Self-Improving AI-Native Organization - A Grand Unified Theory

Thesis. An organization is a set of journey graphs over distinct populations - prospects, customers, candidates, hypotheses, dollars, claims - composed over one sovereign substrate, improved by one loop discipline, and governed by one constitution. The predecessor proved the mechanism on the function with the fastest feedback and the deepest empirical tradition. The organizational theory requires almost no new machinery: it inherits the questions, the layers, and the substrate, and adds only what composition itself demands - junction contracts between graphs, an internal signal economy, org-altitude allocation, and a doctrine layer that amends across every function at once.

1. The function catalog

Every business function is a journey graph over a population, and differs from its siblings only in the parameters below. The catalog is open (vendors, compliance, and community extend it); these six are load-bearing.

Note the recursion in the last row: the knowledge function's journey graph over claims is Layer 6 - the loop that improves every function appears in the catalog as a function, improvable by itself. The tangled hierarchy is not an appendix to the org; it is a row in its chart. And note its latency: the shortest loop in the organization, which is why the corpus compounds faster than anything it describes.

2. Composition: graphs join at edges

Functions never read each other's internals; they meet at junction edges whose crossings are published, dimension-rich events from a shared taxonomy family. Canonical junctions: the close edge (Marketing → Operations), the hire edge (Talent → every function's seats), the ship edge (Product → Marketing's claims and Operations' playbooks), the renewal edge (Operations → Finance), and the learning edge (every function → Knowledge). Junction contracts are versioned; a function may swap its internals freely so long as its edges keep their promises - the bounded-context discipline at organizational altitude.

3. The internal signal economy

The flywheel, generalized: each function's deepest outcomes are another function's priors. Closed-won and closed-lost reasons re-weight the product roadmap; churn reasons re-draw the ICP; support transcripts seed copy; sales objections re-price the offer; seat-trial results recalibrate hiring filters. Depth-weighted signals route across junctions on a declared cadence, and the organization's real competitive variable is signal transit time: how fast a truth learned in one function lands, grounded and canonized, where a second function can act on it.

4. The org chart as versioned graph

Roles, not people, at every node; occupant timelines; human, agent, or paired occupancy per seat. The automation frontier becomes a queryable assignment policy with a history; earned autonomy is granted per seat on evidence. Reorganizations are occupant reassignments, never graph rewrites. Staffing is a loop the system runs: the seat trial is experimental selection over people, founder included, with pre-registration and sample floors.

5. Allocation: the meta-loop

The organization's scarce resources - founder attention, audience, cash, trust - flow to the constraint that is binding across functions, not to each function's local optimum. Function north stars are sub-metrics of the organizational north star, and the sub-metric law holds fractally: a function maximizing its own star can damage the whole non-linearly. Exploration budgets are declared per function in that function's native currency; the allocation currency across them is information per unit of scarce resource.

6. One doctrine layer

A single corpus spans every function under one amendment protocol and per-artifact mutation laws. Coherence waves cross function boundaries in one pass: a corrected funnel constant amends the revenue model, the hiring gates, and the playbook together, or phantom canon multiplies. The corpus is the organization's continuity - its identity across personnel, vendor, and model change - and the onboarding property (a newcomer, human or agent, operational in thirty minutes) is its acceptance test.

7. Governance: the constitution

Invariants beyond any amendment's reach. Blast-radius approval altitudes generalized: instance → policy → constitution. Trust tiers on every artifact, with derivation never laundering trust across agent fleets. Counterparty protection per population - prospects, customers, candidates, employees - as hard constraints, not preferences. The human's terminal role: constitutional editor.

8. Org-level failure modes

Function silos (the sub-metric trap, fractal). Phantom canon across boundaries. Reorg-as-rewrite. Goodhart at the organizational north star. Conway drift - the org graph and the system graph diverging until neither is true. Cross-function attribution wars. And the sophistication trap's final form: building the organization-OS while no single function's loop yet fires against reality.

9. The maturity axis

Functions climb L0–L6 independently; the organization has its own ladder, and it measures composition: O0 isolated, tool-resident loops → O1 one shared substrate → O2 cross-function signal routing live → O3 unified doctrine under one amendment protocol → O4 the asymptotic organization, whose improvement rate durably exceeds its environment's decay rate and whose leverage is bounded by the physics of insight capture, not by headcount or willpower. Sequencing rule: one function to L2+ before composing anything - and marketing first, because its feedback is fastest. That rule is why the predecessor paper exists.

10. The asymptote

The organization whose accumulated, versioned, agent-readable substrate is the asset; where humans govern and agents operate; where every function's contact with reality amends the doctrine that generates every function. The company is a graph of graphs - and the graph learns.