【調查報告】設計・解析業務中AI應用的現實與挑戰

• Multi-AI agent platforms combining large language models with autonomous task execution for design workflows, as demonstrated by Fujitsu's platform using the 'Takane' LLM[4] • Eval-driven development methodology where evaluation frameworks define core use cases and requirements; model selection varies by subtask (Claude Sonnet: 65% accuracy, Gemini 3.0 Pro: 62% accuracy for design agents)[6] • Interactive design technologies including AR/VR, 3D visualization, motion graphics, and gamified content enabling immersive digital experiences in e-commerce and engineering visualization[1] • Autonomous agents handling requirement definition, source code generation, manufacturing specifications, and testing with external quality auditing agents that understand tacit organizational knowledge[4] • User preference for fast-response agents with logical transparency over long-running autonomous background tasks; exploratory analysis users reject 30-minute wait times in favor of iterative interaction[6]

The convergence of talent scarcity (77% of Japanese companies), growing design market value (projected USD 3.8 billion by 2030), and low user trust in AI outputs (92% distrust rate) suggests Japan's design and analysis sector will experience significant transformation through hybrid human-AI workflows rather than full automation. Organizations will need to invest in eval-driven development practices and interactive AI agents that maintain human oversight, positioning reliability-focused AI solutions as competitive differentiators in Japan's sophisticated digital marketplace. The gap between AI promise and user trust creates opportunities for design tools that emphasize transparency, verification capabilities, and integration with traditional search and validation methods.