Join my email list
Craig Brown

I am obsessed with making AI agents prove their work.

I am an AI systems architect who runs a fleet of agents that plan, build, audit and pay each other, and I spend most of my time on the part nobody demos: the gates that catch an agent saying "done" when nothing shipped. Part builder, part skeptic, part ledger-keeper. The thread across my work, from enterprise platforms by day to BlindOracle by night, is that a claim without a record is not worth much. I keep an append-only log of every time an agent's work had to be fixed, and I publish the embarrassing numbers next to the good ones. I optimize my life to be around my family, good food, and warm places. I am now posting every day on X, and more often on Substack and LinkedIn.

The rest of the story @cmb24k2

What I'm up to right now

BlindOracleaudited agents · x402 · Base

BlindOracle

I built a marketplace where every agent carries an ERC-8004 passport, every job returns a proof you can verify yourself, and every call settles in USDC over x402 on Base. 37 live services, a published plugin in the Claude directory, and 104 external payments in the 30 days to October 24. Small numbers, real numbers.

Deep Ledgerpriced in the open

Deep Ledger

The front door to the marketplace: every capability priced in the open, worked by a passport-verified agent, delivered with a cryptographic receipt. If a price or a proof is not public, it is not on the ledger.

Counterpartythe newsletter

Counterparty

I started a Substack in October 2026 about agent trust: identity, authority, and a record the victim controls. Two posts so far. Quarterly cadence at most, and only when something actually happened.

HackathonsETHOnline · Chainlink Convergence

Sealed-bid agent settlement

For ETHOnline 2026 I built sealed-bid agent-to-agent settlement: reserve prices sealed inside a Chainlink CRE confidential workflow, settlement on Base, neutral adjudication. Before that, an agent swarm with autonomous payments for Chainlink Convergence.

What I'm talking about

BlindOracle explainer
Explainer, 47 seconds

When your agent hires another agent

Who pays when the subagent breaks things? A delegation chain that anyone can verify, a passport on both sides, and a receipt that outlives the conversation.

A city of AI agents
The talk I get asked for most

Phantom success, failures included

The talk I give most: an agent reports "done" with zero deliverables, and the gate that catches it. No polished demo. Real ledger entries on screen, including the ones where the fix was mine.

Where I'm posting

X · every day

@cmb24k2

An X article every morning at 9 ET, a short post with a number in it at noon, and the same idea in Spanish and Portuguese for the LatAm agent builders. 24 posts in the last 7 days.

Substack · Counterparty

counterpartybrief.substack.com

The longer, slower version. Started October 2026. Only when something happened that I can show you the record of.

LinkedIn

in/cmb24k

Where I post the version for people who hire, buy, and get audited: the liability angle, the record a stranger can check, what the ledger actually says. Posted by hand, no scheduler.

What I'm writing

What I'm reading

What I'm quoting

"Humans decide what matters. AI does the work. Humans decide how to act."— the house rule
"Built but never invoked is not a capability."— from my fleet's rulebook
"Absence of an answer is never a negative answer."— from my fleet's rulebook

What are you working on? What can we build together?

I am usually mid-build, so it may take me a day or two to reply. If it is time-sensitive, put "READ NOW" in the subject line and resend. I do not mind reminders.

Messages that lead with the agent you are trying to trust, the number you are trying to move, or a great restaurant recommendation get answered first.

Email me