Blog
Engineering notes on governed retrieval, AI agent governance, and on-premises LLM deployment.
-
The Six Substrate Dimensions
What belongs in the substrate and what belongs in the framework? A falsifiable test, six dimensions that survive entity replacement, and why governed agent systems need all six from day one.
-
The Protocol Between Agents Is the Protocol Between Teams
The interface layer gets replatformed. The governance layer underneath survives. Thirteen years of watching enterprise software eat the cost of that pattern, and what it means for multi-agent AI in 2026.
-
Causal Decomposition Under Suspended Judgment: Two Cases from Keystone
Two cases in diagnosing governed AI failures at the mechanism level: a three-factor retrieval bug (FC-005) and a day-one multi-agent substrate decision. The discipline is suspending judgment long enough to find every cause before acting on any of them.
-
Provana AcuteCare: Clinical Copilot with Guardrails
A clinical AI copilot with role-based action guardrails. Three protocols (sepsis, stroke, pediatric fever), dynamic UI generation, four governance outcomes (Serve, Block, Route, Mutate). Built in six hours at the Boston AI Tinkerers hackathon. The governance architecture moved from workplace safety to acute care medicine with zero structural changes.
-
Governed Incident Response: Agentic UI with RBAC and Audit Trails
An agentic interface for incident response where every tool call is authorized by the user's role and logged to a tamper-evident audit trail. The agent generates different UI depending on what the user is allowed to do. CopilotKit + Next.js generative UI. Built solo for the AI Tinkerers Generative UI Hackathon, Boston.