There’s a lot of noise and hype around Agentic AI, and Jason and I fall into a different camp than the majority.

First, Agentic AI could be a new term for you, but it’s easier than you might think. If artificial intelligence can take action, it’s agentic. It’s similar to an .EXE file (an executable file). It can do something. And it doesn’t require a human to do so.

The thought is that if agentic AI can do the things people do now, we’ll need fewer people doing the things. Work things, specifically.

We’re huge advocates and power users of AI, and we read and report on AI news and advancements every week in our AI4 Recruiters program. This gives us a unique perspective on the variance between what’s being hyped and what’s actually happening.

Before I dive into this variance, let me share our seemingly minority opinion.

Agentic AI has tremendous potential, but the potential negative consequences aren’t worth the efficiency. Yet. 

I understand the hype. If they (all the smart AI engineers) can make Agentic AI work consistently, it’ll be incredible and will change the way we do business. It will change life as we know it.

Here’s the problem. We’ve all been trained to trust computers and LLMs are not trustworthy nor capable. Anthropic set up an agentic AI micro company (it was a vending machine) internally and had Claude do all the work. It was a colossal failure. Effective in about 30% of situations. Lost massive revenue.

A couple months ago, after a developer worked for weeks on a new database, having put all the security guardrails in, he lost all of his work because the agentic AI deleted it. But hey, the AI apologized and took full responsibility. And wasn’t able to recover the data.

Klarna went “AI First” in the third quarter of last year and laid off thousands of customer service representatives. They lost $125 million in the first quarter of this year and have now added back human support at a VIP level.

We need to be careful. It feels slick until you have to deal with an agentic AI that kills your deal, or posts something absurd on your behalf, or emails your entire network with nonsense.

Your reputation is your most valuable asset. Why would we ever entrust something so important to a computer program that consistently makes things up and is built to agree with you? We shouldn’t.

I hope to write a different version of this article soon. Once we get a little farther along and we can feel comfortable dipping our toes into low-value, low-visibility tasks that might actually increase our efficiency without so much risk. Stay tuned.