Vibe Coded PM

"Vibe Coded": A Self-Deprecating Name That Tells You Everything

JordanDalton posted a Show HN titled "Vibe Coded Project Management System" — 8 points, 5 comments. No "redefining collaboration" pitch, no "AI-powered intelligent boards" buzzword soup. The name itself is the statement: I coded by vibes, it runs, deal with it.

The project is called PM, open source on GitHub, at v0.0.1 with 12 total commits. The author didn't even bother writing a repo description or adding screenshots — and the HN crowd let him have it.

andrewfromx asked: "screen shots? or video? i have to clone the repo and run it to see?"

bradac56 went harder: "You need to at least put a description of what your code is/does in the github page preferably with screenshots otherwise it's going to get marked as a phishing scam/attack and rightfully so. There is no way I would ever run anything blind."

The author scrambled to add a YouTube livestream link and updated the README after getting called out. But honestly? That sequence of events is the most "vibe coded" thing about the whole project — ship first, document later (or never).

The Technical Bits: Small but Not Sloppy

Digging into the repo: two core models — Board and Task — with Laravel Sanctum handling API auth, factories, and migrations included. Board has a one-to-many relationship with Task. The frontend is either nonexistent or not yet public.

Here's what's notable: despite being just 12 commits, the data model relationships were correct from day one. Too many v0.0.1 projects are Controller-to-database spaghetti with tangled routes. This one had Board→Task relationships, API token management, and proper route configuration wired up during the bootstrap phase. The author clearly had the data model mapped out mentally. "Vibe coded" reads more like self-deprecation than actual recklessness.

Best guess at the stack: Laravel + Sanctum + MySQL. For a solo dev, this combo is a productivity weapon — Sanctum saves you from writing JWT auth from scratch.

What Role Does AI Actually Play Here?

The naming convention — "Vibe Coded" — strongly implies significant AI-assisted code generation. The HN discussion naturally branched into this topic.

Hnrobert42 shared a sharp observation: "It's my experience that Claude, Copilot, etc is good for green field projects. It's less helpful for existing projects with lots of modules. For those, it is really useful to write unit tests. Well, to get started on them anyhow."

That tracks perfectly. Vibe Coded PM is the textbook green field project: simple structure, few modules, clean models. In this scenario, AI genuinely helps you scaffold migrations, generate factories, and wire up relationships fast. But if you asked the same AI tools to refactor a three-year-old codebase with tangled dependencies? Good luck.

The practical takeaway: AI coding assistants are best deployed at the very start of a project, not as a band-aid for legacy code. This project is a clean demonstration of that principle.

Who Is This Actually For?

Let's be real — nobody can use this as a project management tool right now. You have to clone it and run it yourself, there are no screenshots (still), and the feature set is essentially "two models with an API."

But the value isn't the product. It's the demonstration: one solo dev, using Laravel + AI assistance, built a project management skeleton with full API auth, data models, and proper relationships in roughly a dozen commits. This isn't a technical marvel, but it's a solid "just ship it" case study.

If you want to learn how to scaffold a SaaS app quickly with Laravel, or you're curious what AI-assisted coding looks like in practice (not in a marketing demo), this repo is worth browsing. If you want an actual project management tool, go back to Linear.

Limitations: It's Very Early

v0.0.1. No frontend (or none shared). No README description until the community complained. Nearly flagged as a phishing link by HN commenters. The biggest problem here isn't technical — it's communication. The author skipped the most basic documentation, and the community's first instinct was "is this malware?" For open source, your README is your first impression, not an afterthought.

There's also the "vibe coded" quality question. AI-generated code works fine for simple scaffolding, but once business logic gets complex, those "this feels right" code paths can hide real bugs. At 12 commits, there's nothing obviously wrong, but the real test comes when the project grows beyond two models and an auth layer. The author will need to shift from vibing to actually reviewing what the AI produced — and that transition is where most "vibe coded" projects either mature or collapse.