Date: September 11th, 2026 5:00 PM Author: lavender rambunctious church
any ideas for services agents would pay to use themselves, that they can't reproduce themselves in their own workflow as cheaply, accurately or quickly?
Date: September 11th, 2026 5:36 PM Author: glassy vengeful halford
IRL world actions that lead to something physical
sending letters, book buying, art generation etc anything that you can hook up to an api or ideally if demand - create your own api/mcp and build a micro home factory to deploy the service.
Date: September 11th, 2026 5:41 PM Author: lavender rambunctious church
have thought of things like this, but they are all really slow physical processes. I feel like physical world-stuff only works if the physical object is the endpoint or the workflow can persist and resume when the physical result arrives. rentahuman is interesting, but it seems to still mostly just be humans ordering stuff or maybe they tell an agent to get a human to deliver a pizza to the principal when its finished with its long process. but i guess it could still be worht it if it just becomes a thing, where someone asks a consumer agent product like gpt on the webapp "find me this" and it cdan go find it through an endpoint like that. so the product getting sent to the human is the endpoint and the point is they can do it just by asking a chat instead of going and finding and ordering it.
I think internet-facing services for agents are more promising. Seems like now most of what they are buying is compute and data and maybe access to articles from expensive databases that would normally require 30K per year subscription and they just want one article.
Date: September 11th, 2026 5:57 PM Author: glassy vengeful halford
the physical delivery thing is more around ambient discovery - ideally SaaS is able to extract users preferences and deliver something unique on a continual basis.
"articles from expensive databases that would normally require 30K per year subscription and they just want one article."
y this is a big opportunity - tons of old school businesses are slow to pivot and are charging $50k+ for minimal service with tons of extras no one wants/uses. Good call.
Date: September 11th, 2026 6:23 PM Author: lavender rambunctious church
Inference/model calls and compute seem to be the biggest so far. AWS already exposes a control surface for this. There are a few others forget off the top of my head.