【Sakana AI Applied Case Interview】銀行業務へのAIエージェント実装に向けた開発の舞台裏

2026-03-19 · Source: Blog · Field: Finance & Economics — Banking & Financial Services, FinTech & Digital Financial Services · Depth: Intermediate, medium

Summary

Sakana AI and Mitsubishi UFJ Bank (MUFG) have advanced their joint "AI Lending Expert" project into the real-world verification phase. This system, developed by Sakana AI's Applied Research Engineers and Project Managers, supports the complex lending workflow by autonomously constructing thought processes for information collection, structuring, analysis, and integrated decision-making. The AI Lending Expert provides consistent support from initial loan consultation to execution, including financial simulations and drafting approval documents, aiming to free bank employees to focus on customer communication and critical issue verification. The development process emphasized structuring "tacit knowledge" into a workflow, avoiding single-prompt solutions, and rapidly improving the AI through nearly 1,500 pieces of field feedback, classified and refined using AI itself. This approach, combined with MUFG's strong top-management commitment, enabled unprecedented speed in quality improvement.

Key takeaway

For CTOs and VPs of Engineering evaluating AI agent implementations in complex operational workflows like banking, this case demonstrates that a structured, iterative approach, combined with strong organizational commitment, can rapidly elevate AI performance. Your teams should prioritize building AI systems that mimic human thought processes and integrate AI-driven feedback loops to accelerate improvement, rather than seeking single-prompt solutions. This strategy enables AI to become a "buddy" that enhances human capabilities and decision-making.

Key insights

AI agents can structure complex human thought processes to support critical financial workflows, enhancing human decision-making.

Principles

• AI should augment, not replace, human capabilities.
• Structure complex tasks into AI workflows, not single prompts.
• Rapid iteration with field feedback drives AI quality.

Method

Develop AI agents to trace human thought processes step-by-step, using AI to classify and refine feedback, and calibrate AI evaluations against human assessments for continuous improvement.

In practice

• Implement AI for information organization and document drafting.
• Use AI to analyze and categorize user feedback.
• Design AI to incorporate "human sentiments" as input.

Topics

• AI Agents
• Banking AI
• Lending Automation
• Tacit Knowledge Structuring
• Human-AI Collaboration

Best for: CTO, VP of Engineering/Data, Executive, AI Engineer, Director of AI/ML

Related on AIssential

• The agentic moment in banking: A blueprint for better customer experiences — Agentic AI transforms banking customer experiences from reactive to proactive, goal-oriented interactions.
• Collaborative AI Systems: Human-AI Teaming Workflows — Effective human-AI collaboration combines AI's pattern recognition with human judgment for superior outcomes.
• AI just BROKE the ENTIRE INDUSTRY... — AI is rapidly becoming the core infrastructure for global finance, transforming both consumer and enterprise operations.
• MUFGとSakana AIの「AI融資エキスパート」、実案件での検証フェーズへ — AI agent systems can significantly enhance complex banking operations like corporate loan approvals.
• Gradient Labs gives every bank customer an AI account manager — AI agents can manage complex banking support workflows by adhering to SOPs with high accuracy and low latency.
• Autoresearch, Agent Loops and the Future of Work — Agentic loops, like Karpathy's auto research, enable autonomous, iterative optimization based on objective metrics.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Blog.