The open methodology for human–AI collaboration
AI executes. You orchestrate.
Your models will change. Your methodology won't. Cognitive Orchestration is the layer above the tools — the deliberate design of context, boundaries, and judgment around the machine, so intelligence compounds instead of evaporating at the end of every session.
§01 · Why now
The first era of AI asked can the machine do it? That era is ending in a tie.
Every serious tool now writes, plans, remembers, and acts. Capability is becoming a commodity — the same power in everyone's hands, priced toward zero, replaced every few months.
The second era is different. It asks how to turn what a machine can do into something a person, a team, an organization can trust, repeat, and build on. That isn't a cleverer prompt or a bigger model. It's a craft — and it has a name.
- prompt engineering optimizes a turn.
- context engineering optimizes a session.
- cognitive orchestration designs the institution — get it right once, and make it hold across every session, teammate, and model that comes next.
§02 · The manifesto
Nobody set out to build a methodology. We set out to go fast.
We pushed AI agents as hard as they would go — to write, to build, to research, to decide — chasing nothing but speed and quality. And somewhere past the point where the machine could do almost anything, the real question surfaced. Not can it? but should it — and who decided, and how will we know it's still true tomorrow?
That question doesn't live inside any tool. It lives above them.
Orchestration is that layer. It works with whatever you already use — any model, any agent, any vendor — because it governs the collaboration, not the software. Its core claim is quietly radical: your job was never to type faster than the machine. Your job is to be the architect. To define what good means here. To hold the boundary no model can hold for itself. To make this session's lesson the next one's foundation.
Governance, we learned, was never the brake on autonomy. It was the engine.
The models will keep changing. The methodology is what you keep. This is open. CC BY 4.0. Yours to build on.
Orchestrate.
§03 · How it works
Five layers. Six phases. Two human gates.
Cognitive Orchestration is small enough to recite and deep enough to run in production. Five layers hold what your organization knows. Six phases move work from idea to shipped. And of those six, only two need a human — the rest converge on their own.
High autonomy. Human authority exactly where it counts.
Five layers. Each encodes a decision a human makes — and the machine then executes inside.
- L5 Learning The system watches, captures, and gets better across sessions. compounding
- L4 Instructions Work moves through phases with approval gates — not one giant prompt. process
- L3 Guardrails The rules that must never break live outside the AI's reach. risk tolerance
- L2 Context Write down what your organization knows, in a form the machine can read. institutional knowledge
- L1 Intent Route the work to the right specialist — not one generalist doing everything. structure
§04 · One methodology, five front doors
It isn't a coding trick. It's a family.
The same seven principles, adapted to whatever work you do with AI. Each domain instantiates the five layers for its own three failure modes — and the list is a template, not a limit. Derive your own.
- COC Codegen AI-assisted software with institutional memory. in production · proof →
- COR Research Human-directed inquiry; AI teaches, you author. in production · in practice →
- COE Education Assess the collaboration, not just the output. piloting · the approach →
- COG Governance Standards and decisions that show their work. in production · self-hosted →
- COF Finance Judgment-heavy operations, run on the loop. in production
- CO·? Your vertical Map your three failure modes, design the five layers, define your gates. New domains are born in the open. build your own →
§05 · The seven first principles
The nucleus. Recite it.
- 01
The model is a commodity. What your organization knows is the moat.
Institutional Knowledge Thesis - 02
Every AI session is a genius who just walked in — with total amnesia. Onboard it like one.
The Brilliant New Hire - 03
AI fails three predictable ways: it forgets, it drifts, it's blind to danger. Design for all three.
Three Failure Modes - 04
Not in the loop typing. Not out of the loop hoping. On the loop — setting the boundaries.
Human-on-the-Loop - 05
Don't ask the AI nicely to follow the critical rule. Enforce it where it can't talk its way out.
Deterministic Enforcement - 06
The more the AI does, the more you must understand. Automation deepens expertise — it doesn't retire it.
Bainbridge's Irony - 07
Every session should make the next one smarter. Capture what you learn, or repeat it forever.
Knowledge Compounds
§06 · Who builds with CO
A movement, not a marketing page.
- convener Terrene Foundation Publishes and governs the CO specification of record
- your name here Be among the first. The founding cohort is forming. Sign the manifesto and list your team or organization as a builder-with-CO. Sign the manifesto →