@SpaceandTime AI has started buying data on its own, but it can’t even tell when it gets the math wrong
Over the past couple of days, I hooked a data-query agent up to a demo environment and ran dozens of questions through it. The most unsettling thing wasn’t that it got answers wrong; it was that it sounded exactly the same when it was wrong as when it was right. The Enterprise version that Space and Time launched on November 9 hits that pain point head-on. The idea is straightforward: run computations on private data, produce a cryptographic Proof of SQL receipt, and have the application verify the receipt before feeding the result to the agent. Once it checks out, let the agent get to work. The data stays in the database, and the computation can be verified. That promise is going to get very valuable now that agents are starting to make API calls on their own.
Here’s the workflow I tried: connect a business table, send a query through the SDK, get the result and proof, then verify it in an on-chain contract. It took me about an afternoon to get it working the first time. The sticking points weren’t cryptography, but environment setup and figuring out which table was worth proving in the first place. That learning curve is real, but what you get in return is something solid: whether a result is correct no longer depends on who you trust.
Compared with a price-feed solution like Chainlink, which tells you what a number is right now, SXT tells you how that number was calculated and which table it came from. Compared with a read-only dashboard like Dune, the difference is that the latter can only ever give you a screenshot, while the former can give you a receipt. Letting AI make decisions based on screenshots is just asking for trouble.
There’s another side effect I didn’t think of at first. With proofs, every decision an agent makes can be replayed. If something goes wrong, you can trace it back to the specific query and the exact data it used. That makes accountability much cleaner than digging through logs, though it also costs more. So I’m not too concerned with how compelling its story sounds right now. What I care about is this: when agents really start spending money on API calls to buy data, who gets to collect the toll at the receipt layer? #sxt $SXT
Over the past couple of days, I hooked a data-query agent up to a demo environment and ran dozens of questions through it. The most unsettling thing wasn’t that it got answers wrong; it was that it sounded exactly the same when it was wrong as when it was right. The Enterprise version that Space and Time launched on November 9 hits that pain point head-on. The idea is straightforward: run computations on private data, produce a cryptographic Proof of SQL receipt, and have the application verify the receipt before feeding the result to the agent. Once it checks out, let the agent get to work. The data stays in the database, and the computation can be verified. That promise is going to get very valuable now that agents are starting to make API calls on their own.
Here’s the workflow I tried: connect a business table, send a query through the SDK, get the result and proof, then verify it in an on-chain contract. It took me about an afternoon to get it working the first time. The sticking points weren’t cryptography, but environment setup and figuring out which table was worth proving in the first place. That learning curve is real, but what you get in return is something solid: whether a result is correct no longer depends on who you trust.
Compared with a price-feed solution like Chainlink, which tells you what a number is right now, SXT tells you how that number was calculated and which table it came from. Compared with a read-only dashboard like Dune, the difference is that the latter can only ever give you a screenshot, while the former can give you a receipt. Letting AI make decisions based on screenshots is just asking for trouble.
There’s another side effect I didn’t think of at first. With proofs, every decision an agent makes can be replayed. If something goes wrong, you can trace it back to the specific query and the exact data it used. That makes accountability much cleaner than digging through logs, though it also costs more. So I’m not too concerned with how compelling its story sounds right now. What I care about is this: when agents really start spending money on API calls to buy data, who gets to collect the toll at the receipt layer? #sxt $SXT
