Emergence: A Programming Paradigm for Constraint-Shaped Agency

Executive Summary

The programmer does not write the program. The programmer writes the conditions under which the program emerges. Emergence is a programming paradigm in which the fundamental unit of computation is an autonomous agent that interprets messages within a constraint architecture (schemas, thresholds, scoring functions, and topological rules), and the system's behavior arises from the interaction of agents operating within those constraints.

The paper argues that this recovers Alan Kay's original vision of object-oriented programming, which required a computational substrate with contextual understanding that did not exist until large language model agents.

Four Well-Formedness Conditions

To make the paradigm engineering rather than hope, the paper defines conditions a constraint architecture must satisfy:

• WF1 Completeness: every agent output is evaluated by at least one scoring constraint
• WF2 Liveness: the lifecycle rules (fork, decay, merge) are each reachable, so the system can both grow and shrink
• WF3 Dissipation: energy is bounded and decreases whenever an agent acts, so no agent can process signals indefinitely
• WF4 Contraction: the Q-dynamics operator governing the system's acceptability distribution is a contraction, guaranteeing convergence to a unique fixed point

Contraction also addresses the control problem: the system converges to a predictable operating regime not because a controller forces it there, but because the constraint architecture's own dynamics are contractive.

Evidence

The paradigm is grounded in cybernetics (Ashby, Beer), modular design theory (Baldwin and Clark), organizational theory, and complex systems. It is validated against two existence proofs (a working stigmergic mesh and Tessera, a self-validating executable document format) and five empirical studies of the mesh. In those studies, distance between runs fell 91.6% across twelve trials, and the measured contraction rate was faster than the Gaussian model predicts, making the Gaussian rate an upper bound.

The paper treats the paradigm's limits with equal weight: latency, cost, non-determinism, and the new discipline of designing constraint spaces.

Key References