WAF: A Text List Workflow Language for AI Agent Orchestration

WAF is a concise text list workflow language designed specifically for orchestrating AI agents, Shell commands, and human decisions. It aims to address pain points of traditional workflow solutions such as complex configuration, difficulty integrating AI agents, and insufficient support for human intervention, providing a lightweight DSL solution. Its core concept is to simplify workflows into plain text list form—intuitive and lightweight, similar to Makefile but optimized for the needs of the AI era.

Existing workflow orchestration solutions have pain points such as high configuration complexity (e.g., complex DSL/XML), difficulty integrating AI agents (failing to consider features like asynchronous responses), and insufficient support for human intervention. WAF sets the following goals: ultra-simple syntax (plain text list), AI-native design, native support for human-machine collaboration, and seamless integration with Shell commands and external tools.

WAF uses a flat text list syntax where each list item is a workflow step. It supports three types of executors: AI agents (for tasks like natural language processing), Shell commands (for integrating system tools), and human decision nodes (which pause and wait for input). It also has basic flow control capabilities (sequential execution, conditional branching, error handling).

Automated content production (AI topic selection and draft writing → human editing → Shell publishing); data analysis pipeline (Shell data retrieval → AI cleaning and extraction → human verification → automatic report generation); intelligent customer service (AI initial response → manual transfer → Shell ticket update); DevOps automation (AI code risk assessment → Shell build and test → human approval and release → AI status monitoring).

Technical advantages include high readability (understandable even by non-technical personnel), version control friendliness (suitable for Git), high portability (cross-platform), and good scalability (reserved expansion space). Comparison: vs Airflow (WAF is lighter and suitable for small projects); vs GitHub Actions (WAF is more general, not platform-bound, and has native AI orchestration); vs LangChain/LlamaIndex (WAF focuses on workflow orchestration and can complement them).

WAF is an open-source project using a permissive license to encourage community contributions. Future directions: visual editor, dedicated execution engine, template market, IDE support plugins. If the concept becomes popular, it may promote the development of workflow orchestration towards a concise and AI-friendly direction.

WAF demonstrates new possibilities for workflow orchestration through a concise text list form, maintaining tool simplicity and usability in today's complex AI application landscape. It is worth attention from developers seeking lightweight solutions or AI agent orchestration, and we look forward to it playing a greater role in the future.