USDC as the payment mechanism for AI
Typically, Bitcoin is the reigning payment method for AI's to pay each other or to pay web2 services, perhaps even people themselves.
The appeal makes sense given the decentralized and permissionless nature.
While I think Bitcoin will become the payment mechanism for sovereigns to pay for things amongst themselves, country to country (military services, agriculture, weapons, I would need to look up what actually falls in this bucket), this will be due to Bitcoin's neutrality. It will also be to the inevitable game theory of nation-state treasuries needing to protect themselves from being left behind in currency pricing power wars.
However, for the same reasons Bitcoin's volatility make it near-impossible for peer to peer payments, and how its long-term monotonic rise drive people to save not spend, Bitcoin as a payment mechanism for impending agenic world is unlikely.
What does AI-payments, specifically those by AI-powered agents, enable and look like?
The present case is the access to data, primarily by scraping, but perhaps it's being done in other means, like direct ingestion (I have no idea, but I think the concepts are still the same even if it is a direct sync into a private data warehouse or data lake).
If we constrain AI to scraped data, the 402 error-code could become the start of a rich handshake between AI and content server. 402 is a Payment Required error code that is largely unused.
But it could.
For example, Reddit, Stackoverflow, the US Government, whoever has data on the open web could shift to a payment requirement response. Perhaps this is determined by the User Agent (spoofability is a concern) or IP or something else such that regular people can access without paying (but more on that later).
But the response could include things like: payment terms, prices, and a crypto payment address.
A credit card would be pretty suboptimal for something like this.
One reason is the counter-party risk for the content provider. This may not be that big of a deal given that content is largely user generated, and so acquired for free. But as I'll share in a later essay, when it comes to true agenic supply chains, compute is not free.
But under a credit card, someone could dispute the charges, do a charge back against the content provider, and now they have waged an economic attack.
The second, but related to the settlement finality, is the granularity and speed of this.
Imagine content providers (and definitely compute-intensive services) would want to be paid per-byte, maybe in a streaming fashion.
The agent would not be able to make that many high-volume micro-payments against a credit card.
So counter-party risk and consumption-based pricing ultimately lands on some kind of streaming, near real-time finality of settlement....which is not the nature of credit cards.
However, USDC with to 6-decimal fractionality, pretty-fast settlement, and a legible point-to-point protocol (a sender address and a recipient address) could by this mechanism.
The AI agent would initiate a request to Reddit, and Reddit would respond with pricing, a payment address (a wallet or more likely a smart contract), and code-legible terms of service.
The agent would then start to consume data. The Reddit server could provide responses of consumption, and the agent would start to stream payments.
If the agent stops payment, the server blocks access immediately.
When the agent wants to re-engage, it makes the request, perhaps even including parameters on what type of data it wants, and the server responds. An example of a payment plan could include higher costs for filtered data, for data by data, or even become very granular and provide it on semi market basis if lots of AI's are competing for the data.
USDC would likely easily become a de-facto standard, even if the content providers are in other countries, because this content providers likely don't have such strong nation state provisions (since they are largely acting globally anyway).
Now, imagine the agents spawn additional agents.
Meaning, a primary coordinating agent for data and services (to act upon data or inferences) contracts out to other agents.
The other agents need two things: they need to be paid for their work; they need a budget and authorization to pay for access to the services that they need.
Clearly a credit card wouldn't work easily; the tokens would need to be transferred and the authorization would be difficult to trace.
However, in a USDC model, the actors could receive batched budgets drawn down from an escrow, for example, to do work; the agents now have a budget that they can pay from their own sending address to other services, and the cascading effects of a rug on the original payor aren't possible because of the settlement.
What's also appealing is the fractionalization of payments; as services become more specialized, the consumption model may need work performed at $.000001 per unit of work or something like that.
This is especially important when agents are paying other agents, where the work isn't basically freely produced content, but actual compute.
What would it take to implement this?
To me, it would entirely depend on the willingness of the current gatekeepers, the content providers, to implement this approach.
This would actually relieve them of doing meat-space licensing deals.
It's not just Reddit.
Every book publisher should have a service with a payment gateway and end-point and make it a free market for their data.
Every music publisher could expose their actual music for payment via the service.
The key go to market would be for the Gateway provider in the USDC ecosystem to get deals with the core data service providers.
Now this should force the hands of the AI providers....and not just the big ones.
As open source models and niche providers emerge, a long-tail of consumers for data will likely expand, and the concept of a "data market" finally becomes the unlock.
Data markets have been monetized CSVs or, in the case of large sums of data, Snowflake sinks for money.
While there probably have been machine learning models that did want to use this, it was a small minority because of the downstream complexity of actually doing something useful with this data.
AI has become a more legible machine learning tool, abstracting away the harder bits of making sense of it.
This means data markets can actually be robust two-sided markets because demand can be unlocked.
In fact, governments could develop data policies which keep certain types of data open for individuals, reporters, watchdogs, but hedge funds, content creators, commercial entities would need to pay.
It could work similar to the way governments license spectrum, but now I am way off topic.
So: what do you think?
Will AI start to pay for data in a streaming micro-payments model? And will AI supply chains of agent to agent and agent to service become a thing?