The paper introduces notlob, an environment for building literate programs together with language-aware machine agents. It provides a grammar for executable program essays, a parser that treats names as first-class objects, an internal name-graph relating prose, names, and executable artifacts, and bindings to existing programming languages and testing toolsets. A working implementation includes bindings to three established languages and several example programs.
The design is motivated by three observations: natural language dominates AI-assisted programming, machine-checkable artifacts are valuable feedback surfaces for , and the LLM context window is the dominant constraint. By co-locating code with its most relevant natural language and structured data, notlob aims to use LLM context windows more effectively and give coding agents IDE-like symbol-aware search and usage tools. The authors emphasize design and examples, leaving experimental evaluation to future work.