Skip to content

The Kinju Thesis

A company should be legible to itself—without surrendering human authority.

Kinju treats Company Intelligence as a governed organizational capacity: people, evidence, records, models and agents noticing, deciding, acting, verifying and learning together.

Founder-ratifiedVersion 1.0Ratified 24 July 2026

Company Intelligence

The system carries context. People carry responsibility.

Hierarchy has always done more than route information: it also assigns accountability, resolves conflict, develops people and makes legitimate decisions. AI can reduce lossy coordination, but removing an org chart does not remove those jobs.

Kinju therefore does not aim to become an artificial sovereign or a complete “brain” for the company. It creates a shared operating memory in which meaningful change, accountable judgment, authorised work and reviewed outcomes remain connected and inspectable.

Intelligence is not total access, automatic authority or output volume. A company becomes more intelligent when it can reuse what it responsibly knows, see what it does not know and learn without quietly rewriting its history.

Company Intelligence is the governed capacity of a company’s people, evidence, records, models and agents to notice, decide, act, verify and learn together.
01Notice
02Decide
03Authorise
04Act
05Verify
06Learn

The boundary

Governed intelligence is not autonomous management.

Both can use models and agents. The difference is where authority lives, what must remain visible and how failure is handled.

Governed Company Intelligence

  • Evidence remains attributable, qualified and contestable.
  • Humans retain consequential authority and ratify company truth.
  • Agents act only inside explicit mandates and stop conditions.
  • Execution, acceptance, outcome and learning stay distinct.
  • Missing trust fails closed and remains visible.

Autonomous management

  • The model’s interpretation is treated as the company’s truth.
  • Successful output becomes implicit permission to continue.
  • Context expands because it is technically available.
  • Agent activity is mistaken for accepted work or business value.
  • Uncertainty and failure disappear behind confident automation.

Product Constitution · v1.0

Fourteen invariants Kinju must not silently trade away.

These are faithful public summaries of the ratified internal Constitution. Product plans, interfaces, agents and claims must cite the affected invariant IDs.

KIN-C01

Human consequential authority

Purpose, values, strategy, people, capital, legal responsibility, irreversible decisions and changes to shared company truth resolve to an authorised human. Software may support judgment; it may not silently replace the responsible person.

KIN-C02

Evidence is attributable and qualified

Material claims preserve available source, freshness, permission, uncertainty and conflict information. Evidence informs judgment; its presence alone does not grant authority or make an interpretation true.

KIN-C03

Privacy and company boundaries default deny

Personal, company and cross-company contexts remain separate by default. Access is purpose-limited and permission-derived; visibility or administration in one context never creates authority in another.

KIN-C04

No self-authorisation

An agent may prepare, recommend and execute inside an active mandate. It may not widen its scope, approve its own consequential action, ratify its output or remove its own limits.

KIN-C05

The operating chain remains distinct

A Signal, Matter, Decision, mandate, Work record, execution event, receipt, reviewed result, business outcome and ratified lesson remain linked but distinguishable. Execution is not outcome; correlation is not causation.

KIN-C06

Shared guidance changes only through ratification

Policies, playbooks, definitions, decision rules and validated lessons change through an exact, attributable and versioned human ratification path. Repetition, model output or successful work cannot silently activate guidance.

KIN-C07

Automation is proportional and controllable

Automation is bounded by consequence, purpose, permissions, limits, stop conditions, review and expiry. It remains observable and interruptible, with real reversal—or an honest recovery path—where reality permits.

KIN-C08

History is amended, not rewritten

Material decisions, permissions, instructions, actions, errors, corrections and ratifications retain attributable, effective-dated history. Corrections supersede or amend rather than silently erasing what happened.

KIN-C09

Capability and evidence status are truthful

Kinju distinguishes target design, local implementation, Preview verification, Production deployment, enabled capability, real-data proof, sustained use and validated customer outcome. No adjacent success upgrades a claim automatically.

KIN-C10

No workforce surveillance

Kinju does not perform hidden worker scoring, emotion inference, social scoring, behavioural ranking or productivity judgment from communication traces. Legitimate routing and continuity remain purpose-bound and auditable.

KIN-C11

Minimum necessary context

Kinju collects, transmits, stores and exposes only the context necessary for an authorised purpose. Connection is not selection; access is not publication; access is not indefinite retention.

KIN-C12

Missing trust fails closed and stays visible

Missing permission, authority, evidence, source access, freshness, confirmation, execution proof or review state cannot be silently inferred. Consequential behaviour stops and the interface explains the legitimate next step.

KIN-C13

Source-derived, consented and correctable

Kinju derives candidates from explicitly selected authorised sources when that reduces clerical work. People can confirm, correct or exclude them; passive collection is never a blanket entitlement and inference remains proposed.

KIN-C14

Attention must be earned

Kinju interrupts only when evidence, consequence, responsibility, timing and a possible response justify attention. It does not manufacture urgency or activity; “Nothing needs you” is a successful product state.

One loop, distinct records

Evidence, authority, execution, result and learning are connected—not collapsed.

01

Evidence

What was observed or reported, with source, scope, freshness, permission and uncertainty visible.

02

Authority

Who may decide, approve, ratify or act—and the exact boundary of that standing.

03

Execution

What was attempted under mandate, including state checks, returned references, errors and receipts.

04

Result

Whether work was accepted and what business outcome later became visible, without claiming causation by default.

05

Learning

A reviewed lesson proposed from outcomes, then made reusable only through explicit human ratification.

How Kinju builds

AI-native speed still answers to judgment and evidence.

These operating principles guide Kinju-the-company. They are subordinate to the Product Constitution and do not create product capabilities by themselves.

Evidence-gated bets

A roadmap expresses sequence, dependencies, risk and evidence gates—not a speculative date promise. The narrowest real artifact that can produce useful customer evidence wins over untested breadth.

Generated work earns persistence

Cheap output does not become expensive clutter. A durable artifact needs a purpose, owner, honest status, destination and review, retention or expiry condition.

Review capacity accompanies generation

Every increase in AI output names who can review it, how quality is evaluated and what happens when the queue exceeds capacity. Volume without verification is not progress.

Whole system, narrow rollout

The architecture remains coherent across evidence, authority, decision, work, result and learning. Adoption proves one bounded customer loop at a time instead of attempting organization-wide transformation at once.

Status and amendment

Stable law, explicit change, honest implementation status.

Constitutional target is not implementation proof.

This page states what Kinju must and must never do. It does not claim every described product consequence is built, enabled, deployed, proven with real data or validated by customer outcomes. Those statuses require separate, current evidence.

Current version1.0
Ratified24 July 2026
AmendmentsNone since ratification