Using Agentic AI to create smarter solutions with multiple LLMs (step-by-step process)

The concept of "agentic AI" can be understood by imagining a team of specialized assistants working together on a task, like creating a marketing plan. Instead of one AI model trying to do everything, you have multiple "agents," each performing a specific part of the job. Here's how it works using the marketing plan example: Drafting Agent: You first give a prompt to an AI agent (which could be a Large Language Model or LLM) to create an initial draft of the marketing plan for a new product. This agent takes your instructions and background information and produces the first version. ( , ) Critique Agent: This initial draft is then passed to a second AI agent. This agent's role is to review the draft and provide a critique, pointing out areas for improvement or missing information. ( ) Refinement Agent: Finally, the original draft and the critique are sent to a third AI agent. This agent's task is to update the marketing plan based on the feedback from the critique, resulting in a more polished and comprehensive final version. ( ) This multi-step process, using different agents for different sub-tasks, generally produces a much better result than relying on a single AI to do everything. ( ) To fully grasp "agentic AI," the speaker suggests a few "mental leaps": LLMs as Agents: Start thinking of these LLMs not just as models, but as "agents," each responsible for a specific part of the task. ( , ) Agents Beyond LLMs: Recognize that not every "agent" in the workflow needs to be an LLM. Some agents can be simpler tools, like a Google search tool to find data, an API to schedule an appointment, or a calculator. These tools can assist the LLM agents. ( , ) Orchestrated Workflows: Imagine building complex workflows where one agent might act as an "orchestrator." This orchestrator agent could manage the overall process, deciding how many drafts are needed, when to seek external data using a tool, and when the final product is ready. ( ) Essentially, agentic AI is about creating systems where multiple AI agents and tools collaborate in a structured workflow to achieve a complex goal. ( ) Would you like to explore how these agents might communicate or be coordinated in more detail?