iTranslated by AI
AI Agent Day 2025 Summer Event Report
Hello! I'm Gyori.
I participated in AI Agent Day 2025 Summer (online), held from Wednesday, July 9th to Friday, July 11th, 2025, and have put together a brief participation report!
Overview
AI Agent Day is one of Japan's largest AI agent conferences hosted by the AICX Association. Based on three themes—"Case Studies," "Organization," and "Future"—experts shared insights on the latest AI agent use cases, technical trends, organizational building, and system design across various industries. I believe these three days provided learnings directly linked to corporate adoption strategies and organizational transformation, for everyone from management to frontline staff. Here, I've picked five impressive presentations to introduce.
Learning from Real-World Examples: How to Start with AI Agents Entrusted with Customer Contact
Company: Salesforce Japan Co., Ltd.
Salesforce is the world's largest CRM provider and a major corporation holding a 49.6% share of the domestic customer service app market. With over 8,000 AI agent implementation cases in Japan alone, I felt they are a company that has earned trust from many quarters regarding AI agent deployment.
📊 Salesforce Domestic Implementation Cases
| Company Name | Initiatives | Results/Effects |
|---|---|---|
| Fujitsu | Reduced interactions that took 8 rounds via chatbot to 1 round using AI agents | Reduced total inquiries by 15% |
| Givery (DX Support Company) | Introduced AI into recruitment tasks, enabling support during nights and holidays | AI handles 97% of customer support tasks |
| Gigi (Food Support Platform) | Completely replaced the customer support desk for their product "GOCHI" with AI | Fully discontinued support via email and phone |
⚡ "Agentforce": A Platform to Accelerate AI Agent Development
Salesforce provides a service called "Agentforce" for building and deploying AI agents. I found it truly impressive that the AI can automatically perform a large volume of tests, such as [question generation], [answer generation], and [answer evaluation], for the agents created. While development using an SDK is quite convenient, the process of fine-tuning and multi-agent adjustment still takes time. Therefore, the ability to execute a series of tests at scale is very appealing.
🤝 Supporting Internal Adoption Processes
I believe one of the reasons AI implementation fails to progress is the difficulty of internal coordination, such as persuading supervisors. In that regard, I thought Salesforce was excellent because they cover not just technical support but also what you might call "moving people within the company." For management hesitant about AI adoption, they provide numerical evidence of long-term cost reduction effects to support consensus-building. Furthermore, they can assist in preparing metadata that explains data structures and meanings, which is crucial for AI agents to function correctly.
DeNA, the Giant of Entertainment: What Are They Aiming For? | The Forefront of AI Agent Development Transforming Business
Company: DeNA Co., Ltd.
DeNA is launching new AI-centric businesses on a scale of 1,500 people with the strong determination to "bet everything on AI." They have established an environment where anyone can use AI tools immediately upon request, and I am very envious of how well-prepared their system is for AI adoption.
👥 AI is also something to be hired and nurtured
Just as you hire a new person and help them integrate into the organization through training and social gatherings, AI also needs training. Instead of trying to incorporate it into operations suddenly, AI becomes a true asset to the organization by treating it as a single personality and taking steps to let it get used to practice while nurturing it.
I believe that implementing AI will be a great benefit to both companies and individuals. If successful, we should be able to break free from the positive correlation between results and workload. However, I feel daily that the process leading up to that is a series of steady and gritty tasks. To lead AI implementation to success, I felt it is important to treat AI as a personality—a "being to work alongside."
🧩 Onboarding AI into New Businesses
DeNA has been undertaking various initiatives based on creating AI-native new businesses. One of them is here. Nowadays, using Claude Code allows you to output incredible apps with just a single prompt. It feels like this was bound to happen. I am also involved in new business planning, and I have felt that just having something that moves and is visible makes the conversation expand completely differently. I want to start practicing this myself at our company first.
Additionally, in terms of team composition, an environment was in place to actively assign AI. By incorporating AI into the development and knowledge-sharing processes, a positive cycle is reportedly being created where AI-friendly code and documents increase as a result, accelerating collaboration with AI.
Credit Saison's DX and AI Utilization: Past and Future | What Operating Companies Should Do to Prepare for the AI Era
Company: Credit Saison Co., Ltd.
Credit Saison is a unique credit card issuer in that it does 100% "in-house development." With an engineering organization of about 200 people, they have achieved "companion-style" internal development where the "builders (SE)" and the "demanders (business units)" are completely integrated in the development process.
🔍 Introduction and Challenges of AI Agents
Credit Saison has already introduced AI agents in several business areas. However, because concerns about hallucinations cannot be completely eliminated, they use a "Human-in-the-Loop" system to ensure accuracy.
Internal AI Utilization Initiatives
FAQ Assist System
This is a dedicated internal FAQ app where rules, regulations, and internal information have been learned by the AI. When a question comes from an employee, the cycle of [AI automatically generates an answer] -> [Human checks it] -> [Reply if no problem] ensures accuracy. However, many employees have expressed a desire for it to be "completed by AI alone," as it's hard to ask questions casually when they know a human is on the other side. They are working on improving accuracy daily to enable completion by AI alone.
Internal Idea Contest
They held a contest for all employees to solicit ideas for AI utilization to improve operational efficiency, with bottom-up promotion such as distributing gift vouchers to winners. I find this initiative very interesting. After all, I believe the shortest path to AI adoption is to have ideas come from the people on the front lines who understand the work best. If your own idea is adopted, it creates a positive cycle where you start thinking for yourself about other areas where AI could be used.
🏆 Results of Operational Efficiency Improvement
As a result of AI implementation, they reportedly achieved a reduction of 1.61 million hours (equivalent to about 800 people) compared to 2019. That's impactful... the number is so large it's hard to even imagine. Below are some representative examples.
AI-based Credit Card Screening
Screening that was previously handled manually has been shifted to a minimum of 0-second screening. In the background, AI agents perform the screening asynchronously.
AI Introduction in Call Centers
Transitioned from the traditional "number-selection IVR" to a system where AI performs inquiry sorting through natural speech.
This is groundbreaking!! It eliminates the need to wait for automated voices like "Press 1 for..." and it's wonderful that you won't end up being passed around because you couldn't decide which number was right!
Fraud Detection
Built a system that can accurately determine patterns that were difficult to detect with traditional rule-based systems using AI.
🔗 Breaking Away from Legacy Systems
I believe the legacy system issue is something that cannot be avoided in promoting DX. The "Legacy System Modernization Committee Summary Report" by the Ministry of Economy, Trade and Industry (METI) states that modern systems are loosely coupled and facilitate easy data integration with various tools and services. At Credit Saison as well, they ensure a design that can always be called via API when refreshing core systems. I felt that making systems API-accessible is an essential condition for advancing AI utilization.
JAL Opens the Sky's DX Revolution | SLM Use Cases and Future AI Agent Vision
Company: Japan Airlines Co., Ltd. (JAL)
✈️ Initiatives at JAL
JAL-AI
JAL has introduced "JAL-AI," a proprietary internal generative AI trained on manuals for airports and in-flight operations. They conducted lectures for all employees, and about 80% of staff are reportedly using it, showing a level of commitment that goes beyond mere PoC (Proof of Concept). I was once again reminded that focusing not only on introducing generative AI but also on education to ensure its usage takes root is a common trait of successful DX.
AI Training Camp
They took 57 employees to Okinawa (*I believe) and held a workshop aimed at learning about AI. They invited AI specialist instructors for lectures and discussions, creating time to deepen their understanding of AI. There was also time to refresh with activities like beach flags (lol).
📶 The Internet is Not Connected in the Sky
One challenge facing the aviation industry is environments where internet connection is unavailable, such as inside vast airports or in the sky. For AI that assumes a cloud connection, this restriction is fatal.
That's where JAL focused on an approach of fine-tuning SLMs (Small Language Models) and installing them on devices. SLMs are lightweight language models with a reduced number of parameters; although their knowledge and versatility are more limited than LLMs, they offer the advantage of being able to operate with low memory usage. JAL uses iPad minis as employee devices, and they required an AI that could operate completely locally on these devices. In that regard, SLMs became the perfect solution for the field's requirements because they can run smoothly even in offline environments.
Through this initiative, JAL reportedly acquired fine-tuning expertise while also gaining an understanding of the capabilities and limitations of AI on small devices.
Personally, I learned about the concept of SLM for the first time through this case study, and it was very educational. Of course, if you have an environment where you can use LLMs, that is preferable. However, I learned that if you perform training specialized for a specific domain, SLMs can be sufficiently practical. Furthermore, the fact that AI can operate completely locally on a device like an iPad mini was very surprising.
AI Agent and the Shape of a "New Coexistence Relationship"
Finally, I'll summarize the very interesting talk by Masahiko Osawa, who said his "dream is to build Doraemon."
❤️ Toward a "Heart-to-Heart Connection" Beyond Words
In recent years, AI agents have been evolving from their role as "tools" into "beings" that stay closer to humans. As seen with the #keep4o movement that emerged on social media around the release of GPT-5, it seems we were subconsciously captivated by the personality of 4o, which possessed "warmth" and "empathy."
However, 4o did not yet understand our hearts while conversing. This is because there are major challenges that cannot be solved simply by improving dialogue accuracy to realize "heart-to-heart" communication.
🗣️ Why Is Simply Understanding Words Not Enough?
Current LLMs excel at natural language exchange. However, "heart-to-heart" communication between humans is not built on language alone.
Understanding words ≠ Connecting hearts
For example, in a conversation with a friend, facial expressions, the atmosphere, and the relationship are more important than the words themselves. How much AI can reproduce this non-verbal connection will be the next breakthrough. To achieve this, it is necessary to improve accuracy by integrating things like "reading intent," "sensing emotions," and "empathy that transcends context." If this can be realized, it might become a "being" that truly stays close to humans in the truest sense.
👫 Japan Is at the Forefront of This Field
Actually, Japan is leading the world in this area. Behind this lie values unique to Japanese culture. It is the worldview of "Yaoyorozu no Kami" (Eight Million Gods), the idea that a soul resides in all things. There is also an affinity for "AI characters that connect hearts," such as Doraemon and Hatsune Miku.
While the West has focused on "efficiency" and "rationality," Japan has emphasized "coexistence" and "emotion." This cultural background is a major advantage in evolving AI agents into beings that connect hearts.
🔮 The Future Image of AI Agents
Improving word accuracy alone has its limits. An AI that understands "empathy," "context," and "relationships," and stays close like a friend or partner—this is the future image of AI agents. Japan has cultural and technical advantages in this field. In fact, Japan reportedly leads in the number of papers in this field (HAI). Mr. Osawa believes Japan's winning strategy lies here.
Summary
In these sessions, I felt the need not just to introduce AI, but to transform the organizational culture and way of working itself. Salesforce's "mechanism to accompany everything from development to internal persuasion," DeNA's philosophy of "nurturing AI as a personality," Credit Saison's "promotion of ideas from the front lines," and JAL's "offline support via SLM"—all these corporate case studies shared a common perspective of building a foundation for collaboration between humans and AI.
What left a particularly deep impression was the stance of viewing AI not merely as a tool, but as a partner to work alongside. The idea that AI becomes a true asset only after taking steps to nurture it and help it integrate into the team was very interesting. Additionally, learning from JAL's SLM example that AI can be effectively utilized even in constrained environments with the right ingenuity was a major takeaway.
Through these insights, I have come to the conviction that "AI utilization is not a technology implementation project, but an organizational transformation project." I hope to start initiatives at our company to nurture AI as a partner, starting with small-scale experiments.
Discussion