From 'Building' to 'Mass-Producing and Operating' AI Agents: Mizuho Financial Group Launches 'Agent Factory'

Mizuho Financial Group (President & CEO: Masahiro Kihara, hereinafter 'Mizuho') has built 'Agent Factory,' a system for mass-producing and operating AI agents that allows for rapid development and continuous improvement after deployment. Full-scale deployment began in February 2026, starting with the Digital Strategy Department.

Agent Factory standardizes the process from development to deployment, operation, and improvement, reducing the development time of complex AI agents from the previous two weeks to as little as a few days—a reduction of up to 70%.

With a view to utilizing thousands of AI agents in the future, this shift to a mechanism for mass-producing and improving agents in short cycles will accelerate operational efficiency and improve customer service quality across the group.

Background: Accelerating AI Adoption and the 'Wall of Scale'
In December 2025, Mizuho released its next-generation AI infrastructure, 'Wiz Base,' to create an environment for safe and efficient business use of generative AI. Beyond simple adoption, Mizuho is now moving into a phase of automating and enhancing entire business processes through 'AI Agents' that autonomously execute specific tasks.

However, as the company continued to launch AI agents with different business requirements, the reliance on individual development methods made it difficult to ensure speed and reproducibility. Therefore, it became necessary to standardize and unify the following aspects:
- Development Reproducibility: Reducing development time and quality variance through the reuse of common elements.
- Deployment Reliability: Reducing the burden and errors in configuration through shared knowledge.
- Continuity of Operation, Evaluation, and Improvement: Standardizing quality management, operational status tracking, and log management.

Given this situation, 'Agent Factory' was built as a mechanism to safely and rapidly scale AI agents across the company and to improve and expand them in short cycles.

Overview and Objectives of 'Agent Factory'
Agent Factory is a system that integrates the infrastructure (things) and the organization (people) to support the development and mass production of AI agents. It is designed to maintain consistency in design, quality, security, and operation even as the number of developers increases, and to ensure a continuous improvement cycle after production deployment.

The infrastructure is positioned as a platform that provides end-to-end support for AI agents used within Mizuho, from development to evaluation and improvement. It is supported by three main elements:
1. Common Template 'Agent Template': Developed to allow standardized development and deployment. It supports development with an eye toward post-deployment evaluation and improvement.
2. Proprietary Design Principle 'AI Oriented Architecture (AIOA)': Formulated to prioritize governance and security, ensuring both speed and mass-producibility while maintaining the high reliability required of a financial institution.
3. Optimal Development Platforms: Utilizing 'Amazon Bedrock AgentCore' for advanced processing and complex integration, and 'Dify' for rapid development needs, allowing flexibility for everything from simple workflow automation to advanced multi-agent systems.

Future Outlook
Mizuho will continue to update and evolve Agent Factory. By gradually introducing memory functions for personalized agents, enhancing mechanisms for multi-agent collaboration, and expanding operational aspects, the company aims to broaden the scope of utilization and further strengthen its ability to mass-produce advanced AI agents in a short period. This will lead to more efficient and sophisticated internal processes and enable the provision of high-value financial services to customers more quickly.