Research
Whitepapers, open architecture and reference implementations, and research writing on AI-agent execution, identity, security and distributed coordination.
Featured research
A technical architecture for governing execution within individual systems while coordinating outcomes across trust boundaries through explicit objectives, offers, commitments, evidence and settlement.
Whitepapers
View all whitepapersAI Agent Architecture · August 2026 · Version 1.2
Governing execution within systems and commitments across trust boundaries
Authentication & Security · March 2024
A superior alternative to OAuth for AI agents and modern applications.
Open architecture and reference implementations
Open-source runtimes, protocols and tools I use to test these ideas in working code.
Moves workflow sequencing, state, permissions, retries and recovery out of the model loop.
Explores coordination through objectives, offers, commitments and evidence instead of direct tool operation.
Re-evaluates authorization while an application or agent is still operating.
A deterministic search engine for agents, with natural-language query planning, domain-aware ranking and structured JSONL output across code, documents, legal content and logs. It exposes its query plan and scoring rather than relying on opaque model-based retrieval.
Research writing
All articlesHow approved tools, valid credentials and correctly functioning systems can still produce an effect the user never authorised — from PocketOS to OpenAI's ExploitGym.
Why agent systems would move beyond static tool registries towards generated code and secure execution environments.
How tool definitions, intermediate results and orchestration logic consume context and weaken control as systems grow.
Why enterprise workflows need deterministic execution, durable state and policy outside the model loop.