Intention Coding: From Vibe Coding to the Executable Realization of Persistent Human Intent

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

Vibe coding has made a profound change in software creation visible: humans can increasingly express what they want in natural language while artificial intelligence systems generate, modify, test, and repair the code that realizes that intention. This paper argues that the deeper transition is not simply from manual coding to AI-assisted coding, but from code-centered software development toward intention-centered software realization. We introduce Intention Coding as an architectural concept in which persistent human intent becomes a first-class engineering object and source code becomes one executable realization of that intent rather than necessarily its primary or enduring representation. Under this view, a prompt is not identical to an intention, and a generated implementation is not identical to the software identity it temporarily realizes. The same persistent intention may produce multiple implementations across languages, architectures, revisions, models, and execution environments while remaining subject to common constraints, expected behaviors, tests, policies, and governance requirements. This shift creates new engineering problems around intent representation, intent preservation, intent–code divergence, regeneration, realization equivalence, provenance, verification, drift, and reconstruction. It also changes the meaning of software maintenance: modifying or regenerating code should not be evaluated only against the previous implementation, but against the persistent intention the implementation is supposed to realize. The paper positions vibe coding as an important entry point into this transition while extending the earlier concept of vibe tech toward a more specific software-engineering architecture. In enterprise environments, Intention Coding requires more than successful generation. Persistent intent must survive architecture, security, testing, compliance, observability, operational change, delegated execution, and responsible promotion. The central question therefore becomes not whether AI can generate working code, but whether software engineering can preserve, verify, and govern the relationship between persistent human intent and the changing executable realizations through which that intent becomes operational.