Programming is language design: Every defined name, be it a method or a data structure, extends the language. Hence every language seems to be a perfect framework for DSL development. Recently, however, several techniques that rely on syntactic abstraction mechanisms outside the programming language (such as code generator frameworks or meta-programming) have become popular. I will analyze these techniques and show that they do not scale, because DSLs developed this way lack basic composability properties. I argue that a programmers' answer to the DSL challenge is to embed DSLs, using the semantic abstraction mechanisms of the host language rather than external syntactic abstractions. Using concrete examples, I will show that embedded DSLs are easy to develop, simple to reason about, efficient, and - most importantly - composable. Moreover, DSL embedding requires only the weapons of a real programmer - a text editor and a compiler - rather than a complex toolchain. Based on the idea of DSL embedding, I will propose a principled architecture and design method called "polymorphic embedding", which can be used to compose and decompose DSLs in a plug-and-play manner. Finally, I will discuss the influence of the host language on DSL embedding. The programing languages we will discuss include Java, C#, Scala, and Haskell.