Topic hub
Agentic Architectures
How do we build agents whose loops, tools, memory, and checks produce reliable work?
The technical architecture around agents: habitats, harnesses, feedback loops, tool boundaries, intermediate state, verification, and the conditions under which autonomy remains inspectable.
Research threads
02What can deterministic machinery genuinely establish, and which uncertainties remain outside its tests, specifications, and repeatable outputs?
openHow do checks and balances make human and AI decisions trustworthy?What institutional machinery lets an organization delegate reasoning while preserving evidence, authority, accountability, and the ability to intervene?
Writing
03Draft
Designing the Agent's Habitat
A framework for roles, harnesses, authority, and handoffs in agentic organizations
Working note
Agentic Organizations Through Control Theory and Cybernetics
A crosswalk between habitats, harnesses, handoffs, and the language of governed systems
Working note
A Unified Theory of Epistemic Engineering
A four-dimensional theory of AI system difficulty, with practical machinery for designing and diagnosing production systems
Code and models
02Cataloguing
Practical AI Catalogue
Mapping deployable multimodal AI capabilities to real business functions, constraints, and evaluation methods.
Prototype
LLM Ranker
An experiment in comparing coding models using cost, reliability, tool-use accuracy, and task-specific tradeoffs.
Still unresolved
Open questions
- Which intermediate states must an agent expose for its work to remain diagnosable?
- When should a dynamic handoff become a deterministic contract?
- How should feedback loops terminate, escalate, or hand responsibility to a person?