The Year of the AI Agent: Understanding and Building Your Own

What is an AI Agent?

The year of the AI agent—that's what their calling it.

But what exactly is an AI agent, how does it differ from AI automations, and how can you start building them yourself? Well, when it comes to the definition of an AI agent, the waters are murky. But no fear, Anthropic, makers of Claude, came out with an article at the end of December (https://www.anthropic.com/research/building-effective-agents) that gives us a good place to start from:

At Anthropic, we categorize all these variations as agentic systems, but draw an important architectural distinction between workflows and agents:

• Workflows are systems where LLMs and tools are orchestrated through predefined code paths.

• Agents, on the other hand, are systems where LLMs dynamically direct their own processes and tool usage, maintaining control over how they accomplish tasks.

Understanding Through Real Examples

I love this breakdown. The key difference between AI workflow and AI agents is this idea of "dynamically directing their own processes". Now, what does that really mean? Well, lets look at an example.

Imagine you run your own business, and you have your own AI personal assistant that helps you send emails to potential leads. You generate leads by having people fill out a submission form on your website and, once the lead information enters your systems, its automatically routed to your AI assistant, and it creates a personalized email based on the submission data. This would be an AI workflow—the LLMs tasking is already predetermined (create an email) and it doesn't need to make any "decisions" about what tools to use. It's very linear and simple in design.

Now, imagine the same situation, but this time we want our personal assistant to do more than just turn lead data into emails—we want it to do a wide range of tasks—create calendar events, search the web, message others in slack, etc. To do that, we give this agent access to a number of tools (Google Calendar, Google Search, etc) and provide it an internal system prompt that instructs the agent to choose the correct tool for the job based on what we tell it. So, this agent is now no longer limited to one simple task and it can now execute anything we tell it, provided we supplied the agent with the correct tools. We don't even have to explicitly tell it to use one tool or another, it will know the best tool to call based on our conversation.

Building Your Own AI Agent

So now that we now what an AI agent is, how do we actually build one? Well, the answer to that depends on how technical you are. For those with a deep coding background, framework like LangGraph of PydanticAI give the user the most control and customization possible.

Low-Code Solutions

But for the rest of us who haven't been steeped in software development for the past decade, there is still hope. You have three main options when it comes to building AI agents in a low code environment, and ill quickly go over them here.

Make and Zapier are your first two options and they are extremely similar. Both use similar UIs that let you hook "modules" together to create easy to understand but powerful workflows that leverage almost any AI tool currently available. If you have zero AI building experience, this is where I would start.

Next up is n8n. Think of n8n with the training wheels off—you have more room to customize your modules but you still get to operate in a low code environment which caters to the non-technical user. N8n is my personal favorite in this space—their design promotes an extremely flexible AI agent building approach that lets you mix and match tools between separate agent flows you've created. Again, this is slightly more technical than Make or Zapier, but it's the place you should be striving to get to you as you enter this arena.

If you're ready to get in there and start building, make sure to check out my skool community (https://www.skool.com/chase-ai) where I go through how to build agents, step by step, with zero code!