Is Open Sourcing Still Meaningful in the Vibe Coding Era?

Two months ago the author rediscovered "Kup", a small Node.js command‑line tool that publishes local Markdown files to GitHub Issues, and decided to test whether OpenAI Codex could resolve the project's long‑standing TODO items.

Within a few evenings the missing features were added, tests were written, and the project’s architecture was modernised (npm dependencies upgraded, migration to ESM, test framework added). The rapid progress prompted a deeper question: if the code is no longer written line‑by‑line by a human, does releasing the code still constitute genuine open source?

Kup’s core functionality is illustrated by the following commands:

# Install
npm install -g kup-cli

# Publish a new issue
kup ./post.md --repo cssmagic/kup

# Update an existing issue
kup ./post.md --repo cssmagic/kup --id 11

Using Codex, the author fed each backlog Issue to the model, allowing the AI to iteratively implement the requested feature. When Codex misunderstood a requirement, a brief correction brought it back on track. Before adding new features, Codex also performed an architecture upgrade: updating dependencies, handling compatibility, and migrating the module system to ESM, all within a test‑driven development (TDD) workflow.

The experience shifted the author’s focus from writing code to describing goals, defining boundaries, reviewing results, and correcting direction—essentially managing the AI’s output rather than hand‑coding.

The term "Vibe Coding" originates from Andrej Karpathy’s early‑2025 remarks about a new norm where developers describe intent in natural language and let AI generate, run, and debug code. In this paradigm, open source is no longer just about sharing code; it is about sharing the "human intellectual labor" that guides the AI—requirements, prompts, judgments, and error corrections.

To explore this, the author turned each TODO into a closed GitHub Issue that not only listed the functional requirement but also embedded the full conversation with Codex. The issue list (e.g.,

github.com/cssmagic/kup/issues?q=is%3Aissue+state%3Aclosed+label%3AFeature

) shows the transformation from a vague request to executable code, mirroring the step‑by‑step reasoning of solving a math problem where the answer is the code and the prompt history is the derivation.

This experiment demonstrates a hybrid open‑source model: publishing both the code and the AI‑driven development narrative, thereby exposing the decision‑making process behind the final artifact.

While the author does not claim this as the definitive open‑source approach for the Vibe Coding era, the experiment serves as a first step toward a more transparent sharing of AI‑augmented software creation.