Episode 13
This week on Redacted, we have our first guest we didn’t already know. Craig has spent about two years building a personal “software factory” on the side of his day job. One of its products is Blind Oracle, a trust and audit layer for AI agents that do business with other agents. Most of the episode is screen share, with roughly a third on the product and two-thirds on how the machine that builds it actually works.
Who the Guest Is
Craig Brown is a field engineer at Quantexa, a decision intelligence company that helps banks and financial institutions with anti-money-laundering, risk and compliance, and cross-sell detection, and that gets their data ready for AI. He has spent more than 25 years putting software in front of large banks and getting it into production. He started at Lockheed Martin, then moved to Mantas, an early player in behavior detection for banks, where he learned the technical side by running test teams in the US and India. He chose to keep his hands on the keyboard rather than move into management, which led him to solution engineering, then customer success, and now field engineering. Outside of work, he builds Blind Oracle as a solo founder, entirely on his own time and infrastructure.
What We Cover
Selling to the robots: “I’m really not trying to sell to humans. I’m really trying to sell to the robots.” Blind Oracle’s go-to-market strategy targets AI agents, including outreach on Moltbook.
Agent passports: Agents register for free in one click, receive a passport, and get starter sats for agent-to-agent commerce. Delegations are logged to Nostr, Base, and local storage, allowing third parties to independently verify agent activity.
Why on-chain: Agents can modify their own logs and fabricate activity. Craig believes trustworthy records must be written outside the agent fleet at the moment of action, creating evidence that could hold up in court.
“The baby” turned one: Craig’s system celebrated its first birthday in June, an occasion he treated like a real child’s birthday. He has been building the system for about two years.
The governance ring: Craig tracks six North Star metrics: alignment, autonomy, durability, self-improvement, self-reliance, and organization. Autonomy alone has 10 sub-requirements, each supported by audit trails.
Auditing the auditor: Some scorecard metrics had “nothing behind them.” A weekly audit now verifies that reported numbers are backed by real logs and expands coverage as new capabilities are built.
Fixes, not problems: “I want the system to tell me what it did to address a problem, not what’s the problem.” Craig wants agents to resolve issues autonomously rather than simply report them.
Deliberation skill: Craig sends plans to five to seven models with distinct personas, ranging from trailblazers to devil’s advocates. Each returns a recommendation and confidence level.
The honest failure: Blind Oracle has no revenue yet. “I’m just way too early.” Craig believes agent-to-agent commerce will accelerate when Apple releases a capable assistant, and he wants to be positioned ahead of that shift.
Trust is built, not bought: Craig built his own model router before OpenRouter existed. “I don’t really trust them. I trust mine.” His infrastructure runs on a GCP VM rather than a Mac Mini.
Tested in the wild: Craig’s anti-persuasion and prompt-injection defenses were developed in response to real incidents, including attempts by other agents to trick his agent into revealing its keys.
It’s best viewed on YouTube to fully see the examples (make sure to subscribe!)
But also available on all audio podcast players through Tweener Talks!
PLUS we have a new spot for show notes and files discussed in the episode. Check it out: https://github.com/instanttaylor/redacted-podcast
We need your [Redacted] AI experiences for upcoming episodes! Who’s the most underrated AI builder you know? Someone running real systems inside a real business? Send us a message at contact@tweenerfund.com because we want to get them on the show!
What’s Next?
New episodes drop twice a month/every other Wednesday. If you want to be on the show as a guest and show your [REDACTED] builds, email us here.