Sakana AI Develops "Proposal Auto-Generation Application" Using Multiple AI Agents in Collaboration with SMBC Group

Sakana AI just gave Japan's banking sector a very concrete answer to the 'what are AI agents actually good for' question. Working with SMBC Group since a partnership signed in May 2025, the Tokyo-based lab has shipped its first real deployment: a proposal-generation tool now live inside Sumitomo Mitsui Banking Corporation's wholesale business.

The pitch isn't just faster PowerPoint. Sakana built a pipeline of specialized agents that each own a slice of the job — gathering financial and non-financial data on a client company, building hypotheses, drafting the narrative, then checking quality and facts before anything reaches a human. The system reportedly figures out its own workflow rather than following a fixed script, chaining these roles together to produce something more consistent than what a rushed banker under deadline pressure might turn out solo.

The numbers are the real headline here. SMBC says proposal work for large corporate clients that used to eat one to two weeks of staff time can now be done in minutes to a few hours. That's not an incremental productivity bump — that's restructuring how relationship bankers spend their week. Sakana frames the value less as automation of grunt work and more as a research assistant that surfaces angles a human analyst might miss, freeing bankers to focus on the client conversation instead of the deck.

This is also a signal of where Sakana AI, the startup founded by former Google Brain researchers David Ha and Llion Jones, wants to compete: not chasing frontier model benchmarks, but embedding agent orchestration into unglamorous, high-value enterprise workflows specific to Japan. SMBC and Sakana say this SMBC rollout is just the opening move, with plans to push similar agent tooling into other parts of the banking group over time.