Fujitsu Limited (6702) Earnings Call Transcript & Summary

Thank you all for waiting. We will now begin the press briefing on AI-driven development that will transform system development. The first presenter is Mr. Hideto Okada, Head of AI Strategy and Business Development Unit, Fujitsu Limited.

Hello, everyone. Today, we are pleased to announce the completion of a new technology that will transform the entire system development process. This achievement is the fruit of our efforts over the past year built through repeated trial and era. Fujitsu was founded in 1935 as a telephone switching equipment manufacturer. In the 1950s, it succeeded in developing computers and has been developing system together with customers ever since. And over a long period of time, we have thoroughly learned our customers' operations and refined our system development technologies. System development business is one of the pillars supporting Fujitsu's business today. Meanwhile, with the evolution of generative AI, automation of certain system development tasks such as source code generation has already become a reality. The current challenges are how to enable AI to understand tacit knowledge and how to transform old, complex large-scale systems into AI-driven ones. Furthermore, as experts, when it becomes AI-driven, what kind of roles should we play is where the discussion has shifted now. In other words, what the industry is truly being challenged is for AI to understand complex massive legacy systems that have been in use for years and to transform the entire system development process. Fiscal year 2025, we referred to as the year 1 of AI agents. By leveraging Fujitsu's proprietary AI, Kozuchi and Takane, we announced use cases for multi-AI agents such as supply chain enhancement or about a year ago to all of you, through co-creation with start-ups, new AI agent businesses are emerging is what we have announced to you. Over this past year, the societal atmosphere has also changed significantly. At the beginning, there was AI. What is AI? What can be done with AI? We were searching for answers. But now we have shifted to an era of achieving astonishing results with AI. So given this environment, which unresolved areas should Fujitsu tackle? Where is the unsolved area? What we have to do after all is transforming the entire system development process. Specifically speaking, we will take on the challenge that AI still cannot solve, understanding and automatically modifying complex existing systems. Takane-driven initiative. A special team was formed in April 2025 under the top-down leadership of our CEO, Tokita and targeting Fujitsu's health care and government packaged products and leveraging all Fujitsu's assets and capabilities, including proprietary AI technologies like Kozuchi and Takane, we launched Takane-driven initiative to transform system development into an AI-driven process by utilizing all the assets and all of what we have done up till now. And this time, we have completed technology that automates the entire process from a requirements definition to integration texting for complex existing systems. Today, we announced this achievement to all of you. We declared this year's year 1 for AI agents. However, we now report that this year has become the year AI agents actually began powerfully driving system development. That is for sure. That is what we want to report to you. Now let me introduce Mr. Kokubu, who has led the Takane-driven initiative, who is a responsible person of this business to present the specific details.

Next, Mr. Izuru Kokubu, Head of Measures for Specific Projects Unit at Fujitsu Japan Limited, will introduce the Takane-driven initiative.

Hello, everyone. My name is Kokubu from Fujitsu Japan. Today, I will explain the initiatives we have been developing over the past year in the health care and government domains. Let me now introduce the Takane-driven initiative. Why do we choose to begin with the health care and government domains? First, I would like to emphasize that regulatory reforms are positive initiatives that address social challenges and improve people's lives. We have consistently engaged with these positive changes in earnest and have continuously pursued improvements in operational convenience at the front line. These accumulated efforts have resulted in an extremely large and complex package software portfolio, 67 packages totaling 150 mega steps. In the recent direction, the consumption tax cut was a major topic and continuous regulatory changes require repeated modifications to these large and complex package software systems within a single year. Each time, our customers are also required to change their operational processes. This creates a significant burden for municipalities and hospitals as well as for us system vendors. We regard this as one of the most challenging domains. That is precisely why we must reduce the burden on the front line and ensure that systems continue to evolve along with social progress. We concluded that this is exactly where the power of AI is most needed. So what fundamentally differentiates TDI from conventional AI-assisted development? Today, most AI use in system development takes the form of human AI interaction, often referred to as vibe coding. However, TDI aims to achieve end-to-end automation where development proceeds without human intervention. Once regulatory changes are input, the process runs from requirements definition to integration testing with no human involvement. Each agent automatically triggers the next stage, connecting the development process like a relay. System modification shifts from continuous human AI interaction to fully autonomous nonstop AI execution. This is the fundamental difference from conventional AI utilization. Now let us show you a demo to see what this actually looks like. When a regulatory document is specified and the command is issued, execution begins automatically. What you are seeing now normally takes about 4 hours, but we are showing it at 20x speed. The AI regulatory document identifies the 10 programs requiring modification from approximately 68,000 program assets and applies the necessary changes. As you can see, multiple AI agents advance through requirements definition, design, modification and testing in relay sequence. There are no additional human instructions or approvals. The AI autonomously connects the workflow and completes development. It is an outstanding performance and under optimal conditions, work that would normally take 3 person months was completed in about 4 hours. This is an extraordinary result. This is not merely efficiency improvement. It was the moment when system modification driven by regulatory change shifted from human work to AI work. This is how we interpret this transformation. Let me now introduce the 3 technological breakthroughs that made this overwhelming performance possible. The first is AI-driven regulatory interpretation through requirements definition. As shown on this slide, the 2024 medical fee revision exceeded 700 pages. For health care professionals and system engineers alike, interpreting such documents and translating them into operational and system changes is an enormous burden. However, this AI agent extracts the required changes from regulatory documents and cross-references them with package software design documents and source code, systematically identifying modification points such as screen displays, decision logic, report layouts and data updates and automatically generates requirements at the external specification level. This is a critical breakthrough, translating complex legal text into actionable system change requirements. At this point, you may feel that AI can do anything. However, in reality, there are many things AI appears to know, but actually does not. What AI particularly struggles with are the things not written in specifications or people on the ground take for granted, in other words, tacit knowledge. For example, a registration application system. This is a nationwide system, so the specification required the same functions. However, the actual system design needed in a major city with millions of residents is completely different from that of a town with only tens of thousands. Large cities must handle massive simultaneous submissions requiring batch processing, distributed processing and complex workflows. In contrast, for smaller municipalities, simple online processing is enough. This design considerations are not written in the documents that have already been completed. They are implemented based on the experience of SEs who understand the field. Without this classic knowledge, AI would apply the same design to every municipality. That is why a mechanism to structure system engineers tacit knowledge and teach it to AI in a form you can understand is required. The second breakthrough is the mechanism we call multilayer quality control. When AI encounters ambiguous instructions or tacit knowledge, it inevitably generates plausible answers called hallucination. This is an unavoidable challenge for even the most advanced LLM. Therefore, we created -- recreated real-world development environments and the behaviors of veteran SE professionals within the AI. Autonomous design layer repeatedly observes things and acts autonomously, continuously materializing and detailing the target to be developed. Next is Guardian layer. Like a real-world quality audit expert, it audits the results produced by the autonomous design layer. If it finds deficiencies, ambiguities or contradictions, it instructs to redo. However, it doesn't stop there. It also points out exactly why it's inadequate, helping the autonomous design layer redo it correctly. Next, the knowledge layer systemizes and accumulates knowledge about the business, knowledge about program development and even test knowledge, the unspoken norms of the field. Next or lastly is information access layer without compromising accuracy, extracts the information necessary for AI thinking from vast amounts of documentation and millions of lines of codes. Through these layers, the AI continuously learns on its own, builds on its own and evaluates on its own, then completes the system. Now let's revisit the demo to see how this multilayer quality control operates in practice. After receiving legal amendment requirements, observe health design and implementation proceeds autonomously. Please look at the demo. The requirements definition set document is now being imported. This is a medical pad to actually shifted to prescribed medicine. And based on this requirement, the design has started and the analyzed content of this design is shown here with the cycle of observation, instruction and action and design, this is made. And now the layer is being confirmed, and there is a point that needs to be corrected and that is done. And they are specifying the non-embodied area within this system. So this agent is redoing the design, accumulating the pointed out content and designing the next content. At the third design, all the findings have been corrected and has decided that it has reached a level of acceptance and then the source code part will start. So the logic of the medication has been correctly modified. This is the second breakthrough that the output of the AI is brought up to the expert level. And the third breakthrough is an autonomous relay type architecture that keeps the development process running without interruption. As explained earlier, once you input the content of legal revisions from requirements definition to the completion of integration and testing without human intervention, AI agents will run the development process repeatedly, reading steps until they achieve human SE level quality. It even automates failure causes analysis and modifications ensuring developed never faults. Just as if human SE engineers are working, development progresses 24 hours a day. We have received significant anticipation for this initiative from our frontline customers. It holds the potential to transform the very structure that has exhausted both frontline staff and systems with every policy revision. That is the evaluation we have received. This is the feedback from Shimane Prefectural Central Hospital, a practical reliable approach addressing long-standing challenges faced by our medical institution, highly commendable for being designed with a deep understanding of frontline operations. Strongly feel it holds significant potential to contribute to overall hospital operational efficiency in the future, a promising initiative that should be positively considered for implementation. We believe such feedback is proof that this initiative holds real meaning for the frontline sites. This fiscal year, we have advanced technical verification. And in fiscal year 2026, we will deploy it across all 57 health care and government administration packages. This will dramatically accelerate the pace of service evolution and significantly shorten time to market. Furthermore, through the evolution of packaged software, the accumulated domain knowledge will lead to solving our customers' more advanced management challenges. This is the closing part. Takane-driven initiative is not simply about improving development efficiency. It is a challenge to create a world where systems keep pace with societal changes. We will extend this practice company-wide and to our customer site. From here, we will begin full-scale deployment.

Thank you very much.

How did you enjoy the demo? Takane-driven initiative transforms Fujitsu system modification in health care and government into AI-driven operations. However, these results are not confined to the health care or government domains. We intend to extend these outcomes to all system development. System that continuously evolve are precisely where an AI-driven development platform fits best. For example, so within health care and government, there are regulatory changes that happen on a regular basis, and we need to keep pace with them. We also need frequent updates to continue delivering value to customers. In addition, rapid modification and release directly generates business impact. Above all, modification requires accurate understanding of large and complex system assets. These conditions exist across industries such as health care and government, but also in finance, telecommunication, retail, logistics, manufacturing and beyond. We call this achievement, the AI-driven software development platform. The AI-driven software development platform combines Fujitsu's research technologies and domain expertise to orchestrate multiple AI agents and automate the entire process end-to-end from requirements definition through integration testing, as you can see here. In this initiative, we have built the technology primarily around Fujitsu Kozuchi and Takane, while also testing technologies from Microsoft and Google to further enhance accuracy. That said, however, we will continue to work on the evolution of this AI-driven development platform, of course, centering around Fujitsu's technology, but integrating other available technologies. As Kokubu-san mentioned, we were able to enhance productivity 100-fold. This will change how the engineers work. On the left-hand side, we have typical generative AI tools, which are interactive and they can automate tasks. Of course, these carry highly expected performance, which can increase expert productivity twofold, threefold or tenfold. They support you during your 8 working hours for sure. And that is a traditional setup. However, our AI-driven development platform is fundamentally different. Once a task is assigned, AI autonomously executes the workflow through to completion of integration testing. So after you assign a task, it continues running after you go home while at sleep, 24 hours a day, 365 days a year, it works nonstop. This overwhelming productivity frees engineers from a routine modification work, allowing them to focus on evaluating AI outputs, engaging with customers and create new value. So the experienced engineers will make sure that they conduct evaluation and thereby including humans in the loop. However, we need to go into new territories as well. We need to reengage with customers and deepen our dialogue. We can create new value by doing so. So we want the engineers to be spending more time on these creative tasks. In addition, I want them to be using AI to their own advantage. That is where we would like to head. So in other words, we are talking about AI that executes on its own, not AI that supports. Now let me ask you, what determines a truly delicious bowl of ramen noodle? Many of you would likely answer the broth or some kind of taste that has been handed down to them over generations. However, think about it, preparing the broth can take more than 24 hours in some cases. It takes a long time. Yet finishing a single bowl of ramen takes only about 5 or 10 minutes, which means that the finished bowl is prepared quickly, but its flavor is determined almost entirely by the preparation beforehand. Our AI-driven system development shares the same characteristic. The moment when code is modified and tests are executed, I did mention that we can do it in 4 hours. This is just the final 5 minutes of the ramen preparation. But what determines the quality is preparation, organizing assets and knowledge so that AI can operate correctly, and they need to be updated. So preparation matters, and that is the core of AI-ready engineering. AI-ready engineering is a process of establishing a state in which AI can correctly understand existing systems and execute viable automation. Specifically, I have listed many, but let me mention just 4. Structuring system assets and design rules. So what you already have needs to be learned by the AI. And you need to evaluate whether the AI output is actually correct. So preparing accurate data is key. We need the information about the 2024 medical fee revision. You need that data so that you can cross-reference that to check for accuracy. And you need to automate the testing as well. If we can get this preparation work done in a proper manner, you can have a situation in which system assets can be traced accurately and monitored. As Kokubu-san mentioned, what happened when AI gets the information? Well, it doesn't work as it should all the time, unlike the expert system engineers because they don't have the asset knowledge that these engineers have. And in this fitting phase, you could figure out we are lacking assets or we are not having sufficient development rules. These are pet knowledge that can be documented and learned by AI. And this type of learning process needs to be repeatedly conducted in order for the AI to work autonomously and be fully automated. So the maturity of this preparation determines the accuracy of automation. This repeated process is key. That is what we are going after. Fujitsu holds a strong competitive advantage in AI-ready engineering. We are confident in saying so. This is because what is required for this preparation is not merely AI technology or the expert use of the technology, but practical knowledge of system development itself. the on-site experience and expertise accumulated from supporting complex large-scale systems over many years that we have. And as Kokubu-san mentioned, the execution capability of AI demonstrated through TDI, that is our strength. The ability to integrate human practical knowledge with AI execution capability is Fujitsu's strength and AI-ready engineering is an area that cannot be easily replicated. Thus, to automate large-scale system development, 2 elements are required, AI-ready engineering as a preparation and the AI-driven development platform as the execution engine. Only when these 2 work together, we can truly realize end-to-end automation. Right now, one of Fujitsu's core business pillars, system development is now at a turning point. As generative AI becomes widespread, the value of simple development is rapidly declining, driving waves of price competition and in-house development on the customer side. That's the direction we believe it is going to go towards. More specifically speaking, by using generative AI, all our competitors' productivity will improve and the price competition will accelerate. The democratization of generative AI is expanding customer-driven in-house development. And maybe system development might disappear. However, with this fast speed AI evolution, how can we catch up with that speed? What is necessary for AI is not just technology. Business knowledge is also necessary at the same time. How are you going to maintain that? So when you think it in that way, the AI-driven system development moving forward, which is a new domain is not basing it on simply the number of people or man hours and sell the product. That is not going to exist anymore. The technology will run the system and the system will continuously evolve the system. That is the key point. The speed and the capability of adapting to that. That is going to become the value. AI-ready engineering and AI-driven development platform and the capability to implement that. These 2 will be the competitive edge. And Fujitsu would like to change the system development business itself. This AI-driven system software platform for all system development, it will be started to be available in fiscal year 2026, not just Japan. We will transform the entire system development process worldwide. This is what we are aiming for. Before this announcement, we have conducted activities to receive some expectation voices from our customers. So Kawasaki Heavy Industry, the business knowledge that the company has accumulated for a long time and that technology itself, not passing it on to the next generation, but in order to evolve it in the future, it is an important challenge to take on is what they have said. Furthermore, Sumishin SI Network last fiscal year, within the Uvance update with Sumishin AI Network with the executive, Mr. Aikawa, we got on stage and we received the voices that you really can realize this. And since then, we've been working together up to now. Furthermore, Panasonic from the requirement definition to modification, that done in one stop and that being done autonomously, this approach is going to become an effective answer to the challenges that all the legacy systems have in the Japanese corporations. So the preparation work by setting up the domain knowledge, the customers have expectations beyond that point for us. And this initiative is not -- we have no intention to confine this just inside Fujitsu. Our policy and direction is totally opened. We want to foster this new standard for AI-driven development platforms across the entire industry. And as the first step of this, after this explanation session to all of you, starting next fiscal year, the target once a month, we will launch an open development communities or event so that we can establish this open development communities, so all of us can participate in cultivating this. And furthermore, in 6 months of time, how far have we progressed? This is just a starting point. It is going to evolve more and more. So in about 6 months' time, we would like to share with you how far we have come to. The new system development standard is what we will continuously share actively. Now through these open initiatives, we aim to build the next growth model together, not just within Fujitsu, but with our customers and partners. Customers gain the speed to keep pace with change and the certainty of execution through this. And Fujitsu Engineers will shift into new specialized domains where Fujitsu excels, such as AI-ready engineering and for deployed engineer by utilizing resources that have opened up. So the new service such as for deployed engineering, the new service area is where they should head to. This is not just for Fujitsu, but with our partners, we would like to grow by having the same new roles and same new growth opportunities. As you can see, for us, customers, partners and Fujitsu, we will build a new ecosystem where everyone continues to evolve. And today, we share the technology that will allow us to take the first step towards that. As explained to you now, this technology that we have explained to you today, this is not the goal. This is the starting point. The transformation towards AI-driven system development is for sure a turning point, but apologies to repeat myself. But still, we are at the starting point, meaning that what we are aiming is not only to improve development efficiency. Our aim is the very mechanism itself that enables system to continuously keep pace with ever-changing business operations and society. To achieve this, we will continue to evolve our technology and transform our own business. We will continue to support the sustainability of our customers and society through technology. This vision, there are no changes to it. The AI era has already started. Fujitsu's business transformation today is going to start. Together with the AI-driven software development platform, from here, we will fully deploy the transformation of system development processes worldwide. This concludes my presentation. Thank you very much. [Statements in English on this transcript were spoken by an interpreter present on the live call.]