Nagare: A Human-Centered, Evidence-First AI Workflow Kanban System

Nagare is an AI workflow kanban system developed by hachiware-labs. Its core is the evidence-first design philosophy, aiming to build workflows that support human-AI collaboration and achieve seamless integration between humans and AI. The project addresses the shortcomings of traditional tools in human-AI hybrid collaboration, improving work transparency, traceability, and collaboration efficiency. It is applicable to multiple scenarios such as content creation and software development.

With the rapid development of AI, human-AI hybrid collaboration has become the norm (e.g., AI generates drafts and humans review them). Traditional tools cannot track human and AI work or record decision-making basis. Nagare emerged as a solution; its name comes from the Japanese word "流" (nagare), symbolizing smooth work and natural connection between humans and AI.

1. Traceable Decisions: Task changes and decisions require clear evidence (AI reports, data results, etc.); 2. Separation of Human and AI Contributions: Distinguish between human and AI contributions to support quality improvement; 3. Verifiable Progress: Automatically update status by linking to actual outputs, avoiding manual update issues.

• Kanban View: Multi-dimensional classification, real-time synchronization, evidence display, human-AI identification;
• AI Agent Integration: Registration management, task assignment, result verification, feedback learning;
• Evidence Management: Collection, association, display, archiving;
• Workflow Orchestration: Conditional branching, human-AI handover, parallel processing, exception handling.

1. Content Creation: AI draft → Evidence collection → Human review → Iterative optimization → Quality verification → Publication;
2. Software Development: Requirement analysis → Code generation → Review → Testing → Deployment decision;
3. Data Analysis: Data exploration → Hypothesis proposal → Verification → Interpretation → Report generation;
4. Customer Service: Ticket classification → Knowledge retrieval → Response generation → Review → Satisfaction tracking.

• Tool Integration: Integrates with Git, CI/CD (Jenkins), document systems (Notion), and communication tools (Slack);
• Data Model: Designed around tasks, evidence, participants, workflows, and decisions;
• Extensibility: Plugin architecture, API-first approach, multi-tenant support.

• Evidence Collection Completeness: Relies on tool integration; improvement directions include flexible import, manual supplementation, and unstructured data processing;
• AI Agent Generality: Requires targeted configuration; improvement directions include standardized interfaces, automatic matching, and benchmark testing;
• Learning Curve: Team adaptation cost; improvement directions include template guides, gradual introduction, and training materials.

Nagare proposes a new paradigm for workflow management in the AI era. By prioritizing evidence, it addresses transparency and traceability issues in human-AI collaboration and provides a data foundation for continuous AI improvement. In the future, such tools will become infrastructure for the co-evolution of humans and AI, offering a reference framework for AI-assisted teams.