Show HN: Vibe Kanban – Kanban board to manage your AI coding agents

Original Article ↗ Hacker News Discussion ↗

Overview & Perceived Value

  • Several users report strong productivity gains, likening it to their first experience with AI-first IDEs.
  • It’s seen as a “coding agent orchestrator in the shape of a Kanban board”: write tickets, run Claude Code/Gemini directly from cards, and review generated diffs/PRs.
  • Some have already forked and run it in locked-down environments; they find it competent but question whether it adds much beyond terminal + git worktrees for advanced users.

Workflow, Parallelism & Reliability

  • Pain points: dependent PRs, stacking changes, and needing to revise earlier cards; better support for stacked/linked work is requested.
  • Parallel agents editing the same checkout are a recurring concern; suggestions include separate git worktrees/branches per card. The project claims it already uses worktrees per attempt, but at least one user reported clobbering.
  • Users note “compounded false affirmatives”: agents add brittle fallbacks, tests pass, but issues are buried; models can help find these but must be manually verified.
  • Many are skeptical about running lots of agents in parallel today: quality is uneven, review load explodes, and domain difficulty matters.
  • Typical agent runtimes are cited as 2–5 minutes for small tasks, up to 15+ minutes or hours for complex builds/tests.

Positioning vs Other Tools

  • Compared to Backlog.md, this emphasizes tight, in-board interaction with coding agents and an MCP server that can auto-generate plans and tickets (“CTO agent” behavior).
  • Seen as potentially disruptive to traditional PM tools (Linear, Monday, ClickUp), though legacy tools are also moving toward AI workflows.
  • Alternatives mentioned: GitHub/GitLab issues + PRs, glab for GitLab, personal agent setups, and other experiments in AI+human Kanban.

Security, Permissions & Telemetry

  • Strong criticism of GitHub OAuth requesting broad access (including private repos and deploy keys); several argue a GitHub App with granular permissions is more appropriate.
  • Telemetry defaults-on sparked a long privacy debate: it was found to collect email, GitHub username, and detailed usage events.
  • Some see any opt‑out analytics as “spyware” and potentially illegal (GDPR/PIPEDA); others argue pseudonymized analytics are essential for product improvement.
  • The maintainers responded with a PR to make analytics clearly opt‑in, which was positively received.

“Vibe Coding” & Marketing Claims

  • The tagline that “AI coding agents are increasingly writing the world’s code” and humans mostly orchestrate is widely challenged as aspirational marketing.
  • Critics worry about glorifying being “abstracted from your own code,” especially for serious/enterprise systems (e.g., COBOL→Java migrations).
  • Concerns: more AI code means more security issues, technical debt, review overhead, and testing, while junior devs risk learning only to prompt models.
  • Supporters see this as a forward-looking bet: today they use agents for the easier half of the backlog; they expect the statement to become true soon and are exploring interfaces now.

UI/UX, Integrations & Feature Requests

  • Mixed feelings about Kanban as the long-term UI: good starting point, but many columns feel redundant when AI moves cards quickly; a new human–agent interaction paradigm may be needed.
  • Feature requests include: richer keyboard shortcuts, GitLab and Linear integration, GitHub App support, better documentation for the planning/auto-ticket features, and clear multi-user/hosted options.
  • Some are curious whether the workflow is sustainably satisfying over long periods or if it leads to loss of important “feel” for the code and project.