Vibe Kanban is a local web-based tool that can implement the operation, management, and result review of multiple AI coding agents (Claude Code, Gemini CLI, Amp, etc.) through a single Kanban-style interface. The tool allows AI agents to run in independent Git working trees, with the option of parallel or serial execution mode, real-time display of running progress and code differences, and quick functions to start development services and create pull requests, keeping your main codebase safe and organized at all times.
Core advantages: By uniformly orchestrating each AI coding agent and visually reviewing its code modifications, only verified results are merged into the main database, replacing the tedious operation of manually switching terminals and managing branches, greatly saving time and reducing operational errors.
Kanban Kanban tools have changed little in the past decade or so.
Whether it’s Trello, Jira, or Linear, they all default to one premise:
Tasks are created by people, processes are maintained by people, and tools are just record boards.
And vibe-kanban tries to overturn this premise.
It raises a radical but extremely AI-era question:
If AI is really involved in “doing things”, who should task management tools be designed for?
Designed for AI.
One “AI-first Kanban”
The goal of this project is not to make a “stronger Trello” but to explore a new paradigm of work.
With the advent of large models, agents, and automation tools, the way of working is changing:
- Many tasks are not clearly structured from the start
- Projects are often “thinking and doing, changing while doing”
- AI can already take on the role of disassembling tasks and performing subtasks
But the problem is:
We are still managing AI work with “tools designed for humans.”
It was in this contradiction that Vibe-Kanban was born.
Not “Mission-driven”, but “Vibe-driven”
The process of using traditional Kanban is:
- Think about what to do first
- Split the task again
- Fill in the card again
- Start enforcement again
And vibe-kanban, in turn:
You just need to describe “what I want to do”, even if it is vague.
For example:
“I want to make a gadget that translates PDFs to Chinese.”
The AI will do it automatically:
- Task disassembly
- Kanban card generation
- Status assignment (Todo/Doing/Done)
Here vibe refers to:
- Not precise
- Incomplete
- But the real expression of intention
This is very much in line with the way humans think in the creation, exploration, and research stages.
AI is the “first user” of Kanban
This is the most critical and counterintuitive point of vibe-kanban.
In this project:
- People are not the main “operators”
- AI is the one who really “uses Kanban”
AI can:
- Create a task
- Split and merge cards
- Adjust priorities
- Continuously modify the task structure as understanding changes
The human role became:
- Give direction
- Calibration AI
- Judge “do it right”
In other words:
Kanban is no longer “people show themselves”, but “a shared memory structure for people and AI to collaborate”.
Built for exploratory work, not process management
Vibe-Kanban is clearly not a tool designed for enterprise processes.
It is not suitable for:
- Projects with strong construction period and strong KPIs
- Large team collaboration
- Strict approval and responsibility tracking
But it’s perfect for:
- AI project prototypes
- Personal side project
- Research-oriented / creative work
- “Think and try” workflows
The biggest features of this type of work are:
You don’t know what’s next, but you know the direction.
And that’s where AI excels.
From a tool perspective
The real value of vibe-kanban is not in the functionality itself, but in the questions it raises:
- When AI can break down tasks, do humans still need to write to-dos?
- When AI can maintain processes, do they still matter?
- When AI is the doer, should management tools be “AI-friendly”?
It is more of a way of working experiment than a full-fledged product.
Epilogue
If traditional Kanban addresses:
“How people can better manage tasks”
That vibe-kanban really wants to ask:
“How do people manage AI when they are working?”
This question is probably more important than any feature update.
Github：https://github.com/BloopAI/vibe-kanban
Tubing: