I really liked the dash buttons for two reasons:
1. It made ordering something you regularly and only buy online anyway easy for an entire household, like for my use case, hookah coals roughly every month.
2. You could place the button next to where you store said consumable and most folks would prefer using it over asking for a link or reminding someone else to re-order.
I do think it was a mistake to launch with an initial limited set of branded buttons for common household goods most people pickup at the store anyway (like detergent) instead of a generic button for my point 1's usecase but thats just plain-old BigTech thinking.
I can imagine an AI agent grappling with all that but the first time a Dash button price gouged me that way I'd throw the thing straight into the trash.
There is still the issue of who is paying for said item and building an open platform that allows another individual to use my funds at the press of a button would require a lot of investment (identity, trust, limits, etc) though I do see the value you are proposing showing a "mini market" to select from.
I never got one, but they certainly made sense in my mind for something like laundry detergent, etc. You notice you are getting low, hit the button, forget about it and 3 days later it shows up and you go "oh, right perfect".
Envisioning the gigolo opening up their nightstand drawer, shocked they're down to their last couple rubbers... slapping the button and getting back to some other slapping.
Yes, you'd need to sort out pricing. I'm unusually price insensitive, but too many sellers play stupid games and win stupid prizes (me shopping around instead of blindly buying from them every time) by changing unit prices too much too often (e.g. my sons preferred shower gel litre price regularly rotates between prices that adds up to a factor of 3 difference, with any given size container at any given time potentially being a massive rip-off). You could fix that by letting users configure price bands and make it pick from a set of products based on price parameters or give an error. Or only include products where the sellers agree to not play stupid games.
But now your UI is massively complicated and you have error scenarios you didn't have before.
Even then, though, the bigger problem is that there just aren't enough product categories where we frequently order the same thing often enough separately from e.g. the weekly grocery run, to be worth it.
In my household I think it basically boils down to shower gels - because I have a teenage son - toilet rolls, dishwasher tablets and washing machine tabs.
That's not nothing, but it's not going to make me order materially more from Amazon to be able to press a button to reorder vs. pull out my phone, press "orders" in the Amazon app, and find the last order I made and reorder the product.
And I say this as someone who did order a couple of Dash buttons when they came out and promptly forgot to even use them (and I don't even remember why).
It's just such a very minor convenience that to work it'd need to have absolutely zero friction, and even then the revenue it'd bring would likely be totally marginal.
A noteworthy proof of this is the failure of Amazon Dash buttons [1] that blindly ordered your consumables for you. Amazon gave away these devices essentially for free ($5 off with your first order, or some such offer), and people still didn't seem to have been comfortable to use it.
[1]: https://en.wikipedia.org/wiki/Amazon_Dash