The Vibe Quality Model
From ad-hoc to professional vibe coding.
Your AI already knows how professional apps are built: tests, a second look at every change, security scans, a safe copy to try things on before they go live. But it does that work only when someone asks for it. Ask for an app, and you get one that runs, not one that was tested, reviewed and made safe. That is why so many AI-built apps break, leak data or can't be fixed later.
The Vibe Quality Model is a set of guardrails: rules your AI reads at the start of every session, telling it to work the way professional teams do and to keep the records that prove it. You keep building the way you do now.
Before AI, software teams already had ways to raise their quality one level at a time, by making good practice a habit and showing it with evidence. All of them assumed that people did the work. VibeQM is built for work done by AI, in five levels.
It's time to force AI to do things right the first time, and to keep improving all your projects over time with the lessons learned.
Available now: a free self-assessment for your project, npx vibeqm-check. Coming in November 2026: the full reference for the model.
Code is prompted, accepted and shipped when it runs.
AI autonomy earned None yet. Everything it writes is a draft that you read.
Requirements, tests, review, security scans and records on every change.
AI autonomy earned Edits code in its own working copy while you watch.
One way of working across all your projects, enforced and audited.
AI autonomy earned Takes a task to a finished, proposed change on its own.
Bugs, security and delivery tracked against limits.
AI autonomy earned Adds its own low-risk changes to the main code, while the numbers hold.
Every bug that slips through becomes a rule the AI must follow.
AI autonomy earned Adds most changes itself under a written agreement, with a kill switch.
The free self-assessment
One command. Your level, your gaps, and the minutes to close them.
A sample run on a small project, shortened. Free and open source (MIT), available now on npm. Read-only: it changes nothing in your project and sends nothing to us.
The ideas behind it
Professional practice, done by the AI.
Autonomy is earned
The AI gets more freedom only as the evidence of its quality grows, never on trust alone.
The AI keeps the records
You keep working the way you do now. The AI follows the practice and writes down the proof.
Evidence, not opinion
Every level is a short list of requirements, and each one is met by facts anyone can see in the project.
A person owns every change
However much the AI does, a named person answers for what ships, and can stop it.
Coming in November 2026
Free, for everyone.
The model
The full reference for all five levels and what each one takes, free to read and use.
The self-assessment
The tool above, free and open source and available now: run npx vibeqm-check. It can also lay down a ready-made starter set of files
that puts a new project on the road to Level 2.
Already open: the companion repository, github.com/vibeqm/vqm, with the model's framework, the templates for each level and the current setup for each coding agent.
The book · November 2026
AI can write a feature in minutes. Speed was never the hard part. Trust is.
The Vibe Quality Model is a quality system for software written with AI agents: what to ask of the agent at each level, the evidence that shows it was done, and how an agent earns more freedom instead of being handed it.
With Claude Code, Codex, Cursor and Copilot
From ad-hoc to professional.
November 2026