true... but why would my personal-finance app trick me into suboptimal behavior? (I suppose if it's all integrated/same bank etc then yeah, OK).

But quite often my payment app (=Revolut) and I are pretty aligned in our interests. (I had a full paragraph recounting a story w RevolutPay etc, and their fake points, but I took it out from this post).

...or I just cross-run my agents, right; like, give one access/permission to act on the other's. I dunno, I'm dreaming, I guess

I fully expect to see some problems due to underestimating how many biases are unknowingly loaded into these models.