LLM’s are part of everyday life
With the rise of LLM’s over the past decade there has been surge in development/usage of textual chat applications. Chatbots using these LLM models are becoming ineradicable parts of human life.
Considering this organizations are trying to build bots, softwares tailoring the user experience for their businesses. Simply put, Agentic Systems are just LLM’s on loop, and they represent the next step in this evolution. Instead of simply responding to prompts, they can plan, reason, make decisions, use tools, and execute multi-step workflows to achieve a goal.
For eg. You ask gemini to add an event on your calendar.
An Agentic System
As defined by Google
Agentic AI is an advanced form of artificial intelligence focused on autonomous decision-making and action. Unlike traditional AI, which primarily responds to commands or analyzes data, agentic AI can set goals, plan, and execute tasks with minimal human intervention. This emerging technology has the potential to revolutionize various industries by automating complex processes and optimizing workflows.
Agentic system is an AI system that is focused on orchestration and operation of agents to perform actions by using LLMs as underlying technology. To simplify, an agentic system is nothing but a real world organization where each worker (Agent) performs a specific goal contributing towards the broader organizational goal.
As the AI systems are evolving there is need of bringing complex workflows within AI systems to cater varied customer use-cases. These workflows involve multiple actions and decisions which traditional LLM’s are not capable of performing. This leads to the introduction of agents which is an
autonomous entity designed to perform specific tasks
An agent is a standalone system which uses LLM at its core and is tailored to → understand complex user objectives → break it down in small and precise tasks → perform analysis → plan execution → perform the tasks with provided context and tools → reach the desired objective by completing all sub tasks.
Working of Agents
An agent works as an individual worker or part of multiple workers in the system. An individual agent is designed to perform a single task with higher efficiency and correctness. An agent has following components:
Goal → Reasoning Engine → Context → Tools → Feedback
- Goal : Goal can be defined as a objective which needs to be achieved.
- Reasoning Engine : A reasoning engine is a LLM which is called using a special prompt to get the specific task done.
- Context : Every agent has context of its own which is related to the task it’s supposed to achieve.
- Tools : Tools are external integrations that helps the agent to complete it’s task.
- Feedback : With every objective an agentic systems learns and becomes more accurate in future.
Every agentic system starts with a base agent (In Single Agent System) that starts the initial flow of the task. A general use of base agent is to understand the user prompt and generate the plan to achieve desired goal.
In case of multi agent systems a base agent is responsible to break user query down into completable smaller tasks or goals. It acts as a orchestrator for the entire multi-agent system. Once the tasks are decided, a base agent calls respective agents and assigns them sub tasks along with input if any. Others agents may be single or again a part of multi agent system. The worker agents complete the sub tasks either by reasoning, calling tools, interacting with external systems or even fetching data from various sources. The base agent then combines the outcomes of sub-agents in calling order → and synthesizes the final output by transforming the results.
Agentic System In Action
A real world use-case of Agentic System is “Github Copilot Agent Mode” which is available today in most of the IDE’s. It is the next evolution in AI assisted software development. It can achieve multi step tasks including reading files, analyzing the code, performing code edits, running cli commands and many more.
It can be used for:
- Researching the architecture
- Fixing Bugs
- Improving test coverage and code quality
- Updating Documentation
- Resolving code errors
Details on how it works can be found here -
introducing-copilot-agent-mode#_how-it-works
Here is a sample working agent code in Java built with ADK - https://github.com/Deepcodr/JobPilotAgent
Beyond the prompt
Agentic Systems represent a significant shift in AI development. Rather than simply responding to prompts, it enables AI to tackle complex, multi-step tasks that traditional applications struggle to handle.
For developers, understanding agentic systems is becoming increasingly important as organizations move toward autonomous workflows and AI-driven operations. The future of AI is not just about generating answers; it's about taking meaningful actions to achieve outcomes.
Top comments (0)