The Role of Code and Programming Languages in the Era of Agentic LLMs

Redefining 'Code' with LLMs

The advent of agentic Large Language Models (LLMs) has initiated a discussion on the traditional understanding of "code" as the final product of software engineering. With LLMs capable of generating code, some suggest that the specification, rather than the generated code itself, becomes the primary artifact, as code can be regenerated on demand.

The Persistence of Specification

Despite the ability of LLMs to produce machine instructions, the article posits that the instructions or prompts provided to the LLM effectively serve as the new form of "code." This represents a higher level of abstraction but still requires a clear and unambiguous specification to guide the LLM's output.

The Enduring Role of Programming Languages

The article argues that programming languages will not become obsolete. Their fundamental purpose is to reduce ambiguity inherent in natural language specifications by providing a structured and semantically precise means of communication. This precision is crucial for ensuring predictable execution, a role that remains vital even with LLM-driven code generation.

Future of Development

While programming languages might adapt to integrate with this new paradigm, their core function of providing unambiguous specifications will persist. The ability to generate binaries directly from LLMs is considered feasible, but the verification of what the binary will do still relies on a form of higher-level code or specification.