Agent Kanban: Persistent AI Task Management for VS Code
Article: Very PositiveCommunity: PositiveMixed
VS Code Agent Kanban is an extension that manages AI coding tasks through version-controlled Markdown files to prevent context loss. It leverages GitHub Copilot Chat to provide a structured workflow for planning and implementing features directly within the IDE. By treating AI interactions as repository assets, it ensures that development history remains accessible and team-friendly.
Key Points
- Solves context rot by storing AI task history and decisions in persistent, version-controlled Markdown files.
- Integrates with existing tools like GitHub Copilot Chat rather than creating a proprietary agent harness.
- Follows a structured GitOps-friendly workflow using plan, todo, and implement commands to drive the AI.
- Provides a visual Kanban board inside VS Code for a bird's-eye view of AI-assisted tasks.
- Ensures team transparency and audit trails by keeping AI conversations and plans within the repository.
Sentiment
Hacker News broadly agrees with the premise and appreciates the pragmatic implementation. The community overwhelmingly validated the context rot problem, with many commenters sharing similar workflows they had independently developed. The reception was warm and constructive, with mild skepticism around tool necessity and branding accuracy.
In Agreement
- Context rot is a genuine and widely-experienced problem — AI agents lose track of earlier architectural decisions and planning conversations when sessions are cleared
- Storing plans and reasoning as markdown in git is an elegant, git-native solution providing version control, team sharing, and persistence across sessions
- The plan-before-code workflow is essential; forcing the agent to produce a written plan first prevents token-wasting and keeps humans in architectural control
- Leveraging existing IDE tooling (GitHub Copilot) rather than building a custom harness is pragmatic and preserves valuable IDE integrations like diff visualization
- Markdown as a task database is a compelling pattern — simple, editable, diffable, with no hidden state or external dependencies
Opposed
- Existing tools like GitHub Issues or Jira kanban with CLI access serve similar purposes — a dedicated VS Code tool may be unnecessary overhead for teams already using these
- The tool is technically a task board, not a true Kanban board (which requires WIP limits and flow metrics) — the branding creates misleading expectations
- The workflow is VS Code and GitHub Copilot-specific, limiting adoption for developers using other editors or AI assistants
- Concerns about AI-generated comments polluting the discussion, reducing signal quality and community trust in the thread