Applying governance logic and organisational design principles to the problem of AI alignment — where control must be precise, constraints must create freedom, and structure must scale.
AI alignment is not purely a technical challenge. It is a structural design problem — the same class of problem that emerges whenever autonomous agents must operate within boundaries they did not choose.
Organisations have solved versions of this for decades: how to grant autonomy while maintaining control, how to create rules that enable rather than restrict, how to build systems that remain coherent under uncertainty. These solutions are transferable.
My work translates the logic of organisational governance — constraint hierarchies, decision architectures, observational feedback systems — into structural frameworks for AI alignment. The result is a set of tools that complement technical approaches with the kind of systems-level thinking that high-stakes environments demand.
The full argument — governance logic applied chapter by chapter to AI-human alignment — is being written and published as a growing set of chapters. It is a work in progress: expect chapters to be added, revised, and reordered as the thinking develops.
View the Workbook ContentsPreviously published books and papers, offered as context for the workbook.
A governance-first approach to alignment that treats AI control as an organisational design problem. Introduces a 24-point periodic system for mapping the constraint landscape of AI-human interaction.
Read paper →Examines the counterintuitive principle that well-designed constraints create rather than limit productive freedom — and why this matters for AI control architectures.
Read paper →How self-reflective feedback loops and observational restraint create natural governance mechanisms — applicable to both human organisations and artificial intelligence systems.
Read paper →Drawing on experience across McKinsey, Rio Tinto, and the Fred Hollows Foundation — designing systems where governance, autonomy, and accountability must coexist.
Full background →