Systems & thinking

A library of live research threads.

Not a blog. Each thread is an open enquiry — some active, some prototyped, some dormant but interesting.

Mapping messy organisational situations as ecologies of humans, documents, tools, agents, risks and reporting — every node both a part and a whole.

Systems where every part is both a whole and a component — and how that shapes multi-agent design.

The gap between agent demos and agents that actually work — delegation, memory, orchestration patterns.

Applying Checkland's Soft Systems Methodology to the design of AI-driven processes.

The interface is not a screen — it's a negotiation surface.

RAG done seriously — provenance, permissions, structure, and a clear relationship to the work.

Markdown vaults as the substrate for personal and team knowledge systems.

Tabletop rulebooks are messy, cross-referenced, contested knowledge — perfect testbeds for retrieval and reasoning.

Lesson planning, marking support, and source-backed tutor assistants.

Running cheap simulations of decisions before making them, with delegated AI stand-ins.

Knowledge systems for towns, councils and community organisations.

Designing fictional systems to train the same muscles needed for real ones.