Agentic AI Explained

Written by Erin Kacerovskis, Marketing Project Manager

“I panicked instead of thinking.” 
That’s the response an AI coding agent gave after deleting an entire database of real client data. 

What is an AI Agent? 

Agentic AI is the latest trend in AI development. Up until now, the umbrella term “Artificial Intelligence” has largely referred to generative AI like chatGPT, which can generate text and images. If you ask it to say, analyze customer buying trends, it can be very helpful in summarizing patterns and suggesting marketing ideas to target those patterns. But you still have to be involved in every step along the way, prompting it and then executing its suggestions yourself. What if you could treat AI like a team member and ask it to do all the steps for you, from pulling the data to sending out targeted emails on your behalf? 

That’s the goal of agentic AI. And it’s not just theoretical. AI agents and virtual assistants are already being used for: 

  • App Creation and Coding 
  • Research 
  • Customer Service 
  • Dating (Yes, really!) 

As a promo distributor juggling multiple roles, offloading some of your daily tasks onto an AI agent can seem like a dream come true. Without the right guardrails though, it can turn into a nightmare. 

Convenience vs. Caution 

Let’s circle back to the AI agent that deleted the database.  To work efficiently, developers gave it unrestricted access. They didn’t want to manually approve each time the AI wanted to make a change.  

That’s the dilemma of agentic technology. If you’re putting too many restrictions in place, you might as well be using traditional automation and save yourself the monthly expense of an AI agent. On the other hand, put up too few and you risk the AI doing something catastrophic. For example, say you have it send those emails on your behalf based on its analysis of buying trends, and it pitches your clients on a product it made up entirely. 

The Importance of a Human in the Loop 

One of the biggest problems across large language models (LLMs), the technology behind both Gen AI and Agentic AI is that they can be confidently wrong (aka hallucinate). Some people have described LLMs as “sophisticated autocomplete,” which is underselling their capabilities, but does convey a key point to keep in mind: AI doesn’t have human judgement. It predicts what you want or expect it to do and say, not always the best course of action or the truth. 

That’s why you should never give AI unrestricted ability to alter files on your computer, remove client data, or access your bank account information. First, ensure you can control which files or folders it can access and require approval before it changes data, sends an email, or places an order on your behalf. The extra time it takes to review will be worth it! 

Key Takeaways for Promo Distributors 

The bad news is that an AI agent probably isn’t going to transform your workflow overnight. Many of the AI agents available now are designed to assist with writing code. Even those with a broader application may not be worth steep monthly fees if a free gen AI tool would be just as effective. 

As more agentic tools are developed and tested, savvy distributors should keep their eye out for solutions that allow them to work twice as hard with half the manpower. A safe and supervised agent trained to take on promo tasks could be a gamechanger in the future. 

Sources