Kangentic: A Desktop Kanban Tool for Agent Workflow Orchestration

Kangentic is a desktop kanban application designed specifically for agent workflow orchestration. It combines traditional kanban methodology with AI agent automation, allowing agents to act as "participants" in the kanban system and enabling visual orchestration of human-machine collaboration. Its core value lies in addressing the limitations of existing agent workflow orchestration methods—code-first (unfriendly to non-technical users) and conversation-first (lack of controllability)—providing teams and individuals with an intuitive, controllable platform for managing agent workflows.

Definition of Agent Workflows

Agent workflows refer to sequences of multi-step tasks executed autonomously or semi-autonomously by AI agents, including decision-making, tool invocation, state management, error handling, and human-machine collaboration (e.g., the full process of content creation: research, writing, review).

Limitations of Existing Orchestration Methods

• Code-first: Writing code via frameworks like LangChain is flexible but has a high barrier for non-technical users, and the state is not intuitive;
• Conversation-first: Natural language interaction has a low barrier but poor controllability, making it hard to use for critical production tasks; Kangentic offers a third option: visual kanban orchestration, balancing ease of use and controllability.

Kanban as Workflow Abstraction

Each column represents a phase, and each card contains agent metadata:

• Assignee (human/agent);
• Automation rules (card movement triggers agent actions);
• Status feedback (execution logs/output display);
• Human-machine handover (agent transfers to human for review after completion).

Agent Roles and Permissions

Supports multi-role definition (research, writing, review, publishing, etc.), with role separation drawing on human team collaboration models to focus on specific tasks.

Workflow Template Reuse

Provides standardized templates (content creation pipelines, code review processes, etc.) that can be saved and reused to quickly establish consistent collaboration models.

Desktop-First Architecture

• Local agent support: Directly call local models to protect privacy;
• System integration: Integrate with local files/apps/services, allowing agents to operate local resources;
• Offline capability: Usable without network to ensure work continuity.

Agent Framework Integration

Supports frameworks like LangChain, LlamaIndex, AutoGen, CrewAI, with visual execution processes via standardized interfaces.

Real-Time Collaboration

The desktop app supports real-time synchronization for multiple users (using WebSocket technology), allowing team members to view real-time updates of the kanban.

Content Creation Teams

Process: Topic selection → Research (agent) → Writing (agent) → Human review → Optimization (agent) → Publishing (agent), with clear visibility of each phase's status.

Software Development Teams

Process: Requirement creation → Agent generates technical plan → Development implementation → Agent code review/testing → Deployment, integrating into agile processes to enhance automation.

Personal Productivity Management

Agents assist with email processing, schedule arrangement, document organization, learning planning, etc., serving as a personal orchestration center.

Value of Visualization

• Understandability: Users clearly understand agent behaviors;
• Debuggability: Quickly locate abnormal phases;
• Trustworthiness: Transparent execution builds trust;
• Optimizability: Identify process bottlenecks.

New Paradigm of Human-Machine Collaboration

Agents act as "digital colleagues", lowering technical barriers, establishing effective collaboration models, ensuring efficiency while guaranteeing quality and safety, and providing a human fallback mechanism.

Current Limitations

• Ecosystem maturity: Need to continuously follow up on agent framework compatibility;
• Complex workflow support: Kanban abstraction lacks flexibility for dynamic and complex processes;
• Learning curve: Requires users to understand basic concepts of agent workflows.

Future Directions

• AI-driven kanban optimization: Agents analyze data to suggest process improvements;
• Natural language interface: Lower configuration barriers;
• Cross-platform expansion: Mobile companion app;
• Community template market: Share and reuse workflow templates.

Kangentic is an attempt to integrate agent technology with traditional project management methodologies, emphasizing the importance of human understanding and control. It provides an intuitive and controllable orchestration tool for agent applications. As agent technology moves from experimentation to production, such tools will become key explorations in shaping the future of human-machine collaboration.