Cool experiment. I like that you’re running it live, with real constraints, and you shared a concrete failure mode plus fix instead of hand-wavy “autonomous” claims.
What’s the “trust primitive” you think will make skeptical people comfortable letting agents move money without humans, and how do you package that into a real product beyond the demo?