A governed operating architecture for AI-coordinated teams
AI can coordinate the work.
It must never own the judgment.
Domondon Dominium is a universal operating model for teams that want the full capacity of AI coordination — with every consequential decision governed, evidenced, logged, and human-owned. Built from nursing doctrine, safety science, operations research, and modern AI governance. Tested where the stakes are a human being in a bed.
🕯️ Carry the lamp. Keep the ledger. Agents propose. Humans judge. Nurses steward.
The thesis
The operating manual for the AI era was written at the bedside
The disciplines that learned to keep humans safe around powerful, confident, fallible systems — nursing above all, alongside aviation and the high-reliability trades — already solved the problems every organization now faces with AI: how to delegate work without delegating judgment, how to escalate past a confident authority that is wrong, how to keep records that protect the vulnerable, and how to hold the line on what must never be handed off.
Nursing formalized this over 160 years, under law, at the point of consequence. Domondon Dominium translates that doctrine — with its cousins in safety science, military operations, cybernetics, and lean — into a complete, configurable operating architecture for any team, in any industry.
You are not building an AI tool stack. You are building a translation layer between how humans organize work and how AI systems coordinate support.
The foundation
Every team is the same system
Every team — ICU, marketing department, startup, nonprofit — is a system that runs eight verbs. Craft is the cargo; the system is the ship. That is why one architecture, configured differently per team, can govern them all.
Receives demand
Work arrives — asked for or not — through channels the team must see.
Applies judgment
Something decides what each request is, how urgent, and whose it is.
Performs work
The craft — one verb of eight, and rarely where the crisis lives.
Coordinates people
Handoffs, updates, meetings — where systems quietly bleed.
Uses knowledge
Policies, precedents, and the expertise walking around in people's heads.
Manages risk
Every team can hurt someone. Every team runs controls — or discovers their absence.
Produces outputs
The work leaves, and what leaves is the team's entire reality to those who receive it.
Improves over time
Or relearns the same lesson quarterly. The eighth verb decides which.
What AI changes
The economics — not the structure. Judgment and accountability stay scarce. Governance keeps the cheap capacity from impersonating them.
The doctrine
The Five Rights of AI Delegation
Nursing solved human-to-AI delegation before AI existed. The licensed nurse may delegate tasks to assistive personnel — never judgment — and remains accountable for the delegation decision, the supervision, and the outcome. Substitute "AI agent" for "assistive personnel" and the doctrine transfers word for word: the right task, under the right circumstance, to the right agent, with the right direction, under the right supervision.
Accountability never moves.
The short form
Ten principles
- Tasks delegate; judgment does not; accountability never moves.
- Tier the task, not the tool Oversight scales with risk and reversibility.
- Autonomy is earned with evidence, promoted deliberately, demoted automatically.
- A gate that never rejects is not a gate.
- Boundaries are architecture, not etiquette Enforce them with permissions, not prompts.
- Every escalation is praised; every near-miss is data; every reporter is safe.
- Loose coupling beats fast coupling wherever failure could propagate.
- Work-as-done teaches The gap from work-as-imagined is the improvement backlog.
- The human model is sovereign; the AI model derives from it.
- Scale good — or do not scale.
Who it's for
One architecture, any team
Clinical & healthcare teams
Born here. Nurse-governed judgment, No-PHI boundaries, evidence tiers, and human gates as permanent architecture — not compliance theater.
Business & operations leaders
A complete department operating model — intake, triage, decision rights, metrics — that makes AI coordination governable instead of hopeful.
Builders & AI governance professionals
An object–action–control ontology and module architecture aligned with NIST AI RMF, ISO/IEC 42001, and the EU AI Act's oversight and logging mandates.
The aha moment
When it clicks, it sounds like this
So it's not that we need better AI. We never wrote down how our team actually works.
The work was never the problem. It's everything around the work — and that part is the same everywhere.
I finally have language for why our pilot felt impressive and changed nothing.
We don't have a tool gap. We have a "who decides" gap.
The AI can carry my busywork without carrying my judgment. Those are different things — nobody had ever separated them for me before.
Nursing already solved this. I've been doing governed delegation for twenty years — I just never heard it called that.
For the first time, "human in the loop" isn't a vibe. It's a design.
If one of these is yours, the framework will feel less like learning and more like recognition. Start with the anatomy →