TECH AI AGENTS ARE HERE. HOW AFRAID SHOULD WORKERS BE?

AT MOBILE WORLD CONGRESS, the telecom industry trade show held each March in Barcelona, AI agents were everywhere. Or rather, signs touting AI agents were everywhere.

- JEREMY KAHN

At the Google Cloud stand, telecom execs could watch demos showcasing how easily companies can build custom AI agents. At Microsoft's stand, it was the same story. Qualcomm's highlighted its “on-device agentic AI.”

It was a similar scene in January at the World Economic Forum in Davos, Switzerland—where signs pitching tech companies’ “agentic AI” offerings practically obscured the view of the surrounding snowcapped mountains. In fact, almost every place tech companies come to market their wares these days turns into an al-agent fest.

There's simply no buzzier topic in tech and business right now. AI agents are supposed to be what finally delivers AI's long-promised productivity gains. It’s still early days, and businesses are being cautious, partly because AI agents carry more risks than other AI products. But if deployment ramps up rapidly, as many analysts expect, AI agents could radically transform how people work—and possibly lead to millions of jobs being cut.

But what are AI agents, and is the hype around them deserved? For starters, there is no agreedupon definition of an AI agent—which is convenient if you want to claim to be selling one. But generally, it’s an al-powered system that can complete tasks using other software tools. These systems have a generative-AI model at their core, but they can do more than the AI model can in isolation. A gen-AI model may be able to suggest an itinerary for an upcoming vacation. An AI agent could do that, too, but then actually make the bookings and reservations for you. This is obviously a more difficult task that involves multiple steps and use of the internet—and, to do well, quite a bit of reasoning.