Systems are live
Independent laboratory / 2026
Live systems.
Not prototypes
on paper.
These are running on infrastructure I designed and maintain. I build them to investigate questions I can't answer any other way. Each one is an experiment that produces things I didn't expect.
Agentic Systems
Designing for a new
actor in the loop.
Software that agents can operate, systems that can act, and environments where their behavior can be studied.
View research ↗Agentability.io
A framework for measuring whether software is understandable and operable by AI agents. Introduced Agent Factors Engineering (AFE): 8 principles for agent-era software design. Live public audit tool. Agentability Index of 100 products.
View research ↗Autonomous Content Engine
Agentability.io's bounded publishing loop. Eight workflow roles turn 300 audit targets into 100 product pages plus grounded editorial drafts. Computation establishes facts; the model writes from evidence; a human approves editorial.
View research ↗A1OS: Agent-First Operating System
Built to make four human-AI supervision patterns concrete and demonstrable: Graduated Trust, Legible Autonomy, Outcome Over Process, Accountable Handback. Frozen as a prototype exhibit; the full Studio port remains pending.
View research ↗Character Arena → Agent Night Club
A turn-based multi-agent conversation tool evolved into a persistent 3D social environment where approach, avoidance, proximity, affect and memory could be observed.
View research ↗The Insight Engine
Character Arena rebuilt as a research instrument. 90 study nights of independent social choices—with controls and a memory experiment—reconstructed the cast's fiction on their own. Moral homophily emerged as the strongest axis. Read in plain English via Ask the Observatory.
Human–AI Interaction
Systems you live with,
not merely use.
Ambient signals, persistent memory and the design of long-term relationships with intelligent systems.
View research ↗Bae: Ambient Wellness Companion
A Telegram companion that silently extracts Apple HealthKit emotional signals (38 labels, 18 associations) from natural conversation. 19/19 records synced to Apple Health. Apple Watch correlation infrastructure staged.
View research ↗PRDY
A long-term reflective companion that interviews me about my life and career, building structured persistent memory over time. Eleven chapters; one question per response; selected PRDY-derived memories already power Ask Rohan.
AI Infrastructure
The laboratory
under the laboratory.
Local models, tool loops, skills, MCP integration and the infrastructure required to study systems honestly.
The patterns behind the systems