Top 4
Agentic AI
Architecture
Design Patterns
Dipanjan (DJ)
Why AI Agents?
LLMs have increased performance using Agentic workflows
GPT-3.5 powered Agentic AI Systems achieved up to 95.1% on
HumanEval coding benchmark
Easy to connect LLMs, tools and prompts along with external data
to build simple and complex agentic workflows
How AI Agents work?
I. Start with the initial instruction prompt
II. LLM processes the prompt and decides what tool to call based
on the available tools (function calling)
III. The specific tool and tool arguments are then executed by the
system to get some new context information - in this case search
results
IV. This new information is passed along with the initial prompt to
the LLM to give a response or call more tools
Pattern 01 - Reflection
Reflection Pattern focuses on improving the AI agent’s ability to
evaluate and refine its own outputs
This self-critique loop of generating and reflecting is not limited to
a single iteration
System can repeat the reflection process as many times as
necessary to achieve a refined result
Self-Reflection RAG is a popular Agentic RAG system using this
pattern
Pattern 02 - Tool Use
Tool Use Pattern significantly broadens an LLM’s capability by
allowing it to interact with external tools and APIs
Agentic AI systems using this pattern can access databases,
search the web, or even execute complex functions via
programming languages like Python
Very useful to augment RAG systems with capabilties to answer
questions based on real-time searches
Most useful pattern to automate diverse tasks using a set of tools
Pattern 03 - Planning
Planning Pattern enables an LLM to break down large, complicated
tasks into smaller, more manageable components
Planning equips an agent with the ability to react to requests and
strategically structure the steps before execution
Create a roadmap of subtasks, determining the most efficient path
to completion
ReAct (Reasoning and Acting) and ReWOO (Reasoning With Open
Ontology) further extend this approach by integrating decision-
making and contextual reasoning into the planning process
Pattern 04 - Multi-Agent
Multi-Agent Pattern builds upon the concept of delegation, akin to
project management in human teams
Involves assigning different agents (which are instances of an LLM
with specific roles and tools) to handle various subtasks
Several types of multi-agent system patterns:
Collaborative Agents: Multiple agents work together on different parts of a task,
sharing progress and building toward a unified result
Supervised Agents: A central supervisor agent manages other agents,
coordinating their activities and verifying results to ensure quality
Hierarchical Teams: A structured system where higher-level agents oversee
lower-level agents
Detailed Article
Check out the
detailed article
here
