KEY POINTSFujitsu starts early verification of Fujitsu Kozuchi Multi AI Agent Framework on July 15Framework auto-configures business-specific multi-agent AI systems and improves them using execution results and human feedbackInitial focus includes retail ordering, system development and modernization, and sales proposal and booking operations
Fujitsu begins trials for its new multi-AI agent framework.
Fujitsu
Fujitsu said on July 13 it had developed a platform for building and operating business-specific multi-AI agent systems and would begin early verification on July 15.
The platform, called Fujitsu Kozuchi Multi AI Agent Framework, or MAAF, incorporates what Fujitsu describes as self-evolving multi-AI agent technology as a core component. The framework automatically configures groups of AI agents from business knowledge and continuously improves multi-agent systems based on execution results and human feedback.
Fujitsu positions the framework as a way to address a common problem in enterprise AI-agent deployments: business procedures often combine exceptions, judgment criteria and system operations in ways that are hard to capture in a one-time build. The company said changes in regulations, specifications and customer requirements have also made it difficult for some projects to move beyond proof-of-concept stages.
MAAF can ingest business manuals and design documents as well as recordings of commercial discussions and meetings as operational knowledge, and propose multiple automation plans by identifying what should be automated. It also includes an interactive session function that asks focused follow-up questions on key design issues, reducing the need to first draft formal requirement-definition documents.
Fujitsu said the framework is designed to produce business-specific multi-agent systems that have been verified for correct tool invocation. In operation, multiple AI agents can divide roles and work together to support tasks such as ordering, impact analysis, proposal preparation and inquiry handling.
The self-evolving technology treats construction, operation and improvement of a multi-agent system as a single lifecycle. Based on execution histories and human feedback, it generates candidate improvements for prompts, skills, workflows, tools and role assignments. Candidate changes are tested in an execution environment, with only validated modifications reflected in the system. For important changes, human approval or confirmation is incorporated and change histories are retained in an auditable form.
Fujitsu plans to link MAAF with its Fujitsu Kozuchi AI platform and Takane, its enterprise generative AI offering, to accelerate development and operation of business-focused AI agents. Initial applications include retail ordering operations, investigation, impact analysis and testing in system development and modernization, and proposal preparation and sales booking in sales operations.
Multi-agent systems, or MAS, use multiple AI agents that divide roles and coordinate with one another to achieve a task. Fujitsu included a comment in the release from Graham Neubig, an associate professor at Carnegie Mellon University, who described the framework as a strong approach to routing and optimization for AI agents.