Taxonomy of Workflow Automation

A workflow comprises the interdependent processes and people required to reach a result that no single participant can achieve alone [1].

Here is the taxonomy of workflow automation methods, incorporating Agentic Automation as a key evolutionary step.

The classification of workflow automation has evolved from simple task-based scripts to autonomous systems capable of reasoning. This taxonomy categorizes methods based on their technical engine, level of autonomy, and interaction model.

1. Categorization by Technical Methodology

This identifies the underlying “engine” driving the automation.

• Robotic Process Automation (RPA): Mimics human UI actions (clicks, typing) to handle repetitive tasks in legacy systems that lack APIs [2].
• API-based Integration (iPaaS): Facilitates direct data exchange between modern cloud applications using structured interfaces [3].
• Agentic Automation: Leverages Large Language Models (LLMs) as a reasoning core to autonomously plan, select tools, and execute complex sequences [4].

2. Categorization by Decision Autonomy

This distinguishes how the system handles logic and unforeseen variables.

• Deterministic Automation (결정론적 자동화): Follows rigid, predefined $If-Then$ paths; any deviation from the script causes the process to fail [2].
• Cognitive Automation (인지적 자동화): Uses specialized AI (like OCR or sentiment analysis) to transform unstructured data into structured inputs for fixed workflows [5].
• Autonomous (Agentic) Automation (자율적 자동화): Does not rely on a fixed flowchart; it dynamically generates steps based on the goal and can self-correct when encountering errors [4].

3. Categorization by Human Interaction (Engagement Model)

This defines the relationship between the human operator and the automation.

• Attended Automation (유인 자동화): Operates on a user’s local machine, triggered by specific human actions to assist with real-time tasks [2].
• Unattended Automation: (무인 자동화) Runs independently on background servers or the cloud, typically for high-volume, scheduled batch processing [3].
• Human-in-the-Loop (HITL): Common in Agentic Automation, where the agent handles the “heavy lifting” but pauses to seek human approval for high-stakes decisions [4].

4. Categorization by Architecture Scope

This looks at the breadth of the business impact.

• Task Automation: Automates a single, discrete action (e.g., saving an email attachment).
• Process Automation (Business Process Automation, BPA): Orchestrates a series of connected tasks to complete a functional business process like employee onboarding [5].
• Orchestration Layer: Where Agentic Automation sits, acting as a “manager” that coordinates multiple APIs, RPA bots, and databases to solve open-ended problems [4].

References

1. What is a workflow?
2. IBM – What is Robotic Process Automation (RPA)?
3. Gartner – Integration Platform as a Service (iPaaS)
4. Microsoft – Directing AI Agents with Agentic Workflows
5. Salesforce – What is Business Process Automation (BPA)?

---

Gemini