Agentic AI Market Hype is Déjà Vu All Over Again

Reading recent news about the AI industry is like watching a horror movie where the kids choose to hide from the killer in the barn full of deadly weapons rather than drive away in their car. I want to yell that they have it all wrong and should make a better decision. What’s more, we’ve seen this movie before. The market created unrealistic expectations around generative AI, spurring companies to pour billions into technology that is yet to deliver results. And now it’s happening again, this time recasting any form of Large Language Model-calling application as agentic AI. 

If you believe everything you read, you now think that AI agents are ready and able to take over business functions from customer service to cyber security to design and advertising. You also might have the impression that rules-based AI systems are archaic and unnecessary.

What’s an enterprise CEO or CTO to do? Most haven’t yet figured out how to deploy generative AI in a way that delivers meaningful business results with a sustainable ROI. And now they’re supposed to forget about that and focus on agents? 

My message to enterprise leaders and others is to resist the urge to abandon everything you learned from your generative AI experience and pivot into an equally rushed pursuit of agentic AI. 

I personally guarantee that your business will require rules-based, deterministic AI-driven technology used in conjunction with whatever agentic AI capabilities you manage to put in place. Contrary to the hype machine, you won’t turn agentic AI loose on structured business processes like purchases or refunds. Here’s why.

Most of what drives meaningful value for enterprise organizations is tied to structured business processes, like purchases and refunds, credit card transaction disputes, accepting or rejecting health insurance claims, or tasks like adding a wheelchair to a flight. 

All AI that exists today, including agentic AI, is rooted in statistics, just by definition, which means that some of its conclusions and decisions will be wrong. This is why agentic AI, which is supposed to make decisions and take action without intervention, cannot be turned loose on structured business processes because they’re too important to get wrong. 

Furthermore, the best examples of AI “agents” today are merely following the pre-defined steps a human would perform and then using AI to attempt to automate each respective step in the order they’ve been outlined. And that’s okay because it can deliver results. But you know what this is called? That’s right: rules-based AI.

In other words, you might trust agentic AI to manage your calendar or to (slowly) execute a partially scripted walk-through of a website to book you a restaurant reservation, like OpenAI’s Operator, but would you trust it to handle costly refunds, automatically respond to emails on your behalf, or manage air traffic control? All of this is to say that if enterprise leaders approach agentic AI the same way they did generative AI, we’re setting ourselves up for another big letdown when it comes to the impact that AI will have on business.

Instead, why can’t we be practical and measured about what agentic AI can do for us today? It’s all in the science behind it. For NLX, that means using the right tool for the right job at the appropriate time, resulting in a combination of autonomous AI and structured business logic so that companies leverage the latest innovations to drive efficiencies while guaranteeing certainty and safety where needed.

Have a specific use case in mind? Reach out to our team, we're always happy to talk.