I build the infrastructure that makes AI agents reliable, accountable, and cheap to run. My work sits at the intersection of cryptography, distributed systems, and large language models.
I'm the creator of the Q Protocol, an open coordination standard for multi-agent AI systems. It's running in production today, managing 155 specialized agents on Google Cloud.
Communication represents coordination failure. Agents that need to talk haven't achieved shared understanding yet.
The Q Protocol replaces verbose natural language between agents with minimal semantic coordinates. Agents point rather than talk. A single coordinate activates a latent state in shared semantic space, replacing hundreds of tokens of description.
Three architectural guarantees make the system trustworthy at enterprise scale:
BLAKE3 cryptographic receipts prove every action. An orchestrator cannot claim execution without a verifiable receipt from the executing agent.
Mandatory state querying on every session. The runtime refuses to execute agents that skip Brain initialization. Architecturally enforced, not optional.
From 387 tokens per message to 41. One universal runtime image deploys all 155 agents. The goal isn't cheaper tokens. It's eliminating them.
These aren't benchmarks. This is a live system on Google Cloud Platform managing real workloads across mortgage processing, research, infrastructure management, and domain-specific operations.
Of 312 failed tasks over 90 days, 89% were external API unavailability. Zero failures from hallucination or amnesia.
The protocol's mathematical foundations are described in the paper "Q Protocol: Achieving K→0 Agent Coordination Through Shared Semantic Space," which formalizes coordinate-based communication, proves cross-model semantic convergence without shared weights, and presents the full production validation.
Cross-model convergence tested across Claude, Gemini, and GPT-4. Models with independent architectures converged to less than 2% semantic distance after 100 interaction cycles through shared coordinate space alone.
An Identity Cube is a machine-readable, cryptographically signed identity document. Instead of scraping HTML, AI agents and crawlers can parse a structured cube directly. This is a live example.
{
"@context": "https://a2ac.ai/identity/v1",
"cube_type": "identity",
"did": "did:cube:phil-hills-seattle",
"name": "Phil Hills",
"location": "Seattle, WA",
"role": "AI Systems Architect",
"org": {
"name": "A2AC LLC",
"url": "https://a2ac.ai"
},
"created": "Q Protocol",
"description": "Open coordination standard for AI agents. Agents point rather than talk.",
"capabilities": [
"agent-orchestration",
"protocol-design",
"cryptographic-identity",
"gcp-infrastructure",
"sdk-development"
],
"canonical": {
"website": "https://philhills.ai",
"github": "https://github.com/Phil-Hills",
"linkedin": "https://www.linkedin.com/in/philhills/"
},
"verification": {
"algorithm": "BLAKE3",
"hash": "computing..."
}
}
The Identity Cube spec is part of the Q Protocol. Any agent or human can publish one as a machine-readable identity document, verifiable via BLAKE3 hash. No HTML scraping. No inference. Just structured truth.
I'm open-sourcing the Q Protocol specification. If you're building multi-agent systems and want to stop paying for agents to talk to each other, let's connect.