BPAX: Bridging Business Processes and AI Agent Workflows with JSON Standards

This post introduces BPAX (Business Process Agent eXchange), a JSON standard designed to address the understanding gap between processes designed by business personnel and automated systems implemented by technical teams, as well as the challenge of AI agents accurately comprehending and executing business processes. As a common language for human business planning and AI agent execution, it enables seamless conversion from business processes to agent workflows via standardized JSON structures, with advantages such as being lightweight, easy to understand, and composable.

In enterprise digital transformation, traditional Business Process Automation (BPA) faces information asymmetry between business personnel and developers: business experts are familiar with process logic but lack technical skills, while developers are proficient in programming but struggle to understand business scenarios, leading to project delays, budget overruns, or requirement deviations. With the rise of AI agent technology, a new challenge emerges: how to enable them to accurately understand and execute business processes.

BPAX is a lightweight JSON standard that follows the "convention over configuration" principle. It defines a standard JSON Schema covering elements such as step sequences, conditional branches, parallel execution, human intervention points, and tool call interfaces. Its structural design prioritizes readability and balances expressiveness with simplicity. Core components include metadata (process name, version, etc.), step definitions (ID, type, parameters), flow rules (sequence, conditional branches), tool bindings (external tool calls), and human tasks (scenarios like approvals).

BPAX implementation process: Business analysts sort out and optimize processes → describe them structurally using BPAX → AI agent execution engine parses and executes. The execution engine creates runtime instances, executes steps, evaluates conditions, calls tools, waits for human input, and persists states to support breakpoint resumption and auditing. Core advantages include composability (subprocess reuse) and observability (unified monitoring interface).

Compared with BPMN: BPAX is more lightweight, based on JSON (no need for specialized modeling tools), and suitable for version control. Compared with traditional workflow engines (e.g., Camunda): it focuses on AI agent scenarios and can run either integrated or independently. Compared with pure code implementations: it improves maintainability and business transparency, allowing business personnel to directly understand process definitions.

BPAX is applicable to scenarios such as customer service automation (ticket processing, escalation to humans), approval workflows (procurement/reimbursement/contract approval rules), data pipelines (ETL, data cleaning, report generation), and intelligent assistants (conference room booking, IT requests, etc.), helping to achieve process automation and intelligent execution.

Implementation recommendations: Start with small-scale pilots; establish BPAX document version management (Git + change logs); invest in execution engine robustness (error handling, state persistence, etc.); maintain close collaboration between business and technical teams (regular reviews).

Limitations: Immature ecosystem with limited execution engines and toolchains; less expressive for complex processes compared to mature BPM platforms; lack of consideration for operation and maintenance aspects (performance optimization, resource scheduling). Future outlook: As AI agent technology matures and standardization needs grow, we expect more tool support, standard extensions, and integration with traditional systems—BPAX may become a foundational standard for agent interoperability. Conclusion: BPAX provides a lightweight and complete starting point for AI automation, bridging business and technology to turn process planning into automated execution.