Agentic Workflow Implementations

When building AI applications involving multiple tasks implemented by different components (chains or pipelines), complexities can quickly grow as more components and integration points are added over time. Hence, applying an orchestration abstraction layer like Agentic Workflow can reduce development and maintenance burdens by keeping implementations easier to understand and debug.

In this writing, we will explore three different approaches to implementing Agentic Workflows, aiming to reduce development complexities.

A. Event Driven Approach (LlamaIndex Worflow)

B. General Approach (Langchain Agent)

C. Client/ Server Approach (Llama-Agent) — Create Agent that can deploy as micro-service

What is Agentic way?

One explanation from Andrew Ng was an Agentic following set of properties for building intelligent solutions :

Robust Technologies

• Reflection — self critique on response
• Tools use — get more details from additional source

Emerging Technologies

• Planning —systematic approach to accomplish the goal
• Multi-Agent Collaboration — enable communication with other Agents to complete task on hand.

Couple more things I would like to add the definition

• Decision making in the flow should in control by Human