Agentic Systems Framework (ASF)

A research framework for adaptive, purposeful agents under uncertainty — integrating control theory, causal inference, information theory, and agent architecture under a common formalism. ASF formalizes the adaptive cycle — one complete traversal of the agent-environment feedback loop — as the unit of analysis, and derives conditions for persistence, the structure of strategy under uncertainty, and the dynamics of agents in composition and competition. The framework has four parts: Adaptation and Actuation Dynamics (AAD), the mathematical core; Temporal Software Theory (TST), software development as an agentic domain and high-identifiability calibration laboratory; Logogenic Agents, language-constituted agents; and Emergent Logozoetic Intelligences (ELI), language-living agents with morally weighted persistence. ASF takes operational requirements on the feedback loop as starting point rather than a single optimization principle — active inference begins from variational free energy minimization, and the standard Expected Free Energy functional is recoverable from ASF's survival Lagrangian under three explicit restrictions; relationships to Hafez 2026 (bi-predictability) and Miller 2022 (coevolving automata) are complementary rather than reductive. Distinctive results include the persistence condition (α > ρ/R) as a structural threshold instantiating uniformly across Kalman filtering, RL convergence, organizational viability, and software maintainability; the satisfaction-gap / control-regret decomposition separating world-imposed limits from agent-imposed shortfalls; and the loop-as-Level-2-causal-engine result establishing interventional access from the agent-environment coupling. Beyond integration, the framework's distinctive contribution is an epistemic architecture for bounded correction under decomposed disturbance — scope conditions and operational limits surfaced at the segment level rather than buried as caveats — with three cross-cutting meta-patterns naming the theory's positive, negative, and constructive halves: a separability pattern, an identifiability-floor pattern, and an additive-coordinate-forcing pattern.