日立、AI駆動型SI基盤「Agentic AI Integration Platform」発表

Hitachi announced the "Agentic AI Integration Platform," an AI-driven systems integration platform designed to automate the entire SI lifecycle, from requirements definition through design, coding, testing, and operation. The platform deploys autonomous AI agents across each phase, marking a transition from human-assisted AI to fully autonomous AI-led development.

In internal trials using AI-driven specification-driven development, Hitachi reduced a design-to-testing phase that traditionally required 13 person-months to just 10.5 hours, achieving roughly 200× productivity improvement. Hitachi has committed to applying the platform to large mission-critical system projects and targets a 30% productivity gain across all processes by fiscal 2027. According to Atsushi Motoyama, Head of AI CoE at Hitachi's AI & Software Services Business Unit, the platform combines Hitachi's accumulated knowledge of mission-critical systems with cutting-edge frontier AI while maintaining safety—enabling continuous system evolution and contributing to sustained corporate growth.

The platform's architecture rests on three pillars: an "Enterprise AI Base" offering flexible selection of frontier AI models from Anthropic, Google Cloud, and OpenAI; proprietary development tools including Hitachi's own "Hitachi GenAI System Development Framework" and GlobalLogic's "VelocityAI" for agile development, along with security, cost, and reliability management; and an "AI-Readable Knowledge Base" that captures enterprise context, business judgment criteria, and Hitachi's systems integration expertise. When paired with VelocityAI, the platform reduces token usage costs by up to 54% compared to standalone AI use. Crucially, the platform learns continuously: AI agents accumulate knowledge from code reviews and bug reports into rules and reference materials, enabling the system to evolve autonomously over time.

Hitachi's Forward Deployed Engineer (FDE) team will support customers from conception through operation, ensuring sustained business growth. The platform accommodates diverse system types: critical systems that must not go down (accounting, fleet management); peripheral and data systems that can tolerate downtime; proof-of-concept and MVP projects; and small-scale control systems where failure is not an option. Hitachi configures asset application to match project characteristics, covering all domains. For control systems (OT) requiring high reliability, Hitachi is building control-system-specific code generation foundations and has begun internal trials with Anthropic's Claude and OpenAI's latest models. The company is also advancing modernization in partnership with OpenAI, targeting concrete results by fiscal 2026, and plans to deploy the platform to "HMAX by Hitachi."

This announcement follows Hitachi's 50-year history of standardizing development processes (beginning in 1970) and introducing automated code generation via Hitachi Application Framework Justware in 2002, which now supports over 200 large mission-critical systems. From May 2024 onward, the company integrated generative AI into its "Hitachi GenAI System Development Framework," yielding 25–30% productivity improvements at key customers. The Agentic AI Integration Platform represents the next evolution: from prompt-driven AI-assisted development to AI-agent-based AI-collaborative work, and now to fully autonomous Agentic AI-driven development. Hitachi notes that AI adoption rates across the company have reached 60% and commits to raising the SI domain to 100% AI adoption within the coming years, extending deployment across operations and OT environments. Shigeru Tachikawa, Head of the Technology Transformation Division, flagged a remaining challenge: the maintenance phase lags in AI adoption and will require process and mindset changes going forward.

Hitachi's announcement marks a deliberate evolution in how the company applies AI to its core SI business. The company has spent over 50 years building standardized development methodologies (since 1970) and introduced rule-based code generation in 2002 through its Hitachi Application Framework Justware tool, which supports over 200 large mission-critical systems. When generative AI emerged, Hitachi integrated it into a new development framework in May 2024, achieving 25–30% productivity gains at customers like MS&AD Systems, Shizuoka Bank, and Shizuoka IT Solutions. The Agentic AI Integration Platform represents the next step: moving from human-directed AI assistance to fully autonomous AI-driven development. Rather than AI simply helping engineers, AI agents now manage design, coding, and testing independently, learning from enterprise context, business rules, and domain knowledge stored in a machine-readable knowledge base. This shift is significant because it addresses acute industry pressures—Hitachi notes that engineer shortages, IT talent scarcity, and the growing complexity and opacity of legacy systems make AI adoption in development "unavoidable." The platform's integration of Anthropic, Google Cloud, and OpenAI models, combined with careful governance around security, cost, and reliability, signals Hitachi's confidence that AI can safely handle mission-critical systems at scale.

No comments yet. Be the first to share your thoughts!