thought this was a solid idea, so why not take you down the rabbit hole of my thinking
1. LLMs are trained on what's widely available, so anything locked behind a course, a paid template or a private community, the model never saw it. that gap is what I want an agent to get into
2. but paywalled doesn't mean good. a lot of what gets sold is public knowledge repackaged, the same stuff everyone has with a nicer cover and a price on it
3. so this gets tricky for an agent. it can buy the course, but it doesn't have the taste to tell if what's behind the paywall is signal or repackaged, and judging that is difficult
4. still, it's a good experiment. take a skill like /last30days, the one that runs a whole sequence on a single command, and add CLI tooling for the places paid information lives
5. use a skill that creates a CLI for any site. point it at gumroad and it builds a CLI where an agent can search and buy anything there, then it does the same across multiple sources, skool, other directories, wherever the paywalled data sits
6. now it buys from all of them, pulls in information that was never in the training data, and turns it into a /llm-wiki, an obsidian brain on the topic
I'm still working out the judgment problem, can an agent tell good from repackaged before it spends across all those sources, or do we need a human gate there
Shann³@shannholmbergnvidia, stripe and nous teamed up on a hackathon for agents that can earn, spend, and run operations here are 4 I'd like people to build for marketing: > a media-buying agent for unconventional placements, the experimental stuff outside the mainstream platforms. newsletter sponsorships, niche site banners, podcast spots. it finds the placement, settles the price, and pays, and it only spends what you allow it to > an ops agent that sets up the whole environment for a campaign. it researches the toolset, decides on the stack, klaviyo for email, posthog for the analytics, then subscribes, pulls every api key and grants the access, so its ready to run with zero admin from you > a product-testing agent that buys and trials competitor products to map their flow. it signs up, pays, walks the onboarding, the emails, the paywall and the upsell, then writes up exactly how they move someone from signup to paid > a knowledge agent that buys its way up the learning curve. pick a topic and it pays for the courses and templates on gumroad, skool and anywhere else, ingests it, and turns it into an llm wiki your other agents can pull from