A new VS Code extension called Agent Kanban launched March 8, 2026, addressing a persistent problem in AI-assisted development: losing track of planning conversations and decisions when chat sessions end or context limits are exceeded. Created by UK-based full-stack engineer Gareth Brown, the tool gained 79 points and 36 comments on Hacker News within hours of its debut.
Markdown-Based Task Management Prevents Context Loss
Agent Kanban stores every task as a markdown file with YAML frontmatter in a .agentkanban/tasks/ folder, creating what Brown calls "a permanent, editable source of truth." This approach means that when developers return to a project weeks later or hit context limits in their AI chat sessions, the entire history of considerations, decisions, and actions remains accessible and version-controlled.
The extension provides a visual Kanban board directly in the VS Code Activity Bar with configurable lanes (defaulting to Todo, Doing, and Done). Rather than bundling its own AI capabilities, Agent Kanban integrates with existing tools like GitHub Copilot Chat through a @kanban chat participant that injects task context into standard Copilot sessions.
Structured Workflow Guides AI Agent Collaboration
The extension introduces simple commands—plan, todo, and implement—that guide agent work through progressive stages. As Brown explained in the announcement, "A week later you come back to it, or you hit the context limit and clear chat, and that entire history is gone." Agent Kanban solves this by preserving full conversation history within each task file.
The tool's four main features include GitOps and team-friendly Kanban board integration inside VS Code, structured plan/todo/implement workflows via @kanban commands, leverage of existing agent infrastructure rather than bundling a built-in AI harness, and markdown task format resistant to context rot.
Community Response Highlights Real-World Pain Points
Developer discussions on the Hacker News thread revealed both enthusiasm and cautionary tales. One developer noted that "agent kanban is the piece most people skip. they build the agents but have no visibility into what's actually running." Another shared their own experience: "I came to exactly the same conclusion a while back and built the agent Kanban a few weeks back... truth be told - it caused more issues than I wanted and I had to strip it right back - the agents kept getting stuck in loops."
The extension is available now on the VS Code Marketplace and GitHub under an open-source license, allowing teams to integrate AI-assisted development workflows without losing the institutional knowledge that typically disappears when chat contexts are cleared.
Key Takeaways
- VS Code Agent Kanban launched March 8, 2026, created by UK engineer Gareth Brown, gaining 79 points on Hacker News
- The extension solves "context rot" by storing tasks as version-controlled markdown files with complete conversation history
- Integration with GitHub Copilot via @kanban chat participant preserves context across sessions without bundling custom AI
- Tasks are stored in
.agentkanban/tasks/folder with YAML frontmatter, creating auditable source of truth for team collaboration - Early community feedback highlights both the need for agent orchestration visibility and potential challenges with agent loops