Vigorware: New Axioms for How Software Exists and How It Is Built

Conventional software engineering rests on four tacit assumptions: program outputs are deterministic; program structure is immutable after deployment; only humans make design decisions; and development inputs are precise instruction sequences. The capabilities emerging from large language models (LLMs) are steadily compressing the validity boundaries of every one of these assumptions, yet the dominant engineering response (absorbing AI capabilities as libraries, frameworks, or services atop existing languages) encounters structural difficulties in the type system, the runtime, and the representational substrate that cannot be resolved within the old assumption framework. We introduce Vigorware, a working theoretical framework that replaces the four legacy assumptions with four new axioms: Probabilistic Co-existence, Runtime Mutability, Multi-Agent Collaboration, and Intent-First development. Each axiom is given a formal operational semantics drawn from standard PL-theory tools (probability monads, algebraic graph transformation, pi-calculus, and a constructive extension of Hoare logic) and a quantitative degeneration metric in R⁴ (≥0) space that structurally characterises the traditional paradigm as the boundary special case at the origin. We further propose six core features as the key structural supports required for the four axioms to be realised under current technological conditions, analyse their chain coupling through an Experience Solidification Pipeline, and illustrate an initial feasibility point in the design space through the thought-experiment language Capo. The framework is anchored simultaneously in the cognitive profiles of both LLMs and humans, rather than serving solely to unlock AI capabilities. We establish five falsifiable external entry points through which subsequent research may refute specific claims, and identify an open research agenda spanning the PL--SE disciplinary boundary and beyond, positioning Vigorware as a working theoretical framework open to discussion and revision rather than a definitive pronouncement.