Agentic AI
What is Agentic AI? Agentic AI is a framework where multiple AI agents can work independently to achieve a goal. Instead of performing just one task, these agents can handle multiple tasks, make decisions, and even collaborate. The most interesting aspect of agentic frameworks is that agents can communicate with each other to achieve better results.
Why Is Agentic AI Important?
Tasks that we want AI to perform are complex, have many steps, need various applications, and often involve many people. Unlike traditional AI, designed to handle a single, specific task, agentic AI systems allow multiple agents to collaborate, adapt, and solve problems autonomously. You can often see the execution logs of the agents. This is great for transparency and helps you to monitor the process and fine-tune your workflow. Also, you can set a human-in-the-loop step into the workflow, if you need to review the work before it is completed.
Here's why this is a big deal:
Example Workflow
Let’s examine a simple agentic workflow. I have created three agents in Python: A user (me) enters a topic, a Researcher gathers information, a Writer writes a LinkedIn post, and a Manager reviews and approves the LinkedIn post.
Please note that this is a simplistic workflow for demo purposes only. The possibilities are limitless and depend on the user's needs and creativity. You could build an agentic workflow to create games, analyze stocks, and write legal documents, and much more.
Step 1: Create the Agents
Each agent has a role, goal, tools, memory, backstory, etc. In this case, I have added a Researcher agent, whose job is to find good data on a given topic. This agent uses a web search tool to access the latest information on the topic. The Researcher agent has a memory, therefore it remembers previous outputs, instructions, or inputs and can use them to improve or refine its current task. The backstory is like a system message in LLM, it provides context, and personality helping the agent to define how it should behave and make decisions.
How many agents can you create in one workflow? As many as you like. They can also use different LLMs and custom tools. Each agent and step in the workflow increases response time and cost. This is why LLMs with very low latency and cost will be the preferred choice for agentic frameworks
Step 2: Define the Tasks
Each agent has a specific task assigned. For example, the researcher gathers data, the writer drafts a post, and the manager reviews it.
Step 3: Set Up the Workflow
Here is an example of the workflow showing agents, tasks, and the process.
Transparency:
Once the process has been kicked off, you can follow the steps your agents are doing as they happen.
In the below example, the search topic for was ‘Antti Karjalainen WilsongHCG’, and the task was to create a LinkedIn post.
Once the Researcher provided the final response, the Writer decided that it needed more details and delegated the work to the Researcher.
More thoughts. The agent looks at the results and decides to refine the search.
The agent decides to try something else.
The Manager reviews the Writer’s post and asks the agent to verify a few items:
Here is the final output:
Final Thoughts
Agentic AI lets you create flexible agent workflows that can tackle complete workflows and complex tasks in your organization. The possibilities are endless, and you are going to see more and more applications using this type of framework to make processes easier and more efficient.