A USF graduate researcher pushes back on the AI-jobs panic from inside the lab where the technology is being built. ORACLE GRAPHIC / AYRON ESPINO-NARDI

We were testing our AI voice agent on a fake 911 call.

Panic in the caller’s voice, the kind of call our Emergency Command Center project was built to triage during a surge event.

The model got the words right. It found the location, flagged the emergency, and routed the call.

But what it couldn’t do was make the person on the other end feel less alone.

That’s the part I keep thinking about when people ask if AI is coming for my job.

Nearly 40% of enrolled students have thought about changing their major, field of study or coursework because of AI, according to a 2026 CNBC/SurveyMonkey survey.

I get why.

But I’ve spent the last year building AI systems, in a USF lab and across four hackathons, and I think that fear is pointed in the wrong direction.

AI isn’t replacing people, it’s replacing tasks. Those are not the same, and the difference is going to matter more than the panic does.

Related: OPINION: $10 million says USF gets artificial intelligence right

In USF’s MAX Systems Lab, I work with AI models that read images and text together,which are referred to as multimodal models.

One piece of the system reads the images. It’s called the vision encoder.

I once spent a week trying to speed things up by duplicating that piece, since it seemed like a small, easy part to touch.

Measured improvement: zero.

When I actually measured where the model was spending its time, the piece I’d duplicated barely mattered. It accounted for 0.16% of the work. The part I hadn’t touched, the language piece, accounted for 99.84%.

I’d optimized the wrong component, and the system never told me. It just ran, same speed, indifferent to what I’d changed.

That is the part AI can’t do for me. It will happily generate the code to duplicate an encoder in under a second. It won’t tell me I’m solving the wrong problem— I have to find that myself.

I see the same pattern outside my own lab, too. I also work with Dr. John Templeton, in USF’s Bellini College of Artificial Intelligence, Cybersecurity and Computing, on an AI model for a swallowing study.

AI can spot patterns across huge amounts of clinical data faster than any person could, Templeton said, but it can’t tell a real signal from bias or bad measurement on its own.

“AI should be viewed less as a replacement for clinical or scientific judgment and more as a tool for augmenting it,” Templeton said.

Same story, different field. The tool moves fast. The judgment is still ours.

Rishil Shah, a USF graduate student in computer science interning as an AI architect at NMEDA, put it a different way when I asked him about it.

“A job isn’t fixed,” he said. “It’s a bundle of tasks that’s been reassembling itself for as long as people have worked.”

AI is just lowering the cost of some of those tasks fast enough to be disorienting, which isn’t the same as removing the role.

“Producing a first version is cheap now,” Shah said. “Judging whether it’s right is the part that takes real training. That part got more important, not less.”

He sees the same pattern outside software. A model can draft a nurse’s chart note or run an engineer’s calculation, but somebody licensed still signs it.

That responsibility hasn’t moved. It was never a technical thing, it’s a question of who’s answerable.

However, I won’t pretend the fear is baseless.

Research out of Stanford’s Institute for Economic Policy Research shows hiring for entry-level, AI-exposed roles has dropped since 2024, in fields like software development and customer service.

A New Yorker analysis found confidence in the job market among recent college graduates collapsed from 49% to 19% in a single year.

Something real is happening to how companies hire junior people, and it’s worth taking seriously.

But switching your major because AI might eventually touch it isn’t the move.

Every field on that CNBC list is going to change. Medicine and law are next. Computer science already is. There’s no AI-proof major.

What actually helps is treating AI like a skill you’re expected to have, not a threat you’re trying to dodge.

AI skill requirements already show up in 79% of U.S. tech job postings, according to Dice’s tech hiring report. It’s baseline now, not a niche skill.

USF’s Bellini College built a free, self-paced microcourse called AI Whisperer, open to anyone including students, three to four hours, that actually teaches you to work with these tools instead of fearing them.

Get into one AI-adjacent project or class before you graduate, not to become an AI person, just to see where the model’s judgment runs out and yours has to start.

AI can process a panicked 911 call faster than any dispatcher alive, but it still can’t make someone feel less alone while they wait for help.

That gap isn’t closing anytime soon. It’s the reason my job still exists.