Fujitsu Limited is moving to automate the entire software development lifecycle with a new multi-AI agent platform internally deployed across its government and healthcare systems business.
The initiative, officially launched in spring 2025, reflects what company executives describe as a necessary shift away from highly manual, craft-based system maintenance that has long defined enterprise IT in Japan.
Japanese systems, especially those of large corporations, have been built while responding to various rule changes. The current state of system development, maintenance, and modification largely relies on manual work. It’s an area that requires a certain level of craftsmanship,
said Hideto Okada, Head of AI Strategy & Business Development at Fujitsu.
At the same time, he noted, technological innovation is accelerating — and the traditional model is no longer sustainable.
System development must change now,
Okada said.
Multi-Agent Architecture Across the Lifecycle
Fujitsu’s response is the AI-Driven Software Development Platform, designed to automate everything from legal requirement definition to source code generation, manufacturing, and testing.
The platform combines large language models with what Fujitsu describes as agent-based AI technology capable of coordinating multiple tasks across development workflows. The system can analyze specifications, generate system designs, write and revise production code, and execute testing processes with limited human intervention.
We challenged ourselves to realize a world where pressing a button could completely automate the repair of business applications,
Okada said, describing the project’s founding ambition.
Unlike conventional AI coding assistants that focus on generating snippets or supporting individual developers, Fujitsu’s approach orchestrates multiple AI agents to execute broader engineering tasks collaboratively. One agent can define requirements based on legal revisions, others generate and validate code, while an oversight agent acts as a quality auditor — externally checking outputs, validating actions, and even interpreting what Fujitsu describes as “tacit knowledge” embedded within organizations.
The company positions the system as a step toward fully autonomous, end-to-end AI engineering.
Takane: A Domain-Specialized LLM
At the center of the platform is Takane, a proprietary large language model developed in collaboration with external partners. Fujitsu says the model is optimized to understand complex enterprise systems, including legacy architectures with deeply interconnected components.
We needed a container to store the diverse knowledge Fujitsu has cultivated over 40 years. Takane is optimal for that container,
Okada said.
Unlike general-purpose LLMs, Takane has been evolved as a model specialized in specific domains. Taking local government and healthcare as examples, it is trained extensively on operational knowledge such as resident registration, tax systems, and electronic medical records, while also embedding system development processes and software engineering principles.
While there are other LLMs that boast Japanese proficiency, Takane can correctly understand complex Japanese legal documents,” Okada emphasized. “It can be said to be a model that stores all of Fujitsu’s system development assets.
The model is also designed to operate within secure private environments — a key requirement for public sector and healthcare deployments.
Internal Rollout in Government and Healthcare
Before commercial expansion, Fujitsu prioritized internal implementation. The first deployment focused on business software for local governments and healthcare institutions developed and provided by Fujitsu Japan.
Izuru Kokubu, Head of Measures for Specific Project at Fujitsu Japan, said the timing was critical.
We have been engaged in package-based business for local governments and healthcare for nearly 40 years. We learned about operations from our customers and built systems according to their requests, but now, a significant portion is entrusted to us regarding how to change operations or how to respond to legal revisions,
Kokubu said.
As regulations and rules continue to change, their impact on enterprise systems grows each year.
The longer we spend with customers verifying whether this is truly good, the more beneficial it is for both parties,
he added.
When presented with the opportunity to adopt the AI-driven platform internally, Kokubu said he did not hesitate.
I immediately jumped at the chance and said, ‘Let’s do it!’
According to Fujitsu’s internal proof-of-concept testing, the platform significantly reduced development time in a regulatory update scenario involving medical software. A task that would traditionally take “three person-months” was completed in approximately four hours — a productivity gain of roughly 100 times, the company said.
Material by Iva Abadjievа
Image: Fujitsu
