Cutting Through the Haze of AI Agent Hype

By now we all know that the AI Hype Engine is running at full tilt 24/7 and is probably about to overheat (or maybe it’s already overheated and blown up…). The last couple weeks have been no exception, and the focus of the hype has been AI agents, largely due to the release of Manus, a Chinese AI model that “makes its own decisions without requiring specific instructions from a human operator.”

What Exactly is an AI Agent?

What is an AI agent, you ask? Good luck finding a consistent answer. Even ChatGPT, which has invested heavily in creating AI agents, can’t agree with itself on what an AI agent is and does.

Even though experts can't entirely agree on a single definition, there are some core ideas most people agree on, and understanding these can help you cut through the hype.

A Simple Definition of AI Agents

At its most basic, an AI agent is an automated system designed to independently (that’s the key word here) complete tasks on a person’s behalf. Unlike traditional AI tools like chatbots (how are we already using the word “traditional” to describe AI, by the way?), AI agents don't just respond to questions or follow predefined instructions step-by-step. Instead, they can autonomously plan and execute tasks, adjust strategies dynamically, and operate without constant human supervision.

Think of an AI agent as an intern who can manage tasks independently – given a single prompt, it researches, analyzes data, makes decisions, and acts to achieve a goal, even if the goal is complex or requires multiple steps.

Independent Actions, Not Pre-Determined Workflows

As noted above, the distinguishing characteristic of an AI agent is its ability to act on its own. This doesn’t mean that it can (or should) act without human involvement. In fact, it should definitely NOT do that (at least not yet). But an agent is different from a workflow that has been created by a human and then runs by itself.

I encountered this earlier this week when I created an “agent” that, using Zapier, triggers a sequence of events when someone answers the brief “what is an AI agent” survey I created. When I shared it on LinkedIn, I was asked if what I created was a “workflow” and not an “agent.” And it was!

How AI Agents Differ from Traditional AI

So, what’s the difference? Let's clarify with examples:

• Traditional AI (e.g., ChatGPT): Answers your queries and generates content based on explicit instructions. Every action requires your detailed input.
• Pre-determined workflows (e.g., Zapier automation): Executes specific tasks based on pre-defined rules and conditions. In my personal example, I created a “Zap” that when someone completed a survey in Typeform, their contact information was automatically added to my HubSpot CRM, and they were sent an email from my work account. Once programmed, the actions happened dynamically and without my intervention. But a true agent could have done the Zap creation (and the survey creation for that matter) without me.
• AI Agent (e.g., Manus AI or OpenAI's Agent SDK): Receives high-level instructions and then autonomously breaks down the problem, plans the workflow, executes tasks across multiple applications, and adapts on the fly.

True AI agents often use multiple AI models and specialized tools simultaneously, collaborating to deliver more complex outcomes than standalone tools.

Real-world Examples of AI Agents

• Manus AI can independently research, analyze, automate workflows, build websites, manage social media accounts, and even generate comprehensive reports without constant oversight.
• OpenAI's Agent SDK empowers developers to create custom AI agents that automate multi-step workflows like analyzing documents, conducting online research, or automating financial transactions.

What AI Agents Are NOT

It's equally important to know what AI agents aren't:

• Not just chatbots: They're more than conversational tools and can independently carry out complex workflows.
• Not infallible: AI agents still require careful monitoring, tuning, and oversight. Errors and misunderstandings do occur.
• Not Artificial General Intelligence (AGI): While sophisticated, they're far from human-level intelligence or autonomy.

How to Get Started with AI Agents

AI agents represent an exciting leap forward in AI capabilities, but they are largely incapable (for now) of doing many of the things they AI Hype Machine will have you believe they can do.

So for now, my advice is to learn more about what’s possible, try to create some simple workflows or agents of your own, and start to envision what it might look like to have a digital clone of yourself that can do certain things on your behalf.

Upcoming Loop Talks on AI Agents

• On Thursday, May 1, Mary McPheat and I are presenting From Overwhelmed to On It: AI Agents in Your Marketing Workflow at the 30th Healthcare Marketing & Physician Strategies Summit in Orlando.

• One week later, on Thursday, May 8, I'm presenting Deploying AI Agents: Boosting Your Team's Output at Mirren Live, an annual gathering of agency business development and client services leaders in NYC.