Every previous wave of technology replaced what people did with their hands. This one produces work faster than anyone can judge it.
That is the whole problem, and it is not a technology problem. The models work. They draft, summarise, analyse, and recommend at a volume no organisation was built to absorb. What has not scaled is the human capacity to direct that output, verify it, and own the consequences when it is wrong.
So the return on AI is not determined by the tools. It is determined by the organisation around them: where authority sits, how far trust extends, and who carries the decision when the output is plausible but wrong.
Human Intent × AI = Sustained Advantage
The relationship is multiplicative, not additive. Where human intent is absent, the multiplier is zero, and more capability produces more unowned work rather than more advantage.
The Core Agency Architecture is the blueprint for building the organisation that makes the multiplication work. It locates authority, scales trust, and assigns the mandate.
It is in three parts:
Diagnosis. Why organisations are absorbing capability without return, and why the failure modes look like technology problems but are not.
The blueprint. The Core Agency Architecture itself. Purpose and direction, then the decision envelope: scope, authority, accountability, and time. Then the operational design and the relational limits that govern how far any of it can stretch.
The mandate. What leaders have to decide, reserve, and defend for the architecture to hold.
The work draws on established research, such as Gulati on purpose, Ton on standardisation and slack, Collins on discipline, Dunbar on relational limits. It reconstructs that research for a context none of it was written for. The underlying findings hold. What changes is the constraint they now have to survive.
I’ve been drafting since 2026. New sections of Core Agency Architecture published as they’re ready.
Start here with your reading experience.