Long-Horizon AI-Augmented Engineering under Complexity

Until recently, only large organizations could build highly complex systems. What they supplied was not only labor, but the structure that held their work together over many years. With agentic AI, a founder or a team of two or three can now attempt such systems. Capability alone is not enough: without explicit structure, long-horizon work drifts, repeats itself, and loses coherence. This framework provides that structure, so that a small team can keep a complex project coherent across time, let AI agents do most of the work reliably, and keep human authority over what matters.

To apply it to a project, start with the Quick Start.

The Framework for Keeping Complex Work Coherent Across Time is a system of principles, normative requirements, domain bindings, guidance, and checks for complex engineering work. It is domain general and intended to serve as a blueprint for your projects, as well as the next generation of AI tools supporting long-horizon work.

Its core parts are:

• Governing Principles: the authoritative principles that define the governing model.
• Normative Structural Specification: the mandatory project architecture that puts the Governing Principles into practice.

The Governing Principles and the Normative Structural Specification define the domain-general core.

How the framework was developed

The framework was derived from mechanisms found in recorded engineering practice: the methods by which large organizations such as SpaceX, NASA, Microsoft and others have kept complex work coherent over many years. From that record, 79 recurring mechanisms were collected as Generalized Patterns.

The five Governing Principles and the 106 clauses of the Specification were then written against that record. Each clause names its grounds: a Generalized Pattern, a Governing Principle, or a part of the Goal, which states the objective the Framework must meet. No clause stands without a stated reason.

The result was analyzed via a semantic graph which represents the clauses, their grounds, their terms, and their relations to the Goal. Static analysis of this graph shows which grounds each clause rests on, which goals it serves, and what a change to one clause can affect. An internal rebuild from fixed inputs and recorded decisions reproduced the recorded graph exactly.

The paper Long-Horizon AI-Augmented Engineering under Complexity gives the reasoning behind the principles and the clauses, explains their derivation, and reports the analysis.