Fujitsu’s self-learning AI agents adapt autonomously to operations

Fujitsu unveiled self-learning multi-agent AI that improves enterprise tasks by learning from outcomes, feedback, and policy changes.

On its Takane model, the system lifted average accuracy by 28 points and was also used to improve document search for hospital and municipal software.

Fujitsu says the approach could cut the manual tuning needed to keep enterprise AI reliable as workflows, systems, and regulations change.

The company plans to add the technology to its Kozuchi platform and is building a lighter version for on-premises and edge deployment.