On June 11, 2025, Mercari launched an internal radio-style talk session series called “Leaders Talk! AI/LLM Behind the Scenes Open Door.” The guest for the second session was Shunya Kimura (@kimuras), Executive Officer and CTO of Japan Business, with @gomichan from the AI/LLM Office returning as moderator.

During the recording, Kimura spoke about the sense of hopelessness he once felt before leading generative AI adoption, along with the thinking behind it. The talk drew in members who also felt uneasy about the rapid pace of AI development and implementation.

We bring you the conversation with the same sense of presence as if you were right there listening.

*This article was composed and written using AI tools based on the day’s audio recording. This English version was translated from Japanese using a custom generative AI model interface trained using Mercari style guides, glossaries, and past Mercan articles.

CTO @kimuras on how he uses generative AI

@gomichan: To jump right in, @kimuras, how do you usually use generative AI?

@kimuras: Ever since ChatGPT came out, I’ve been using it a lot, both for work and in my private life. I especially use deep research every single day.

@gomichan: What kinds of things do you ask it about?

@kimuras: As the CTO of both Marketplace and Fintech, there are a lot of conversations flying around, and sometimes I can’t immediately catch the nuances. But thanks to generative AI tools, I can now keep up in real time with discussions I couldn’t follow before.

And recently, since technology is advancing at an unbelievable speed, I’ve been using deep research to stay up to date. I’m also a heavy user of Notebook LM.

@gomichan: How do you use Notebook LM?

@kimuras: I usually feed it documents I’ve compiled during my research at night, then listen to them as audio output on my morning commute. Before meetings, I also let it read the minutes so I can think about how I want to shape the discussion. Lately, I’ve been converting update info from X into audio as well, and listening to it on my way home.

@gomichan: I see. So you prepare in advance and then listen to it all on your commute.

@kimuras: Yeah, it’s made my input much more efficient.

From hopelessness to hope

@gomichan: I feel like you encountered AI really early on, @kimuras. How did you honestly feel when AI tools started being released in earnest?

@kimuras: Generative AI actually has a long history—before ChatGPT, chatbots were already trending. For example, we were using technologies like seq2seq, but conversations didn’t flow as well as they do now. That was probably about 10 years ago.

@gomichan: A whole 10 years ago?

@kimuras: At the time, I thought image generation still had a lot of potential for the future. But when it came to conversation, I felt it would be difficult to raise the quality enough for commercial use. But when ChatGPT 3.5 came out, I thought, “Wait, something’s changing here.” Of course, since ChatGPT is still a language model, unless you feed it actual sources, it tends to come up with “plausible but not factual” statements.

@gomichan: Right, what we call hallucinations.

@kimuras: Exactly. Still, it was shocking how well the conversation flowed. That’s when I started using it regularly—about a year ago.

@gomichan: Since ChatGPT came out at the end of 2022, it’s already been two years—that’s surprising. A lot has changed since then. Something I really wanted to ask today; when I joined in July 2024, hardly anyone inside the company was interested in AI. What was going through your mind at the time?

@kimuras: Honestly, at first I felt a sense of hopelessness. I think there are both positive and negative forms of hopelessness, and at that time I felt a mix of both.

@gomichan: What was behind that feeling?

@kimuras: LLMs could do so much that it felt almost hopeless—like the skills and experience people had built up over time might no longer be needed in any profession. A year ago, I was deeply worried that this new technology might bring such a huge impact that it would force humanity to question the very value of work and of living as human beings.

@gomichan: I think that uneasiness was strong across society as a whole too.

@kimuras: Historically, people have always felt uneasy about technological progress—this has been a constant throughout history. That itself isn’t negative; I think it exists to help us safeguard the future. So this past year, I’ve been grappling with how to overcome that fear.

If you try AI and stop right away, you lose out by developing a needless fear of it. But if you approach it with the mindset of “letting it integrate,” it quickly becomes a positive thing. Even with fear and uneasiness, if you keep using AI, I believe you’ll eventually hit a breakthrough.

“It begins with prompts, and ends with prompts.”

@gomichan: Lately I’ve come full circle and realized prompts are incredibly important. For example, I think the real value lies in the workflows, processes, and ways of thinking of people doing excellent professional work.

@kimuras: Exactly. To create a great product and deliver more value to users, you need that underlying knowledge of specialized processes—otherwise, you can’t reach that goal.

@gomichan: Thinking about it that way, which direction do you see Mercari heading in?

@kimuras: That’s a good question. Basically, it’s about optimizing each task so we can focus more on thinking. That’s probably the kind of change everyone already imagines. But as I’ve said before, to release a product it’s crucial that the entire company works as one. Change isn’t only happening on the product and engineering side.

@gomichan: Like in the Security division, for example. If any one part is missing, it won’t work.

@kimuras: Compliance and legal checks, customer service, and more—all of these need to come together company-wide to build our services. As the barriers to expertise lower, I think we’ll see more and more collaboration.

@gomichan: Yeah, totally.

@kimuras: And when collaboration increases, I think all members will start trying out things like service proposals and mockup creation, and we’ll all create new services and value together. Everyone, regardless of their role, will feel more strongly that they’re providing value to users, and we’ll see more proposals toward that.

@gomichan: I see. From the perspective of individual employees, what do you think they should be doing?

@kimuras: The most important thing is simply to master AI. First, master AI within your own area. Once you feel empowered by it, you can then step into other areas and try using it there too. Each of us needs to tackle our roles and these new technologies head-on, and really optimize what we do—otherwise, we can’t move to the next stage.

@gomichan: You often talk about doing a “deep dive.” What does that look like in practice?

@kimuras: Fundamentally, it means mastering prompts. I want people to be able to use AI as a true tool. That means thoroughly using it until they understand both the pros and cons, and constantly asking, in any kind of work, “Can AI do this?” Recently there have been many cases of people developing with AI without relying on engineers. I think it’s important not to hesitate just because you think, “I’m not an engineer.” Instead, you should catch up on the information and challenge yourself.

@gomichan: In the end, the people who actually understand the work create better prompts, since they know the job.

@kimuras: Exactly. In the AI/LLM era, even people who aren’t engineers will be able to develop a fair amount on their own, so I really want everyone to engage with it.

@gomichan: I’ve been reminded lately of the importance of prompts—of verbalizing thought processes and iterating through trial and error. If you just use it in a straightforward way, you can probably get a decent result—say, 60 out of 100—but I think you need to refine your approach to really boost the final quality.

The prompt I’m most proud of creating recently is one that gets the AI to brainstorm topics I’d like. I had the AI trace my thought process, and it generated an output in about four iterations that would have otherwise taken me half a day to develop—it was fascinating.

@kimuras: That’s great! I think it’s also important that memory accumulates through past conversations and inputs. When you really use an AI tool to the fullest, it becomes personalized, and that feels like an asset.

I often get asked, “What should I do to make use of LLMs?” I’ve thought about it a lot myself, and lately I feel like it really does begin and end with prompts.

Growing Mercari by “learning together”

@gomichan: When members feel afraid of learning new technologies, what can leaders do to support them? Any concrete examples?

@kimuras: I think the best thing is to convey the fun of it. If you haven’t used it enough yourself to appreciate the fun of it, you won’t be able to convincingly tell someone, “This will impact how you work in the future.”

Generative AI itself is really useful for studying, so it might be interesting to start by simply asking the AI, “What should I do?”

@gomichan: Not only are the applications for AI diversifying, but the models and tools are evolving month by month. I think a lot of people feel tired just trying to keep up. How do you think we should deal with that?

@kimuras: I’m exhausted by the pace of change, myself (laughs). You just can’t keep up. So I’m really happy members voluntarily share info in the Slack channel #ai-random. There’s a limit to what you can track on your own. That’s why I think it’s great that we can all share information with each other at Mercari. Now, instead of the loneliness I felt at first, I feel a real sense of “teamwork.”

@gomichan: Definitely! It might be good for each person to bring in knowledge from their own area of expertise.

@kimuras: If we’re serious about promoting AI, that’s exactly where we need to start. I believe that the source of continuous growth always lies in new challenges. There might be times when people voice concerns like, “Isn’t that going too far?” or “Isn’t that a bit risky?” And while we must always maintain compliance, I want to continue fostering a culture that actively encourages taking on new challenges.