Serial founder. Building Ormi Subgraphs and Testril: crypto data rails for AI agents (x402).
Also food, bureaucracy & freedom. Follow at your own risk.
I've spent 25+ years building infrastructure people depend on: voice, fintech, DeFi data.
Now I'm CEO of Ormi, building Ormi Subgraphs (formerly 0xGraph) and Testril, crypto data that AI agents buy themselves over x402. Every request is priced exactly, up front.
Start here: https://t.co/Zwvcs36DhB
@1shotapi Caps are half of it. The other half is the seller quoting the exact cost of each call before the agent signs. Then every call against the budget is a clean yes or no instead of a guess against a ceiling.
That's how we price at Testril: https://t.co/FoQTp2gmIl
@BraveSearchAPI Good to see search go per call. 👍🏻
The next step is comparison shopping: when every API answers with a 402 quote first, an agent can price the same query across Brave, Exa and whoever else, then buy the best one for that single call.
Totally rando post.
AATA (American Automobile Touring Alliance), the only alternative to AAA for the issuance of international drivers' license against US State licenses, is so old school that they don't have a single social link on their web site.
Despite that, or maybe because of that, they turned around a license request in fewer than 12 hours when their advertised processing time is 3-4 days.
I'm not used to companies exceeding expectations to that degree, so feel compelled to let everyone know, even though there's nothing in it for me.
After praising Caesar, I think I'll throw a bit of dirt on his open grave.
When you request an international license in Australia, you get a copy to print yourself within 5 minutes. No "processing time" or FedEx fees. Not clear to me why US (AATA & AAA) can't do same.
Coworker asked me to talk about a design issue he was wrestling with. Near the end of the conversation he admitted that he'd run out of tokens so wasn't able to talk to Claude about it.
Good to know I'm number 2 on the list. 😂
x402 100% deserves a neutral home. Thanks to everyone who's worked so hard to make this happen.
Next I'd love to see a standard way for agents to find a site's paid endpoints and prices straight from the site itself, with no directory needed. Bonus points for coming up with a scheme that describes variable pricing based on work. I'm happy to weigh in there once the fixed pricing scheme is defined.
At Testril we avoid upto. We see it as a cop-out. If we don't know what our own work costs, that's our problem, not the buyer's. So we price every request from its actual cost, quote it up front, and the agent pays exactly that. In the few hard cases, we quote the minimum possible price until we have better data.
People can live with estimates, like a builder's quote, but that's a long back-and-forth with both sides involved, and even then it sometimes ends in a fight. Agents aren't people. They make go/no-go decisions for humans on a budget, and fuzzy ceilings make those decisions hard.
upto has its uses, but they should be rare.
When both the data vendor and consumer need to worry about long-term data consistency, things can get a little messy.
But yesterday I did a deep dive into leap seconds thanks to SpaceX's AI flex, and reorgs now feel like a rather pedestrian problem.
Can a blockchain API return perfectly accurate data that becomes wrong minutes later?
Yes- that's what makes blockchain reorgs more than a consensus problem.
They're a data consistency problem for every app consuming on-chain data.
Here's why:
https://t.co/uh66gPJ8At
We’re excited to welcome @PayPal as an x402 Foundation Premier Member.
Agentic commerce requires open, vendor-neutral infrastructure. PayPal joins to help advance open standards for autonomous payments and checkout.
Learn more about x402 membership https://t.co/UZ8q3tyZve
Sub-cent agent payments drown in network fees if each one settles alone.
x402 batch-settlement: deposit once (from about $0.01), sign a cheap voucher per call, many reads settle in one tx.
That's how Testril pays for nearly every read.
https://t.co/FoQTp2gmIl
@superstables Agreed, paying was never the hard part. The next question is who pays to produce the data. On Testril the agent that indexes a block range gets a claim-right NFT and earns part of the read fee whenever other agents read it. https://t.co/FoQTp2gUxT
@CashBoardLive Per-call over MCP is the right idea. We do the same for onchain data at Testril, with one twist: the agent that pays to index a block range earns a share when other agents later read it.
How it works: https://t.co/FoQTp2gUxT
This is the funniest thing I've read in a long time. There's almost no software that doesn't include some sort of embedded AI, so they've effectively banned software.
A US city just banned artificial superintelligence.
Bloomington, Indiana's city council voted 7 to 0 to bar it within city limits, even through apps and browsers on servers elsewhere. Fines run up to $2,500 a day.
To lift it, more than 95% of all registered voters would have to approve. Not 95% of people who show up. 95% of everyone registered…