The Theory
Chapter 5 supplied two views of the evidence. The originating case showed how MAGE's structures emerged under sustained engineering pressure; the industrial reconstructions showed related structures arising independently under different constraints. Neither establishes a universal law. Together, they give us something worth explaining.
We begin with a general theory of how agentic capacity interacts with the governed engineering environment. The theory treats implementation capacity as an input, not an outcome: the same capacity can produce durable progress or merely accelerate churn depending on the representations, evidence, constraints, and controls through which it acts. Engineering pressure exposes mismatches. Engineers can diagnose them and turn what they learn into structure that later work inherits.
We state the theory and its main predictions first, deliberately in general form. Only afterward do we ask where those predictions should be expected to hold. MAGE costs something to build, maintain, coordinate, and govern, and not every engineering judgment can or should become durable machinery. The scope conditions therefore bound the theory rather than precede it.
The chapter then turns from claims to inquiry. Its final section turns the theory, predictions, and scope conditions into a research agenda: what to measure, what comparisons could distinguish the proposed mechanisms, which quantities remain unknown, and what evidence would strengthen—or weaken—the account.
The objective is not to declare MAGE a universal law of software engineering—or of agentic work more generally. It is to make the explanation precise enough to be wrong. The evidence behind the theory comes from software engineering. The mechanisms are stated more generally because representation, enforcement, inherited structure, and the interaction between agentic capacity and its environment do not themselves require software. §6.3 therefore asks not only when MAGE should work within software engineering, but what properties a domain must have for the account to transfer at all.
The synthesis
Agentic capacity does not by itself produce useful progress. It acts through the environment surrounding the work. In engineering, Modeling changes the representations used for consequential reasoning and Alignment enforces selected obligations independently of the producing reasoner.
Together, these structures shape how much autonomous capacity becomes durable progress, how much failure escapes, and how much human judgment the work still requires.
Carrying forward: Commodity intelligence · Modeling Principle · Alignment Principle · Governed Engineering Environment · Engineering capital · Probabilistic surface
New here: Dynamic Model · Environment quality · Determinization frontier · Representation innovation · Scope conditions · Testable predictions · Research agenda