TL;DR
NERD (No Effort Required, Done) is an experimental language designed for code produced primarily by large language models. It favors dense, English-token-friendly syntax, compiles to native code via LLVM, and aims to reduce token use and runtime dependency overhead.
What happened
An experiment called NERD proposes rethinking source code for an environment where LLMs are the primary authors. The creator observed that language models tokenize English words more efficiently than punctuation-heavy programming syntax, and designed a terse, machine-oriented language that uses compact English-like tokens to express logic. NERD is meant to be auditable and verifiable by humans but not edited line-by-line by them; instead, humans issue intent and LLMs produce or modify NERD source. The implementation is a small bootstrap compiler written in C that emits LLVM IR and requires no runtime. The project claims substantial token and cost savings — examples cited include roughly two-thirds fewer tokens compared with equivalent TypeScript code — and positions NERD as a potential next step if machine authorship of production code continues to grow.
Why it matters
- Potentially lower LLM API costs because English-word-heavy syntax can reduce token counts by a reported 50–70 percent.
- Shifts human roles from direct authors to reviewers and policy stakeholders, altering team workflows and developer tools.
- Native compilation to LLVM IR with no runtime could simplify deployment and reduce runtime dependencies.
- Claims of improved auditability: a machine-oriented source that is designed to be human-observable rather than human-authored.
Key facts
- The author frames NERD as short for 'No Effort Required, Done.'
- The project argues LLMs tokenize English words more efficiently than symbols and punctuation.
- NERD is described as terse and machine-optimized, intended to be human-observable but not human-friendly for editing.
- The implementation described is a bootstrap compiler written in C that compiles to LLVM IR and has no runtime or dependencies.
- The source claims token reductions in the 50–70% range versus conventional languages, with an example citing a 67% token reduction compared to TypeScript.
- Workflow proposed: a human states intent, an LLM writes or edits NERD, the code compiles to native, and humans review the results rather than authoring line-by-line.
- Objections addressed in the source include debugging and compliance, arguing debugging happens at the abstraction layer and translated views can support audits.
- The author characterizes the project as an experiment and admits it could be wrong about long-term adoption.
What to watch next
- Wider adoption of NERD or similar LLM-first languages beyond the experiment — not confirmed in the source
- Integration of NERD-style workflows into existing CI/CD and security/audit tooling — not confirmed in the source
- Regulatory and compliance acceptance of machine-authored source that is 'human-observable' but not written by humans — not confirmed in the source
Quick glossary
- LLM (Large Language Model): A machine learning model trained on large text corpora that can generate or transform natural language and code.
- Tokenization: The process of breaking input text into units (tokens) that a language model processes; tokenization affects model input size and cost.
- LLVM IR: An intermediate representation used by the LLVM compiler infrastructure to enable optimizations and generation of native machine code.
- Bootstrap compiler: A minimal initial compiler implementation used to compile or generate code for a new language, often written in an existing language.
- Auditable: Describes code or systems that can be inspected and verified for correctness, security, and compliance, even if not authored directly by humans.
Reader FAQ
Is NERD intended to be read and edited by humans?
The source says NERD is designed to be human-observable for auditing but not human-friendly for direct editing; humans act as stakeholders and reviewers.
Does NERD require a runtime or external dependencies?
According to the project, the bootstrap compiler emits LLVM IR and the language is designed with no runtime and no dependencies.
How much token savings does NERD claim?
The source cites token reductions in the 50–70% range and gives an example of roughly 67% fewer tokens versus TypeScript for equivalent logic.
Will NERD replace existing languages like TypeScript in production?
Not confirmed in the source
The Story The Question 40% of code is now written by LLMs. That number is growing. I was using Claude Code, watching it generate TypeScript. A thought hit me: Why…
Sources
- Nerd: A language for LLMs, not humans
- Why ML Needs a New Programming Language
- Programming Without People: Designing a Language for LLMs
- Code Llama, a state-of-the-art large language model for …
Related posts
- NERD: A Programming Language Built for AI Authors, Not Human Readers
- 2025 Year in LLMs: Reasoning, Agents, Coding Tools, and CLI Uptake
- PyPI in 2025: Year in Review — Security, Organizations, and Growth