Welcome to the Tech Talks series, where we give you an inside look at PayPay’s product development philosophy and atmosphere through the voices of our diverse product members from approximately 50 countries and regions around the world.
This time, we interviewed two engineers from the AI Utilization Office, a department that promotes the use of AI across the entire PayPay Group. They shared the behind-the-scenes insights of developing internal AI tools and the unique rewards of being an AI engineer at PayPay.
Junji Atsumi
Engineering Team, AI Utilization Office, System Division, Product Group
He previously gained experience in system development and operations at an e-commerce company. He then moved to a system integrator, where he was involved in system design, development, operations, and maintenance. After joining a smartphone game company as an internal IT engineer responsible for improving business efficiency through the development, integration, and automation of internal tools, he took on his current role in February 2020. He is now in charge of developing AI tools for employees, including an AI chatbot, at the AI Utilization Office.
Masahiro Hori
Engineering Team, AI Utilization Office, System Division, Product Group
After graduating from university, he worked at a system integrator, where he was responsible for both front-end and back-end development. He then transitioned to a securities firm, gaining experience in operating large-scale financial systems. He has been in his current position since September 2022. At the AI Utilization Office, his responsibilities include building the necessary platforms for AI utilization.
Overseeing AI Promotion, Development, and New Technology Verification Across the Company
Could you tell us about the role and mission of the AI Utilization Office?
Junji:
Our organization’s mission is to drive the adoption of AI throughout the entire PayPay Group. It originally began as a cross-departmental initiative started by volunteers within the company in early 2023, when generative AI suddenly garnered significant attention. As the importance of generative AI grew, the initiative expanded and was formalized into an official organization.
Our current specific roles are centered around three main pillars: AI Utilization Promotion, where we formulate the roadmap for AI adoption; AI Development, where we develop internal applications that leverage AI; and New Technology Verification, where we quickly test and explore pathways to practical application for the constantly evolving field of AI technology.
What roles do the two of you play?
Masahiro:
Both Junji and I are AI engineers driving the development of various tools. My focus is primarily on the platform and infrastructure side, which serves as the foundation for AI utilization. He, on the other hand, often handles the development of front-end tools that employees interact with directly, like the Slack-based chatbot (Slackbot for short) that we’ll discuss later.
The Number of AI Users Company-Wide Skyrockets by Over 5x in Two Years
Could you share a project that has been particularly memorable for you recently?
Junji:
A memorable project for me was the development of our Slackbot, which was created to enable employees to use AI more casually and securely. The single most important thing I focused on during development was how to lower the “psychological barrier” that employees had towards AI. When the project first started, there was still a vague sense of apprehension and resistance to generative AI within the company. So, we started by providing a secure API, where input information is not used for retraining, on the familiar Slack UI that many employees use daily.
First, we had to carefully build a common understanding, starting with the very definition of a “prompt.” We made a concerted effort, such as creating a collection of samples to help people visualize concrete use cases and having the Slackbot permanently reside in various channels so that employees could naturally see how others were using it.
From a technical standpoint, we designed it with future scalability in mind and adopted a microservices architecture. Specifically, we built services like RAG, a batch for saving training data, and a container for interacting with the LLM as independent components. We designed it so that these components could also function as an API hub where they all connect. To improve response accuracy, we chunked the training data from our internal Wiki and company regulations by headings and chapters, creating cleaner, less noisy text. As a result of these efforts, we have successfully obtained multiple patents for the internal information training (RAG) component.
Masahiro:
For me, the development of the “LLM API Hub” was a memorable project. This hub serves as a foundation for efficiently utilizing the various AI tools used internally, including the Slackbot developed by him, while ensuring security and governance. I had never experienced AI system development on this scale before, and I vividly remember feeling a mix of responsibility and excitement. The project kicked off with the goals of maximizing the unique characteristics of each LLM, optimizing costs, and avoiding over-reliance on any single tool. Above all, we aimed to create an environment where developers and business unit members could confidently incorporate the latest AI technology into their work.
A distinctive feature of this project was the introduction of a design methodology called DesignDocs. Unlike a typical design document, a DesignDoc is a method for articulating the design philosophy and decision-making process, including “why a particular technology or architecture was chosen.” By adopting this approach, we eliminated any discrepancies among development team members regarding our starting point—the “why”—which enabled smooth communication. In particular, conducting a short but intensive review session with all members over about a week allowed us to align on the same goal from the project’s initial phase and proceed with development with an unwavering focus.
What were some of the biggest challenges you faced while working on these projects?
Masahiro:
I hit a wall with the development language. Initially, we planned to use an existing, company-approved library for the API hub’s authentication process. However, midway through development, we discovered that a part of the library was not compatible with Python.
The original library was written in Java, so we were suddenly forced to implement that part ourselves in Python. I was worried about the impact on the schedule, but this is where AI truly shined. I asked the Slackbot and GitHub Copilot, which I regularly use for development, to rewrite the code, and they generated nearly perfect code. Thanks to AI, we were able to overcome this crisis without any schedule delays. It was a moment where I truly rediscovered the value of the tools we were creating, right there in the midst of development.
Our hard work paid off, and the internal response immediately after the release was tremendous. Right after we posted the announcement, we were flooded with inquiries about the specifications. We also had opportunities to hear directly from management that they “have high expectations,” and we could feel the immense anticipation and passion from the entire company.
Junji:
For me, the struggle was less about technical challenges and more about the “marketing” aspect—figuring out how to get people to actually use the tool. Simply releasing a convenient tool isn’t enough to reach employees who are busy with their daily tasks. Even if we made a company-wide announcement, it would quickly get lost in the flood of other information.
To overcome this barrier, we launched a company-wide AI utilization promotion project. We recruited volunteers from each department who were interested in using AI and had them think about specific use cases like, “Here’s how my department could use this,” or “I want to automate this task.” They also took on the role of evangelists within their own departments. Thanks to these PR activities, which engineers alone couldn’t cover, our colleagues across the company helped us dramatically close the distance between AI and employees, which in turn led to widespread adoption.
In the approximately two years since Slackbot’s release, the number of internal AI users has increased by nearly fivefold, growing 1.5-fold year-on-year. This has resulted in the reduction of hundreds of thousands of work-hours annually across the company. Every time I see a use case in the company newsletter, like “I was able to improve my work this much using AI!”, I feel truly glad that we did this project. We will continue to aim for even greater AI adoption and further reductions in work-hours.
Riding the Wave of AI to Create the New Standard of the Future
What do you find appealing about being involved in AI development at PayPay?
Masahiro:
The proximity to management and the culture that makes it easy for engineers to make bottom-up proposals for what they want to do is a huge draw. The company is also very open to new technologies and tools. For example, Cursor, an AI coding tool that’s gaining traction, was quickly considered for adoption thanks to suggestions from the ground level. As long as security requirements are met, we can take on new challenges with incredible speed.
Junji:
The fact that the entire PayPay Group is strongly promoting AI utilization is the best possible tailwind for engineers. A clear policy has been set forth to move beyond the development of conversational AI tools and aim for the large-scale development of AI agents. This means that if you have a strong desire to be involved in AI development, the opportunities are limitless. I feel it’s an incredibly privileged environment to be able to immerse myself in development while feeling the high expectations from the company and my team members.
Finally, do you have a message for our readers?
Junji:
PayPay is a company with 70 million users (as of July 2025), yet our development environment is brimming with speed and autonomy. Our team, in particular, is a lean, highly skilled team tackling a major cross-company mission. You get to experience the scale that comes with a large company—in terms of the number of employee users and the investment in AI—while simultaneously moving quickly on unprecedented development projects using the latest technology. I think it’s rare to find an environment where you can enjoy the best of both worlds like this.
Before I joined PayPay, I was one of the users who experienced the “10 Billion Yen Giveaway Campaign” and was amazed by the new payment experience. Now, as a developer, I want to deliver that same excitement to my colleagues within the company. With that in mind, I am now pursuing the new goal of implementing AI agents. I hope you’ll join us in creating exciting products together.
Masahiro:
We’re at a pivotal moment for AI utilization at PayPay. In the fast-paced field of AI, we are shaping the company’s “future standards” with our own hands. I’m personally paying close attention to the AI coding technology that was so helpful in my recent project, and I want to build a system internally that allows the browser to operate automatically, much like OpenAI’s AI agent, “Operator.” I invite you to come and share this exciting experience with us.
Current job openings
*Job openings and employee affiliations are current as of the time of the interview.