Agentic AI 101: What's an AI agent and when should they be used?

In order for AI to become a truly revolutionary force in life and business, we’ll have to trust it with more than just our grunt-work. We’ll have to outsource some of our decision-making as well.

That’s where agentic AI comes in.

An AI agent is a system you can interact with in plain language that has some amount of capabilities it can manage, including taking actions on your behalf. The simplest AI agents are “helper” agents such as support agents or virtual assistants. (Clippy was the godfather of this sort of thing). You can ask these agents to book your plane tickets or revise some text, or which barbecue grill would best suit your needs. 

Other AI agents run when you're not doing anything. They’re constantly listening, and can take some action if they think it's necessary.

That makes them close to what we used to call automations. The key difference between an automation and an agent is that the agent has more than one function, and can make its own decisions and interpretations of what it’s supposed to do.  

Here are a few thoughts about the value of agentic AI as a force multiplier and facilitator, and why companies that are replacing human beings with AI agents are thinking about this technology the wrong way.

Why should we use AI agents?

Not all tasks fit neatly into a simple binary. Think of a governance agent that’s supposed to enforce brand guidelines or highlight when content should get another review. You might want more complexity in its output than a cut-and-dried yes or no. 

Your AI agent might decide that the problems with a piece of content are egregious enough to block it from being published, or that it needs someone else to review it, or there’s some other review state it should go into. Based on what’s happening, it might decide which tools to apply for the job. There are multiple paths that an AI agent can follow, and it has some agency in deciding what they are. 

In the Old World, you'd have to account for each of those branching paths, and if the scenario didn't fall into one, then nothing would happen. The advantage of AI agents is that they’re equipped to deal with situations they’ve never seen before. They’re not forced to operate within what they were “programmed” to do.

Of course, there's a time-saving aspect as well. Humans are very expensive, and compute is not, so there may be certain repetitive tasks that an agent is better equipped for in terms of resource efficiency.

For example, sifting through a great amount of content, summarizing things, identifying outliers — these are all things that can be pretty time-consuming for humans to do. AI agents can do these tasks quickly, and even help you identify opportunities that you might not have discovered yourself.

Contentstack’s rule for AI development

As Contentstack prepares to roll out new agentic AI features this fall, our Product team has benefitted from a simple guiding principle: If it doesn’t deliver significant user value, we don’t do it.

So, what is “user value”? In my opinion, it’s anything that falls into one of these three buckets:

Number one, it does something that you're already doing, but 10 times better. For example, instead of sifting through all my data and having to hand-identify all the audiences, an AI agent can break up my audiences based on lookalike models. I could have done this myself, but with AI I can do it (at least) 10X faster.

Number two, it allows me to do something that I literally could not do before. It's not just helping me with what I could have done otherwise. The AI enables some net new capability that was not possible before. 

And number three, it can turn a regular user into a power user. Let’s say we wanted to add natural language search in our CMS, so you could type in and find what you’re looking for without having to figure it out through advanced search. Rather than having to become a search power user or rely on your developers, the AI functionality gives you the same abilities as the most skilled users of the platform. 

If an AI concept does not meet one or more of these criteria, then it's not worth our team’s investment. And by the way, none of those criteria boil down to “it allows your company to cut the size of its workforce.” Which brings me to my next point…

Why AI agents should enhance human effort, not replace it

So far, companies have taken a couple different approaches with agentic AI. One is trying to literally replace people. So, if you have someone whose job is to do X, they’ll seek to create an agent that does exactly that.

But I think we get a lot more value from AI agents when we use them as facilitators, not workforce replacers. You hired smart people to use their talents, right? If that’s the case, you should probably get rid of all the things they have to do in the course of the day that don't differentiate you and your brand. Are there capabilities that you can give a team member via an agent that provides a force multiplier or allows them to provide more value? Start there.

We’re quickly entering a world where AI will be able to put marketing on autopilot. It will create the campaigns, it will write the content, it will generate the assets, etc. And my question is always, “If everybody has the exact same tools, how is a great marketer going to beat a good marketer?” Because if you have a system that can spit out average content and everybody's doing it, things are going to be very noisy.

It's not that AI can't help you develop novel experiences. But if you're creating content for other humans, then humans need to serve as the judge of taste. When AI tools are generating the same campaign ideas, the most successful marketers are the ones who will be better at deciding which ideas to pursue. 

Looking ahead: AI agents will eat more tasks from the bottom up (and that’s a good thing)

Whether it's industrial farming or computers, new technology always eats jobs from the bottom up. It could be literal jobs that you have to do, or those “jobs” could be abstraction layers.

In the early days of code, we were interacting directly with the hardware itself and writing in ones and zeros. Then we added operating systems, and now you can use a mouse, which is a layer on top of an operating system. We used to have to manage our infrastructure by taking racks to a server farm and plugging in the cables and installing the software, and now we have cloud services. We used to write Javascript and HTML by hand. Then we developed code frameworks where you don't even need to know the underlying Javascript. 

AI is the latest layer on top of all this. Soon, you won't necessarily need to know how to use Contentstack. Our platform is an abstraction of an abstraction of a database that you used to have in order to do all these jobs. 

AI agents are simply the most recent iteration of what technology has been doing for a long time. The real purpose of AI — just like all the advancements that came before it — is to get you to the value part faster. You can now spend most of the hours in your day doing the high value things that are going to have an impact, and less time figuring out how to learn the software.

Conor Egan is the SVP of Product at Contentstack. Follow him on LinkedIn.