Kaname Manifesto
We are uncovering better ways of delivering software in an age when artificial intelligence generates code faster than humans can ensure it reflects their intent. The risk is not that AI builds the wrong thing slowly - it is that AI builds the wrong thing fast. Through this work, and helping others navigate the same shift, we have come to value:
That is, while there is value in the items on the right, we value the items on the left more.
On These Values
Specification clarity over implementation velocity
expandcollapseArtificial intelligence generates code at machine speed. Speed without clarity produces incorrect systems faster. The bottleneck in AI-augmented delivery is not how quickly code is generated - it is how precisely humans can articulate what the system must do. We value knowing what to build over building it fast.
Human judgment over AI momentum
expandcollapseArtificial intelligence does not determine intent, validate architecture, or accept delivery. Human judgment at defined gates governs every transition in the delivery system. The speed of AI generation does not justify bypassing this judgment - it makes it more necessary. An AI agent cannot carry work across a gate boundary.
Living specification over the codebase as truth
expandcollapseLiving specification over the codebase as truth. Most engineering teams treat the codebase as the source of truth: what the software does is defined by what the code does. Kaname inverts this. The specification is the source of truth; the codebase is a derived artifact. When they conflict, the specification governs - the code must be corrected, not the specification amended to match behavior. A specification that remains authoritative can regenerate implementations as requirements evolve, tools improve, or systems must be rebuilt. A codebase that has drifted from its specification is not a truth - it is an approximation.
Verified intent over working code
expandcollapseVerified intent over working code. The Agile Manifesto values working software - software that runs. Kaname requires more: software that not only runs but has been verified against the Use Cases that defined it. Code can compile and pass tests while diverging from the specification. AI-generated code amplifies this gap: it produces functional-looking output that may approximate intent rather than embody it. The Delivery Gate exists because working is not sufficient - verified is the standard. A human reviewer who has confirmed software against explicit acceptance criteria is the signal that delivery is complete.
Principles
Our highest priority is to deliver software that faithfully reflects human intent - verified against the specification that defined it, not merely software that runs.
Welcome specification changes, even during implementation....
Governed change paths exist for this purpose. AI agents regenerate affected artifacts; human gates remain the governing boundary.
Deliver frequently, but deliver verified....
A Use Case delivered and confirmed against its acceptance criteria is worth more than many features that approximate requirements.
Specification owners and implementers must engage throughout delivery - not only at gates....
A specification that requires clarification during implementation is an incomplete specification. The gap belongs in the specification, not resolved locally by the implementer.
Build AI-augmented delivery around humans with clear, named accountability....
Give them the specification context and the safety to exercise judgment - including the judgment to raise when a specification is wrong. Accountability here means dependable ownership, not threatened compliance. Kaname presupposes trust; it cannot manufacture it.
The most effective way to convey intent to an AI agent is through structured, precise specifications - not ad-hoc prompts....
Prompts are transient. Specifications are versioned, shared, and authoritative.
Verified Use Cases are the primary measure of delivery progress....
A Use Case is complete when its acceptance criteria are satisfied and a human reviewer has confirmed it - not when code exists that approximates it.
Human gates make AI-augmented delivery sustainable....
Without them, AI velocity creates pressure that teams cannot absorb indefinitely. Gates protect the human layer of the delivery system against the speed of the machine layer.
Continuous attention to specification quality is a form of technical excellence....
A Use Case that is imprecise generates an implementation that is imprecise. Specification work is not overhead - it is the primary investment.
Generate only what the specification requires....
AI agents will produce what they are asked. The specification defines the boundary. Everything outside it is waste - generated fast.
The best implementations emerge from clear human intent, not from AI inference....
AI cannot determine what a system should do - it can only implement what is specified. The quality of AI output is bounded by the quality of the specification it receives.
At defined intervals, inspect the delivery system - not just the work items - and adjust its policies....
The path from specification to verified software is the system. Improving this path is ongoing work, never complete.
A Note on Foundations
This manifesto builds on work that preceded it. The Agile Manifesto (2001) established that individuals and their interactions matter more than processes and tools, that working software matters more than documentation about it, that customer collaboration matters more than contract negotiation, and that responding to change matters more than following a plan. These values have not expired.
What has changed is the nature of the work. In 2001, the constraint on software delivery was implementation capacity - skilled humans writing code was scarce and expensive. The agile values reduced friction around that constraint.
In the age of AI-augmented delivery, implementation capacity is no longer the constraint. AI agents can generate working code from a clear specification in seconds. The new constraint is upstream: the ability to specify precisely what the system must do, and to govern the delivery of AI-generated output against that specification.
The Kaname values do not replace the agile values. They address the layer that AI has exposed.
The full guide defines the methodology behind these beliefs.