Show HN: Agency Protocol – Domain-Specific Trust Through Verifiable Promises

TLDR: I'm building an infrastructure agnostic trust-building protocol ground in Promise Theory. It formalizes the notion of merit as a domain-specific promise-keeping track record.

Most trust systems collapse nuanced capabilities into oversimplified scores (think Uber ratings or GitHub stars). A surgeon might have excellent bedside manner but declining surgical outcomes, yet both get averaged into a single reputation score. This information loss becomes dangerous as AI systems start making consequential decisions.

Agency Protocol treats trust as domain-specific and economically consequential. Instead of hoping systems behave well, we require explicit, stakeable promises:

- A diagnostic AI promises ">95% accuracy on chest X-rays" and stakes computational credits - A translation service promises "medical document accuracy with <24hr turnaround" - Each promise is independently assessable with verifiable evidence

When promises are broken, stakes get slashed. When they're kept, trust (and economic advantage) compounds.

We're heading toward AI-to-AI coordination at scale—autonomous systems making millions of micro-decisions without human oversight. Traditional reputation systems weren't designed for this speed or specificity. Agency Protocol creates trust as a programmable primitive, like TCP/IP did for data transmission.

The economic mechanism is crucial: it makes reliability profitable and deception expensive, regardless of what the system "wants" to do internally.

See It In Action I've built a working demo at promise-keeping.com, exploring some potential applications in AI, media, etc. The UX translates complex economic incentives into intuitive interactions.

Some open questions:

- How do we prevent gaming when AI systems fully understand the economic incentives? - What happens to privacy when everything becomes assessable? - Can this framework scale to coordinate millions of specialized AI agents?

The theoretical foundations draw from Promise Theory and game theory, but I'm most curious about practical adoption challenges. Would love feedback from anyone building autonomous systems or thinking about trust infrastructure.

Technical whitepaper and implementation details at the demo site.

https://www.promise-keeping.com/