Agentic Workflows: Reusable AI Agent Architecture and Business Process Analysis Framework

The Agentic Workflows project by SMIT-Club is an AI agent architecture and workflow framework focused on business analysis, decision support, and problem decomposition. Its core features include:

1. Agent-First Design: Define runtime behavior through agent specifications rather than static workflow files, enhancing flexibility;
2. PSD Pipeline: A structured multi-stage problem decomposition process that supports systematic analysis of complex tasks;
3. Reusable Components: Provide agent specifications, skill templates, and workflow patterns to facilitate rapid application by teams;
4. Multi-Agent Collaboration: Define clear agent roles to enable end-to-end business analysis support. This thread will introduce the project background, architecture, core processes, and application value in detail across different floors.

Background Challenges

With the improvement of large language model capabilities, AI agents have evolved from Q&A tools to autonomous systems for complex tasks. However, multi-agent collaboration, output consistency, and traceability have become key challenges in practical applications.

Core Concepts

The project adopts an Agent-First design: runtime behavior is defined by agent specifications rather than traditional static workflow files. This design allows the system to dynamically adjust execution strategies based on tasks, enhancing flexibility and adaptability.

The project aims to provide reusable agent specifications and workflow patterns to help teams implement systematic AI-assisted analysis in various fields.

Layered Architecture

The project uses a clear layered architecture, with resources organized by function:

1. Agent Specification Layer: The core module, containing agent behavior definition files that clarify responsibility boundaries and input-output contracts;
2. Skill and Prompt Layer: Reusable procedural capability modules (skills) and task instruction templates (prompts);
3. Documentation and Guide Layer: Includes detailed documents such as agent directories, taxonomies, and PSD pipeline guides;
4. Workspace Layer: A structured directory (inputs/outputs/templates/examples) to help new users get started quickly.

Core Agent Roles

Four key business analysis agents are defined:

• Requirements Analyst: Extracts/clarifies requirements and defines acceptance criteria;
• Stakeholder Impact Analyst: Identifies stakeholders and assesses the impact of changes;
• Process Mapper: Maps existing/target processes and identifies bottlenecks;
• BA Review Expert: Verifies the effectiveness, traceability, and completeness of deliverables.

PSD (Problem Statement Decomposition) Pipeline is the core workflow of the project, which decomposes complex problems through 6 structured stages:

1. Stage A - Normalizer: Cleans the original problem and outputs a uniformly formatted description;
2. Stage B - Extractor: Extracts key elements such as constraints and success criteria;
3. Stage C - Classifier: Classifies problem types and priorities;
4. Stage D - Auditor: Reviews logical consistency and information completeness;
5. Stage E - Packager: Organizes results into a standardized format;
6. Stage F - Excel Formatter: Generates Excel files for easy use by business users.

Orchestrator Functions

The PSD Pipeline is coordinated by a dedicated orchestrator:

• Manages the execution order of stages and data transfer;
• Supports breakpoint recovery (resumes from the point of interruption);
• Ensures that outputs between stages meet handover requirements.

The project is suitable for three types of scenarios:

1. Business Analysis Projects: Requirement clarification, stakeholder impact assessment, process optimization, and deliverable inspection;
2. Decision Support: Decomposing ambiguous problems, identifying decision constraints and success criteria, and evaluating the pros and cons of options;
3. Project Initiation Planning: Quickly understanding scope, identifying stakeholder expectations, and establishing requirement baselines and acceptance criteria.

Each scenario can be implemented with systematic analysis through corresponding agents or the PSD Pipeline.

The project brings four paradigm shifts:

1. From Tool to Partner: Agents collaborate proactively, understand context, and make autonomous decisions;
2. From Single Point to Process: Achieve complex end-to-end tasks through workflow orchestration;
3. From Technology to Business: Focus on business analysis scenarios and verify the value of agents in the knowledge work field;
4. From Experiment to Engineering: Transform agent prototypes into engineerable solutions through specifications, documentation, and contribution processes.

Summary

Agentic Workflows is a model project of AI agents in the field of business analysis. It solves multi-agent collaboration challenges through systematic methods and provides reusable architectures and patterns.

Reference Value

It provides organizations with:

• Methods for defining agent responsibility boundaries and interaction contracts;
• Design ideas for recoverable and traceable workflows;
• Code and document organization specifications for agent projects;
• Collaborative contribution processes and review standards.

Future Outlook

As agent technology matures, more reusable business components will emerge, accelerating the penetration of AI in the knowledge work field. For business analysts and project managers, agents will become important partners in complex task collaboration.