Home GenAI Skills About Us

Agentic AI

What is a "AI Agent" or "Agentic AI"? It can sound scary. But it is important for you to understand what is Agentic AI, along with its capabilities and limitations. The reason being, AI Agents will be everywhere and that is the way we are going to be working with software infused with AI. In the most broadest sense, an AI agent is a software program that can complete a task using LLM, with minimal to zero supervision. Another way to put it is :"An AI Agent is a type of software program, that uses AI(LLM's). It can run autonomously, reason, plan and make decisions to complete a goal." The AI agent can be triggered by a human, or by an event or by another agent. If it needs input or cannot complete a task it reaches out to a human for input or guidance. Here are a few examples:

  • A travel AI Agent creates a custom itinerary for a group of traveler's having mobility issues, translates it to multiple languages and sends it to each member of the travel group.
  • A finance AI Agent, searches the web for all news related to a portfolio of stocks , summarizes it and sends it to finance analyst.
AI Agents

Unlike traditional software systems that follow predefined rules, agentic AI can adapt to different scenarios, analyze and reason based on context to complete a task. As seen in the diagram, AI agents use LLMs to reason, planand solve complex multi-step problems. They may also use LLMs to execute the various steps in the process. AI agents reason based on the data and context allowing them to handle tasks that previously required human judgment. Below are some of the properties of AI Agents

  • Goals: What task can the Agent do? This is important, can serve as a guardrail for the Agent so that it does not go outside it's boundary
  • Channels: Where do you want to interact with the agent? Web, Mobile, Phone, WhatsApp, etc.
  • Data: Agents have access to "just" the data they need to complete their job
  • Context: Knowledge bases and memory (long term & short term) to make intelligent and personalized decisions
  • Action: An AI agent can take action (like calling APIs or updating databases) to complete the goal
  • Safety and Guardrails: Define guidelines and boundaries of what agent can do or not do

AI Agents will come in all shapes and forms. Specialized AI agents which can complete one task. Or complex AI agents which can call upon other Agents to complete a goal. Most SaaS vendors like Salesforce, ServiceNow, Databricks,etc have already introduced Agents into their offerings.

Resources