Agentic AI Architect • Engineering Leader • Scale by Subtraction Pioneer
I design production-ready AI agent systems that are reliable, governed, and truly autonomous, not just sophisticated chatbots. My work focuses on Control Planes, Deterministic Graphs, and the radical idea that systems scale better by removing complexity.
The field guide I wish I had while building large-scale cloud and AI systems.
This book is about the decisions that become expensive at scale: where state lives, how systems fail, when to remove complexity, and how to put real control around probabilistic software. It comes from years of building systems where architecture had to survive contact with production.
Current work: open standards for trusted agents
agentrust-io · hardware-attested governance for autonomous AI agents
The next generation of AI scaling won't come from adding more tokens. It comes from removing complexity.
We don't ask microservices "nicely" to respect rate limits; we enforce it. Why treat AI agents differently? Build the Kernel for the AI OS.
Don't detect hallucinations after generation. Prevent them structurally with multi-dimensional Knowledge Graphs that define what's possible.
Language is for humans. Code is for machines. The best agents are the ones that can't talk. They communicate through structured data.
Open source projects that make autonomous AI systems reliable, governed, and actually useful.
The unified governance platform for AI agents, hosted under Microsoft. 5 top-level distributions, 5 language SDKs, ~10 framework integrations. 7 of 10 OWASP Agentic Top 10 fully covered and 3 partial, NIST AI RMF alignment, OpenSSF best practices.
Trust Runtime Attestation and Compliance Evidence. An open standard for hardware-attested trust records a verifier can check without trusting the operator.
Confidential MCP gateway. Enforces Model Context Protocol tool-call policy inside a hardware TEE, with the Cedar policy bundle measured into the attestation report and a signed proof per session.
A cryptographic identity and provenance standard for AI agents. Hardware-anchors the artifacts that define an agent at deployment, so what is running can be proven, not just claimed.
Confidential agent-to-agent. Attested, attenuated delegation and sealed peer channels as a profile on the A2A protocol.
My methodology for building reliable AI systems. Focus on removing complexity rather than adding features. Via Negativa applied to software architecture.
Deep dives into agentic systems, architectural patterns, and the philosophy of building AI that works. Published on Medium and Dev.to.
Contributing to the frameworks that will define how autonomous AI systems are built and governed.
Contributing to agent interoperability standards through the OASIS Agent Interoperability Technical Committee.
Agent Interop TCAGT aligns with NIST AI Risk Management Framework profiles for responsible AI deployment in enterprise settings.
AI Risk ManagementParticipating in the Coalition for Secure AI and the AI Agent Infrastructure Foundation to advance agent safety.
AI SafetyAGT follows OpenSSF best practices for supply chain security, dependency management, and vulnerability disclosure.
Supply Chain SecurityChallenging conventional AI wisdom with practical, battle-tested approaches.
"Sustainable agent architecture does not require agents that remember more; it requires agents that know how to forget."
"Stop trying to prompt-engineer safety. Shift the burden of constraints from the probabilistic LLM to the deterministic Knowledge Graph."
"We don't ask microservices 'nicely' to respect rate limits; we enforce it. Why are we treating AI agents differently?"
Deterministic policy enforcement, tamper-evident audit trails, and trust-gated agent handoffs, all hardware-attested.
Find me on these platforms for articles, code, and collaboration.
