Fujitsu has developed a self-learning multi-AI agent technology in which multiple agents work together as a team and continuously self-adjust based on performance, human feedback, and policy changes. This technology replaces manual adjustments by experts.
Fujitsu believes this technology is a solution to persistent problems in the business environment. This means regulations are changing, systems are being updated, and business processes are constantly evolving. But traditional AI agents can do little about their own failures. They carried out the instructions but did not independently analyze why something went wrong. As a result, experts had to continually manually adjust prompts, search strategies, and evaluation criteria.
The new multi-AI agent architecture uniquely identifies the reasons behind success and failure during task execution, extracts actionable knowledge, and adapts operational insights accordingly. Therefore, agents take over tasks that previously required manual and ongoing maintenance by experts. Fujitsu positions AI as the new backbone of its digital infrastructure, with internal models designed to support and accelerate all business activities.
After self-optimization, Takane’s accuracy improves by 28 points
Fujitsu applied this technology for the first time to its business-specific LLM, Takane. Multiple AI agents took over the entire optimization cycle, from data selection and training parameter tuning to evaluation and improvement. Each agent developed improvement recommendations based on actual work performance, and only those with proven effectiveness were implemented. As a result, the average accuracy improved by 28 points compared to the pre-specialization version. Takane is optimized for sectors such as manufacturing, healthcare, financial services, and government.
In healthcare, systems now extract structured information from unstructured medical records in a consistent format. This includes diagnosis, disease progression, and treatment strategies. Techzine previously reported on how Fujitsu is leveraging AI to address societal issues, including simulating health policy and testing cyber resiliency on virtual copies of corporate networks.
From document search to autonomous agent improvement
The second application involves document search for design specifications for Fujitsu EHR systems for medium and large hospital and municipal software solutions. Previously, determining the impact of legal changes on software required experts with deep knowledge of regulations, business processes, and system architecture. Using new technology, AI agents learn from previous search results, mistakes, and human corrections. In doing so, they employ search techniques previously reserved for experienced professionals, such as referencing adjacent documents and not excluding seemingly unrelated files from the same corporate domain.
Fujitsu plans to integrate self-learning agent technology into the Kozuchi AI platform and provide it as a core technology for enterprise AI development. The company is also working with researchers at Carnegie Mellon University to develop a lightweight version for on-premises and edge environments with limited memory and power consumption.
