Fujitsu Develops Self-Evolving Multi-AI Agent Technology that Learns While Working
Fujitsu has developed technology where multiple AI agents collaborate and autonomously learn from daily operational results. This eliminates the need for human expert adjustments and improved accuracy by an average of 28 points in the domain-specific LLM 'Takane'.
📋 Article Processing Timeline
- 📰 Published: May 26, 2026 at 00:00
- 🔍 Collected: May 25, 2026 at 15:32
- 🤖 AI Analyzed: May 27, 2026 at 14:19 (46h 47m after Collected)
Fujitsu has developed 'self-evolving multi-AI agent technology,' where multiple AI agents work as a team to execute tasks and autonomously learn from various changes, including daily execution results, human feedback, policy revisions, and specification changes.
In business operations, legal revisions and specification updates occur continuously, and decision-making criteria—such as which information to reference—have historically relied on the tacit knowledge of experts. With conventional AI, professional intervention was essential to organize the reasons for failures and reflect improvements. This new technology enables AI agents to organize their own successes and failures, verify their quality and safety, and learn only from effective improvements.
Evaluation using the domain-specific LLM 'Takane' showed an average accuracy improvement of 28 points in fields such as manufacturing, healthcare, finance, and government. In the healthcare sector, for example, it demonstrated high effectiveness in extracting information from medical records. This allows companies to autonomously optimize AI according to business changes without relying heavily on AI specialists. Furthermore, under the OneFujitsu initiative, the company is accelerating its global data-driven management.
In business operations, legal revisions and specification updates occur continuously, and decision-making criteria—such as which information to reference—have historically relied on the tacit knowledge of experts. With conventional AI, professional intervention was essential to organize the reasons for failures and reflect improvements. This new technology enables AI agents to organize their own successes and failures, verify their quality and safety, and learn only from effective improvements.
Evaluation using the domain-specific LLM 'Takane' showed an average accuracy improvement of 28 points in fields such as manufacturing, healthcare, finance, and government. In the healthcare sector, for example, it demonstrated high effectiveness in extracting information from medical records. This allows companies to autonomously optimize AI according to business changes without relying heavily on AI specialists. Furthermore, under the OneFujitsu initiative, the company is accelerating its global data-driven management.
FAQ
今回開発された技術の最大の特長は何ですか？
AIエージェントが業務遂行の結果やフィードバックから成功・失敗理由を整理し、品質と安全性を検証したうえで有効な改善案だけを自律的に学習・反映する点です。
この技術はどのような業務に適用されましたか？
業務特化型LLM「Takane」の自動強化・継続進化、および「HOPE LifeMark-HX」や「MICJET住民記録」といった大規模業務システムの設計仕様書検索に適用されました。
精度評価の結果はどうでしたか？
製造、医療、金融、行政などの領域で精度評価を実施した結果、業務特化前と比較して平均28ポイントの精度向上を確認しました。
この技術の導入による企業のメリットは何ですか？
AI専門人材への依存を抑えつつ、自社業務に最適化されたAIを短期間で構築し、運用を通じて継続的に進化させることが可能になります。
なぜ従来技術では業務への適応が困難だったのですか？
業務環境の変化に合わせてAIエージェントを適応させるには、プロンプトや評価基準、運用ルールの継続的な調整が不可欠であり、これらを専門家が手動で行う必要があったためです。