Research
Notes and papers on distributed cognition, autonomous systems, and software architecture.
I’m exploring how autonomous systems, human operators, and software infrastructure can be composed into coherent cognitive systems.
This work blends philosophy, cognitive science, and practical engineering, and is grounded in systems I run daily to manage projects, context, and execution across multiple domains.
The goal is not artificial intelligence in the abstract, but practical intelligence - systems that reason, coordinate, and act reliably in real environments.
Featured Work
Ontological Foundations of Distributed Cognitive Systems
A framework for organizing multi-agent systems through ontological clarity - defining what agents are, how they reason, and what boundaries govern their capabilities.
The paper connects philosophy, cognitive science, organizational theory, and software architecture to describe how distributed intelligence can be designed and scaled in practice.
Systems in Practice
These ideas are not purely theoretical. I use a set of experimental systems daily to test and refine these concepts in real workflows.
Ralph
An orchestration and metacognitive system that prepares daily briefings, tracks commitments, and coordinates work across tools and domains.
Lux
A codebase intelligence tool that indexes source repositories for semantic search, enabling agents to resolve their own questions against real code rather than asking humans.
Corpus
A hierarchical knowledge repository designed to support bounded context, selective retrieval, and asynchronous coordination between agents.
WorkStream
An execution pipeline that transforms ideas into specifications, implementation efforts, and completed artifacts through coordinated agent workflows.
Recon
An autonomous technical analyst that evaluates development tickets before implementation begins — classifying work by investigation strategy, generating pushback questions, and resolving them against the codebase.
Argus
An adversarial voice dialogue platform for real-time collaborative review, designed around a floor manager state machine that mediates between human and AI participants.
Pulse
A channel agnostic communications and connectivity layer that coordinates context flow between systems, channels, and participants.
Field Notes
Current Focus
Autonomous Evaluation
I'm currently exploring how investigation pipelines can self-improve through their own output. The system classifies tickets into cognitive archetypes, routes them to specialized investigators, and uses codebase retrieval to resolve its own questions — achieving 70-77% autonomous resolution on production tickets. The current focus is closing the feedback loop: using evaluation quality signals to generate and validate new specialist capabilities without human curation.
This work is being developed alongside Lux (codebase intelligence) and a Faculty sub-system that manages specialist knowledge across workspaces.
Research Method
I treat these systems as a form of applied research.
Ideas are developed, implemented, observed, and refined through use rather than purely theoretical design.
Most writing and drafting is assisted by language models as editorial tools, but the architecture, experiments, and system behavior described here reflect ongoing hands-on work.
Last Updated: 2026-03-20
An agent without codebase search is a questionnaire generator. An agent with it is an analyst.
The archetype selects the investigator, not the category.
The compliance-alignment distinction is well-trodden ground. Watching it play out in a production pipeline gives it a texture the abstract framing misses.