchase the full version through a premium SMS payment (Buy). As these games were embedded in feature phones before the smartphone era, the results were beyond imagination.

A single software developer who had moved from South Korea to Europe was instantly transformed into the CEO and Founder of a global company, generating massive revenue through Samsung mobile phones in about 60 countries worldwide. As the company grew, I hired new colleagues, and naturally, I drifted further and further away from the world of software development that I loved.

Nearly 20 years have passed since then. The business has mostly been wound down, and the last senior developer who stood by me also bid a bittersweet farewell at the end of June 2025. Although I had shifted the business's focus to AI chatbots, I had to spend most of my time wrapping up other scattered projects.

Above all, having been away from the development scene for nearly 19 years, starting to code again was an immense challenge. The world was now filled with technologies completely different from what I knew, such as AWS, Node.js, and TypeScript.

Finally, on July 1, 2025, with the feeling that I could not retreat any further, I began to develop an AI chatbot service called 'AI Talk' entirely on my own. While everyone around me recommended Claude Code, I happened to have a $300 free credit from Google Cloud, so I decided to use Gemini CLI for practice.

With the emergence of ChatGPT, I personally became deeply interested in Retrieval-Augmented Generation (RAG) technology. I was completely engrossed in the process of handling unfamiliar Python code, embedding data into a Vector DB like Pinecone, and generating automated responses with the OpenAI API based on data selected through cosine similarity. Of course, RAG services were already well-established through LangChain, and I also aimed to build a service like Chatbase myself.

While planning a B2B business model, I took notice of the Shopify App Store as a marketplace comparable to Google Play or the Apple App Store. The policy of waiving the 15% commission until an app's lifetime gross revenue reaches $1 million USD was particularly attractive. Furthermore, the default Remix framework and the Polaris React component library provided an excellent development environment, significantly easing the burden of UI/UX design for the admin interface.

However, other AI chatbot services that used RAG had one clear drawback: the inconvenience of having to constantly retrain the data whenever a store's product information was added or updated. I, too, thought this was an unavoidable technical limitation. But after learning about Shopify's Storefront API (MCP), I made the bold decision to abandon RAG and replace it entirely with the Storefront API.

Understanding how MCP worked from the beginning was not easy. I decided to use the OpenAI API, which was slightly more affordable and accessible than the Claude API. I also added an option for store owners to input their own OpenAI API Key to increase the default usage limit by tenfold.

The performance of the implemented MCP was truly satisfying. For general queries like "Hello?", OpenAI handled the response naturally. For specific questions like, "What products are available for purchase?", it searched the store's entire internal data in real-time to provide the most accurate answer.

Most importantly, this system has a special feature that naturally guides the user toward placing an order. This was a key differentiator made possible by leveraging MCP, a feature difficult to support with a RAG approach that simply retrieves and presents information.

If you'd like to experience the live demo for yourself, please visit the link below and click the 'View demo store' button on the left.

https://apps.shopify.com/ai-talk-checkout-assistant

As a next step, I am planning a service to manage customer orders through the Customer accounts MCP server. Once this feature is added, the AI assistant will be able to provide more advanced support, such as checking a customer's order status and looking up order details.

Conclusion

The very beginning of all this coding started with Gemini CLI. However, as the number of features to develop grew, I naturally migrated to Claude Code. The important takeaway is that even if you don't fully understand the syntax of TypeScript, developing a Shopify App like this is entirely possible with the help of AI.

So now, after 20 years, I am back to being a solo developer. The support from AI coding assistants and cutting-edge technologies like MCP is making my heart race with excitement once again.

This was also created with the help of Gemini 2.5 pro and the image was generated.