Fujitsu's AI agents learn from business operations to improve themselves
Fujitsu has developed self-evolving multi-AI agent technology that continuously improves by analyzing its own performance, human feedback, and changes in business rules. The system identifies why tasks succeed or fail, then applies those lessons to future operations without requiring manual reprogramming by experts.
The technology addresses a persistent problem in operations: business environments change constantly. Legal revisions, system updates, policy shifts, and on-site rule changes happen regularly. Until now, adapting AI systems to these changes required specialized staff to manually adjust prompts, search methods, and evaluation criteria.
Fujitsu tested the technology in two operational areas. In the first, multi-AI agents automatically enhanced and optimized "Takane," a business-specific language model, across manufacturing, healthcare, finance, and public administration. The system achieved a 28-point average accuracy improvement compared to pre-specialization performance by autonomously selecting data, adjusting learning conditions, and testing improvements.
In healthcare applications, the system extracted diagnostic names, progression stages, and treatment policies from unstructured medical records and test results in consistent formats. This reduced reliance on AI specialists to design and maintain models while improving response accuracy.
The second test applied the technology to design specification searches in Fujitsu's electronic health record system and local government business solutions. Traditionally, identifying how software changes affected other systems required experts with deep knowledge of regulations, business processes, and architecture. The AI agents learned from past search results and human corrections, then improved their own exploration strategies-mimicking how skilled experts check related documents and avoid excluding potentially relevant information based on domain knowledge alone.
What this means for operations teams
The system reduces manual work for operations staff managing complex, regulated environments. Rather than waiting for technical experts to update AI logic when rules change, the agents adapt themselves based on real-world execution results. This matters in healthcare, finance, and government sectors where regulatory changes and policy shifts are frequent.
Fujitsu plans to integrate this technology into its AI platform and offer it as a core capability for building business-specific AI. The company is also working with Carnegie Mellon University researchers to enable these self-learning systems to run on-premises and at the edge-not just in cloud environments-using less memory and processing power.
For operations professionals, this represents a shift in how AI tools interact with changing business conditions. Rather than treating AI as a static tool that requires expert intervention to adapt, these systems function more like team members who learn from experience and improve their own performance over time.
Learn more about AI agents and automation or explore the AI learning path for operations managers.
Your membership also unlocks:
