Programming language semanticists have been struggling with assigning a precise mathematical meaning to programs. If the meaning of a program is platform independent, the correctness of the program can be established independent of any particular implementation of a specific hardware/software platform. But such an “extensional semantics” is not suitable for addressing the dynamics of program execution including execution time, memory usage, and power consumption. In this study, we propose a new programming paradigm called implicit programming to formally separate the notion of correctness (a semantic issue) from that of performance and resource usage during program execution (a pragmatic one). We shall show that this approach is sufficiently general to encompass approximate computing and probabilistic programming within a single framework. We then focus on its application in approximate computing and build a particular intent-specific programming language, FAST, to show how it allows users to code a variety of performance optimization tasks adaptive to different environments. Next, we will show how we could extend our implementation to support the more general continuous multi-constraint cases and to control multiple adaptive functions simultaneously without mutual interference. And finally, we will give a corresponding formal model of pragmatics (currently called cost semantics in the Programming Languages research community) and show how “intensional” (platform-dependent) properties can be formally established for particular platforms.