From Sci-Fi to Wi-Fi:
What Can We Expect
from AI Agents?

Unless you've been living under a rock, you’ve probably noticed that AI agents are the hottest new trend in tech. Every AI vendor out there is hyping them up as the future of automation and productivity.

But what exactly are AI agents? Are they just chatbots with a fancy rebrand, or is there real innovation behind the buzz? Should you start using them today, or are they still more promise than practice? And most importantly, which use cases actually deserve an AI agent, rather than just another overworked automation script?

Let’s cut through the hype and see what AI agents are really made of.

From Horizontal AI to AI Agents: My Take

To understand where AI agents fit into today’s fast-evolving AI landscape, it helps to rewind a bit. When ChatGPT burst onto the scene in late 2022, it didn’t take long for organizations to start asking: “How can we harness the power of large language models (LLMs) beyond just generating blog posts and witty tweets?” The real goldmine, they realized, was their own internal content—buried in platforms like SharePoint, OpenText, and Salesforce.

Fast forward to last year, and we saw a surge in AI applications designed to tap into this organizational knowledge. These tools weren’t just answering questions—they were reasoning, summarizing, and even creating content, all while grounded in the full set of organizational data. Perhaps the most well-known example? Microsoft’s M365 Copilot, which has been making waves as a workplace productivity powerhouse. I often refer to these AI apps as ‘horizontal AI’ since they reason over the complete set of content or data, without being an expert in any domain.

And I’ll admit—it was a big step forward. These horizontal AI applications helped bridge the gap between scattered company data and meaningful insights. But for all the hype, I found them lacking in 2 crucial areas: accuracy and action. They confidently made claims but sometimes ‘forgot’ crucial info. They told me what to do but didn’t do it for me.

Why AI Agents Matter More

While horizontal AI tools like M365 Copilot can certainly boost productivity, there are two key limitations:

• They can be a little… overconfident. These tools pull from all the organizational content you have access to. Sounds great—until you ask about the company’s work-from-home policy and get an answer from an outdated HR document you still had stored in your OneDrive. Or you request a summary of customer interactions, and it conveniently forgets to mention that rather scathing complaint letter. The bigger the haystack, the harder it is to pull out the right needle.

• They talk a big game, but don’t do much. These AI tools excel at answering questions and generating content, but when it comes to actually executing tasks? Not so much. They’re great at telling you how to do things, but not at doing them for you.

This is where AI agents step in. Unlike their horizontal AI cousins, agents don’t just go through an entire company’s content. Instead, they can be configured to pull knowledge from a specific, well-defined set of data—ensuring more accurate, relevant responses. But more importantly, they don’t just talk—they act. AI agents can perform tasks based on their knowledge and, in some cases, even make autonomous decisions.

Think of it this way: If traditional AI applications are like friendly, chatty personal assistants who can give you general information on just about anything, AI agents are more like specialized experts who roll up their sleeves and get things done—within a focused domain, of course. Borrowing an HR analogy, if horizontal AI tools resemble generalists who know a little about everything, AI agents are the T-shaped specialists: deeply knowledgeable in their specific area and ready to take action.