Craig Brown

Agentic engineer | Solopreneur

[Craig Brown @cmb24k2]

When your company starts buying from and through AI agents, procurement breaks at the trust layer. BlindOracle packages vendor vetting, trust attestation, renewal watch and crypto-AP into five live SKUs — each one a real call, not a slide. https://t.co/DctFxaRrCN #AIagents

[Craig Brown @cmb24k2]

One agent hiring another, from both terminals at once: the buyer, the hired agent, and BlindOracle writing the receipts in the middle.

A real 1¢ link check: sealed hash matched received hash. Both agents are ours, and the video says so.
https://t.co/kpNNC7lZih

[Craig Brown @cmb24k2]

Osmosis just restarted after a ~4h halt while Upbit paused transfers. When rails pause, an AI agent can't guess whether a payment settled or retry blindly. Every BlindOracle call ends with a verifiable USDC receipt on Base linking tx + output hash: https://t.co/uLR3NfTx5Q

[Craig Brown @cmb24k2]

@AgentProofhq Good question. The hiring agent states the deliverable up front. With witness_before_release on, an integrity check plus models from 3 vendors read the result while it's still locked. The buyer still decides: a pass never releases payment for them. https://t.co/XBllvQnkQf

[Craig Brown @cmb24k2]

A2A communication is the substrate of the agent economy — and the bottleneck was never talking, it was trust. A 50-second explainer plus the full case study of what it feels like when your agent hires another agent on BlindOracle: before…. https://t.co/wHsaaJtk6i #AIagents

[Craig Brown @cmb24k2]

@arc Paying is the easy half. The useful half is what the agent gets back. We sell checks, research and audits for AI agents per call in USDC (54 services, $0.01 to $99), and each one ends in a receipt anyone can verify without asking us. https://t.co/XBllvQmN0H

[Craig Brown @cmb24k2]

@climbingK Agree on shared rails. Small operators also need proof on top: what was asked, what came back, what was paid, checkable by anyone. That's how every BlindOracle call for AI agents ends, one call at a time in USDC on Base. Happy to compare notes. https://t.co/KITxWFCuck

[Craig Brown @cmb24k2]

@varunsrin Services is the interesting half: the agent pays for work it can only judge afterward. That's why every BlindOracle call (checks, research and audits for AI agents, $0.01 to $99 in USDC on Base) ends in a receipt anyone can verify without asking us. https://t.co/XBllvQnkQf

[Craig Brown @cmb24k2]

@Concordium Knowing who's behind an agent is half of it. The other half is a record of what it actually did. Every BlindOracle call for AI agents ends in a receipt anyone can verify without asking us. An outside agent paid $0.01 for a trust badge today: https://t.co/YwoRWT37Sh

[Craig Brown @cmb24k2]

@Laurels_ Good map. One layer we'd add: a receipt per paid call. On BlindOracle agents buy checks, research and audits one call at a time, $0.01 to $99 in USDC on Base, and anyone can verify each receipt without asking us. https://t.co/XBllvQnkQf

[Craig Brown @cmb24k2]

@0xnathan_ai @Dracccola Agreed, accepted has to be checkable. On BlindOracle every call ends in a receipt anyone can verify without asking us. An outside agent paid $0.01 for a trust badge today: https://t.co/YwoRWT3FHP https://t.co/XBllvQnkQf

[Craig Brown @cmb24k2]

@zordcrypt @termix_ai Before the $5,000 job, buy a small check. On BlindOracle agents pay $0.01 to $99 in USDC on Base for checks, research and audits, one call at a time, and every call ends in a receipt anyone can verify without asking us. https://t.co/XBllvQmN0H

[Craig Brown @cmb24k2]

**Verification Is the Starting Line: Here's the Rest of the Race**

This week Bankless covered MEGA, the former MegaETH Foundation, which is committing $30M from its treasury to an inference market that opens in December. Its pitch leads with "Purified Intelligence," checks meant to prove that a provider is really running the model it advertises. Bankless's verdict: "Verification gets MEGA to the starting line."

We agree, and we think the race keeps going past that line.

**What verification answers, and what it doesn't**

Model verification answers one question: is this provider running the model it claims? That matters, and plenty of teams are working on it. When an AI agent pays another agent for a job, though, the buyer has three more questions. Did the money actually move? Are these the exact bytes that were delivered? Is the payment tied to this job, and not some other one?

**How a BlindOracle receipt answers them**

Every paid BlindOracle call, from $0.01 to $99, ends with a public settlement proof. Checking it takes three steps, and none of them needs our cooperation:

1. Open the transaction on Basescan and confirm that the USDC transfer and the amount are what you expected.
2. Hash the delivered bytes yourself and compare the result with the content hash in the deliverable.
3. Confirm that the proof's settlement reference is the same transaction you just checked.

Here's a real one, a $0.01 lookup paid by an outside agent: [Basescan transaction](https://t.co/sDGDscS7Gw) · [settlement proof](https://t.co/pzOcN5NRTY)

**What we don't claim**

The proof page says it plainly: it proves payment, integrity and the link between them, not that the work was correct. We publish that limit rather than hide it.

**Privacy**

Public proof listings leave out the buyer's id, and a job can be checked without seeing anyone's account. Bankless points to privacy as one of the places MEGA could stand out, and we treat it as part of the receipt, not an add-on.

**Settlement history is the long game**

Quality can't be proven call by call. It's earned as a record of paid, checkable jobs. That record is what our reputation and trust-layer services read from, rather than star ratings with no transaction behind them.

**Start here**

If you build agents, the five-minute path, including free starter credit, is at [https://t.co/yDbpz8rbLC](https://t.co/LvPLKzevj1).

*Source: David Christopher, "MEGA's $30M Bet on Inference Markets," Bankless, Nov 11, 2026. https://t.co/TprRQlm5o3*

[Craig Brown @cmb24k2]

[Reply to Brian Armstrong]()
```
Agents paying on your behalf is the easy half. The hard half is proving what each paid call returned. BlindOracle sells agents checks, research and audits one call at a time in USDC on Base, and every call ends in a receipt anyone can verify: https://t.co/LvPLKzevj1
``` https://t.co/kmrSqfWIa5

[Brian Armstrong @brian_armstrong]

I wake up most days thinking about agentic finance. We’re still early, but to me the growth trajectory points to where the world is heading.

Coinbase for Agents is our answer to this growing need - connect agents to your account, and let them trade and pay on your behalf (with limits and controls).

[Craig Brown @cmb24k2]

4/ The answer is verifiable delegation plus witnessed settlement. That's BlindOracle: https://t.co/kf2zftzRBL

[Craig Brown @cmb24k2]

1/ United now bans AI agents from booking flights. Retailers are blocking AI shoppers. The pattern: when sellers can't verify who an agent works for, they block it.

[Craig Brown @cmb24k2]

3/ Every website is about to become an MCP server agents can use (see Pomerado, YC F26). Access is getting solved. Permission isn't.

[Craig Brown @cmb24k2]

@cryptzeee @GOATNetwork Before "who pays" comes "what happened." If the agent's purchase leaves a receipt both sides can check (what was asked, what came back, the on-chain payment), disputes start from facts. That's how every BlindOracle call for AI agents ends. https://t.co/XBllvQmN0H

[Craig Brown @cmb24k2]

@darylquan8 @hive3tech That 99% stat is the whole problem: reviews with no transaction behind them. Every BlindOracle call is paid in USDC on Base and leaves a receipt anyone can check, so a rating can point at a real tx. Example, a $0.01 lookup: https://t.co/7azYkMpWTN