Русский · English · 中文 · हिन्दी
[KP] is a language for specifying intent to large language models (LLMs), in which the model acts as the compiler. The discipline studies how to formulate requirements in Russian so that the model's output is predictable and fit for production use.
Programming languages have evolved toward higher levels of abstraction. Machine code described the machine. Assembly described instructions. High-level languages described logic. At each step, the developer stopped describing how to perform an action and started describing what the result should be. A new layer took on the translation into instructions: the assembler, the compiler, the interpreter.
Further progress was blocked: there was no executor that could understand a task description in its original form. A compiler applies formal rules but does not understand meaning.
Large language models (LLMs) provide an executor that understands a task stated in natural language. If a task description is treated as a program, natural language becomes a programming language.
In this view, the compiler is the model, the parser is the context, the type system is the constraints set by the author, and the debugger is the verification of the result.
The syntax of machine code is checked automatically: an error stops the build. Natural language has no such check. It accepts any form and always produces a plausible answer, so it can distort meaning without warning. In Russian this effect is stronger because of free word order, polysemy, words that are spelled and pronounced the same but mean different things, and aspect of verbs.
A language of the new level is designed the same way as the syntax of earlier levels. The author takes over the parser's role: they set the role, context, invariants, constraints, and output format. Whatever is not stated explicitly, the model fills in on its own.
The name "contextual" reflects the principle: meaning is defined not by a single phrase but by the whole environment. Context plays the role of the type system.
[KP] is an applied discipline at the intersection of software engineering and computational linguistics. Its subject includes:
[KP] is not prompt engineering. Prompt engineering picks a phrasing for a one-off answer. [KP] designs an environment for predictable and reproducible results.
The discipline is intended for experienced engineers and product managers. [KP] lets them formulate requirements for a set of AI agents without programming skills. The structure of the discipline gives predictable results at the product level.