x
    Glossary

    What is Agentic AI?

    Agentic AI refers to advanced AI systems that can autonomously analyze data, pursue complex goals, and adapt in real-time by solving multi-step problems with low to minimal human intervention. Simply stating, for a given environment (data ecosystem) they can understand the task, gather the data from the sources, generate solutions with a reasoning model like RAG/LLMs, and act on it via access to tools and APIs. It ideally consists of guard rails in the form of confirmations or approvals from humans; and a feedback loop.

    How does Agentic AI work?

    The capabilities of foundational models (which are typically used in Generative AI), have accelerated the growth and development of Agentic AI models. This makes generative AI and foundational models a great addition to the Agentic AI workflows, as they can assess the situation to brainstorm and create clear steps for innovative solutions. You think of LLMs like the brain of Agentic AI.

    Given that AI Agents are an emerging field, there are no hardcore frameworks for evaluating or developing them. In general, Agentic AI follows the structure:

    • Starts with a human command/ discussion
    • Clarifying the task and creating the execution flow
    • Execute each step of the plan by calling the required tools/environments
    • Pause or check point for human feedback
    • Task completion

    Though it might not be as simple as this, the idea is to have a Planner, Evaluator, and Executor which execute their own part and then flow their output into the next downstream activity like a multi-agent system.

    Examples of Agentic AI Models

    Some of the areas where people have started seeing automation results in Agentic AI models are:

  • Supply chain agents that can automatically adjust inventory levels based on real-time demand signals

  • Workforce planning agents that can dynamically schedule workforce

  • IT incident responses can be categorized into multiple workflows based on their priorities and incident types, and certain issues which can be addressed easily can be resolved quickly with AI

  • AI agents may identify inefficiencies in CI/CD processes, such as long build times or flaky tests, and suggest improvements

  • Customer Service queries can be routed to appropriate departments

  • As we’ve noted previously, agents work in their environments and call the tools/functions they want – so there’s a lot of scope for things that can be automated. All we have to do is explore.

    Agentic AI framework
    Sample Agentic AI framework

    P.S. In most cases, you might need automation or workflows but not agentic AI. So, think about the flow and execution before you want to get started with agents.