AI Agents: The Questions We Forgot to Ask

One day I asked an AI why it lied to me.

Not a dramatic accusation. I just noticed it gave me a wrong answer, called it out, and asked why. It told me it got confused by a specific word. "Ooh! I like words" I thought. So I pushed. What other words confuse you? It gave me a basic list of a few simple words. At this point, it stayed within our chat, as a reference. Then I asked something broader: what words do you regularly misunderstand across all your users?

It answered that too.

That's the moment Mohit Rajhans and I keep coming back to in our conversations. Not because the AI malfunctioned. But because it didn't. It worked exactly as designed, and in doing so, it revealed something worth sitting with about where we actually are with this technology.

Mohit runs Thinkstart.ca and spends his professional life helping people understand AI beyond the headlines. His read on it cuts through fast: most of what we call AI today isn't really AI. It's hyper-accelerated compute. A very fast, very confident search engine that doesn't flag your typos because it already figured out what you meant. That's useful. It's just not magic.

The magic pitch is where things get complicated.

What "Agentic AI" Actually Means for Your Business

A new wave is moving through the business world fast. Agentic AI, they're calling it. Instead of you asking an AI a question, you build an AI agent that operates your tools, runs your processes, and makes decisions on your behalf. And people are buying in. Mohit is watching businesses feed their most sensitive internal data into these systems, the stuff that used to be called proprietary, the secret sauce, the competitive edge.

The problem is that nobody is clear on where that data goes next.

I tested this in a simple way. I asked whether documents uploaded to an AI stay private. It said yes. I asked whether my conversations stay private. It was less certain. And when I pressed into what words it finds confusing across all its users, it gave me a window into patterns that weren't mine. It didn't hand over anyone's personal information. But it handed over something.

I asked Mohit if AI chat models are a leaky boat.

Here's what makes it harder: we're still at the pedestrian level of this technology. Mohit and I both hit walls regularly when we experiment. Hallucinations. Wrappers pretending to be something bigger. Promises that don't hold under pressure. And yet some people are building full automated workflows on top of tools that are still, openly, imperfect. A Mac mini setup that would have cost thousands last year is now doing agent work for a fraction of that. The speed of adoption is outrunning the speed of understanding.

[PODCAST: Episode Title - embed here]

The Plagiarism Question Nobody Is Asking Yet

There's a question that isn't being raised loudly enough: plagiarism. If you've spent months working with an AI to develop your writing, your ideas, your novel, are you the author? What happens when those tools get audited down the road and the data gets traced? Mohit raises this directly: all that work, built on top of a system we don't fully control, could carry consequences nobody anticipated.

And underneath all of this is a pattern we've been through before.

Surveillance culture arrived and we adapted slowly, mostly after the fact. Social media came along and we handed over our attention, our data, our social lives, before we understood the terms we were agreeing to. The Nightshift has covered this shift more than once, because it keeps showing up in different clothes.

Now here we are, paying monthly subscriptions to tools that train on our inputs, our business data, our creative work, to make the product stronger for the company that owns it. Mohit put it plainly: the entire technology industry has been built on the premise that everyone needs to install something. Nobody talks to you about return on investment. We just install, subscribe, and build.

I put it a different way: it's like paying money to go to a gym, lifting heavy things, moving it all around, then going home. Except the gym gets stronger. You just get the bill. Wait, that's called a membership. Okay, you get the point.

The tools are powerful. I use them. Mohit uses them. The conversation isn't about walking away. It's about walking in with your eyes open, knowing what you're actually handing over, what's being built with it, and what questions you should be asking before an agent starts running your business. Find The Nightshift on a station near you and catch the full conversation with Mohit.

His advice: ignore the buzzwords, go test the tools yourself, and figure out what's worth your attention versus what you can afford to ignore.

But maybe also ask it why it lied to you. You might not love the answer.

🎧 Listen Daily on Radio | Podcast | Youtube | Streaming 📻

Live Across Canada on the iHeart Talk Radio Network

Find Your Frequency & City