AI Agents Paying Bitcoin

In this memo, we describe what AI Agents are, how they're using similar sub-systems as cryptocurrencies, and how they are work together.

How It Works Summary:

1. Software that does work for you
2. Give them some control and enable approval workflows
3. Give them a budget
4. Pay with Bitcoin's Lighting Network at lower fees

Depending on whether you work in the ad industry, this might sound incredibly familiar or incredibly foreign. Let's start with the familiar.

Growth marketers log into Google or Facebook's advertising center, upload an image and text, set a daily budget, and push the start button to try to get people to buy their product or service.

That's how Google and Meta make all their money. Theoretically, you could call this an AI Agent because that's what they are. You give them some control to make optimizations across ad units and creative (i.e., A/B testing) to drive higher performance. It's used at a global scale inside nearly every major company in the world.

So why can't you have one that works for you outside of this one use case? Well, you can and the tools are being built now. But the difference is you'll need micropayments to enable some of these workflows and likely need streaming payments.

There are two solutions for that, both on the Bitcoin layer 2 Lightning Network. One is Strike by Jack Mallers which has enterprise-grade software in a consumer-friendly app backed by his own Bitcoin treasury, and the other is Lightspark by David Marcus of Paypal and Facebook fame, backed by major Venture Capital firms.

I'll tell you why these tiny, streaming payments are a fundamental requirement using a simple story. I wrote a book and sold it for $1, and Stripe took 35% of that dollar, leaving me with $0.65. That's a huge tax that should not be required. Moving to Bitcoin's Lightning Network means I pay $0.001, leaving me with $0.999. This represnts a 35% improvement in gross margin for my business.

The marginal cost of these AI Agents will be the computing power required for each calculation, each step, and each job they do. Each of those has an energy cost, a battery storage cost, a chip cost, an operating system cost, and an internet cost. Essentially, cloud computing. That is calculated on a per-second basis currently, but we could go lower level and calculate on a unit-of-compute basis. Each unit of compute may cost $0.00001 so we need a way to pay in real-time that doesn't charge a transaction fee of $0.35, which would be 35000x more expensive.

Thus, you could have an AI Agent theoretically run for you 24x7 as a mini business:

• Transaction fees: $0.001 transaction fee per second * 60 seconds per minute * 60 mins per hour * 24 hours per day * 30 days per month = $260 per month (this is an aggressive calculation assuming you're constantly transacting)
• Compute fees: a medium AWS instance is $0.0336 per hour * 24 hours per day * 30 days per month = $25 per month
• Provider Service Fee: assume you have a company that does this for you and charges $15 per month
• Total = $260 + $25 + $15 = $300 per month
• As long as your AI Agent was earning more than $300 per month in revenue, you are earning a profit.

You need internet-native AI and internet-native currency to do this work, and there are a lot of ways to earn more than $300 for tasks on Fiverr and Upwork, especially for something running 24x7x365.

We fully expect this will be one of the shortest paths to quick cash flow for creative internet engineers.

Do you really want customer leads or would you rather Wake Up To Money™:

https://www.youtube.com/watch?v=u1Fjvhh3c_0

We'll leave you with a few diagrams, one from the AI world and one from the Blockchain world.

• The term in the AI world is Retrieval Augmented Generation (RAG). It's a fancy way of saying "go somewhere else to get facts and incorporate them into my AI chatbot response". This reduces the errors and so-called hallucinations.

• It can be incorporated into a series of chained AI Agents performing tasks in series. Each task is an "expert" or resource that does one small function. Expect a cottage industry of tiny, accurate, but powerful AI agents that do one thing: do them really well, fast, and cheaply. You will still need to pay for them. That's the business model for them to be created after all. See LangChain for more on how these come together.

• The term in the blockchain world is an Oracle. Oracles are sources of accurate information outside of a blockchain that can be trusted and used to bring into a smart contract. In layman's terms, go to the source of truth for something, understand what it says, make a decision on it, and then execute that decision.

As you can see, these are nearly the same use case, just applied in different emerging tech domains. Prompt = "Go somewhere else to get facts and incorporate those facts into my AI or Blockchain transaction".

As time goes on, our vocabularies will converge as the technologies do. And you'll find the ones who win are the ones who message using plain language that busy people can understand.

--Sean