TL;DR
A developer released whenwords, a relative-time formatting library that contains no implementation code—only a spec, language-agnostic tests, and install instructions. The project is a thought experiment about what software looks like when coding agents can produce working implementations on demand.
What happened
The author published whenwords, a utility for converting timestamps to human-readable relative times that ships with no source code. Instead of implementations, the repository contains SPEC.md (behavior and implementation guidance), tests.yaml (125 language-agnostic input/output test cases), and INSTALL.md (a short prompt that tells a coding agent to implement the five required functions and run the tests). whenwords defines five core functions — timeago, duration, parse_duration, human_date, and date_range — and the author reports successfully generating working implementations across Ruby, Python, Rust, Elixir, Swift, PHP, and Bash, plus an Excel formula variant. The release is framed as an experiment in spec-driven development enabled by recent improvements in coding agents (the post cites tools like Opus 4.5, Claude Code, Codex and others), and it raises questions about when binary code libraries remain necessary versus when precise specs plus generative agents suffice.
Why it matters
- Demonstrates how expressive specifications plus capable code-generation agents can produce cross-language implementations without a canonical codebase.
- Highlights trade-offs between on-demand implementation and traditional reference implementations for performance-sensitive or complex systems.
- Exposes testing and maintenance challenges as specs expand to cover more languages and agent behaviors.
- Raises questions about responsibility for bug fixes, security updates, and long-term support when generated code varies by agent or environment.
Key facts
- whenwords contains no implementation source code; it provides SPEC.md, tests.yaml, and INSTALL.md.
- The project defines five functions: timeago, duration, parse_duration, human_date, and date_range.
- tests.yaml contains 125 test cases that any implementation must pass according to the repository.
- The author reports working implementations in Ruby, Python, Rust, Elixir, Swift, PHP, and Bash, and an Excel formula variant.
- Installation is described as a brief prompt to paste into coding agents such as Claude, Codex, Cursor or similar tools.
- The post references recent model improvements and names Opus 4.5 and Claude Code among tools used to generate implementations.
- The author contrasts whenwords’ 125 tests with SQLite’s 51,445 tests to illustrate differences in testing surface area.
What to watch next
- Progress and reliability of coding agents (the post highlights recent gains in tools like Opus 4.5 and Claude Code).
- Whether a spec-only approach scales to larger, performance-sensitive projects: not confirmed in the source.
- How multi-language, multi-agent CI and testing strategies evolve to validate spec-generated implementations: not confirmed in the source.
- If vibrant community-maintained reference implementations continue to be necessary for security, support, and interoperability: not confirmed in the source.
Quick glossary
- Specification (spec): A written description of expected behavior and interfaces that an implementation must follow.
- Test suite: A collection of input/output cases used to verify that an implementation meets the spec.
- Coding agent: A machine learning model or system that generates code from prompts or high-level instructions.
- Unix time: A widely used timestamp standard counting seconds elapsed since January 1, 1970 (UTC).
Reader FAQ
What is whenwords?
A relative-time formatting library published as a specification and tests rather than language-specific code.
Which languages are supported out of the box?
The author reports tested implementations for Ruby, Python, Rust, Elixir, Swift, PHP, and Bash, plus an Excel formula variant.
Does the repository include runnable code?
No — it contains SPEC.md, tests.yaml, and INSTALL.md but no implementation source files.
Is whenwords tied to a single AI model?
The install instructions suggest using various coding agents (the post mentions Claude, Codex, Cursor, Opus 4.5), not a single model.
Is whenwords ready for production use?
not confirmed in the source
Jan 8, 2026 OSS AI SPECS A Software Library with No Code All You Need is Specs? Today I’m releasing whenwords, a relative time formatting library that contains no code….
Sources
- This is the future: A Software Library with No Code
- c++ – Writing a language agnostic API?
- Language Agnostic Programming: Why you may still need …
- Natural Language Is Now the Only No-Code Tool That …
Related posts
- Indonesia Temporarily Blocks Grok amid Surge of Sexualized Deepfakes
- AI Econ Seminar: A Simulated AI Economist Faces a Hostile Faculty Panel
- How the High School You Attend Reshapes Your Chances for UC Admission