I’m sick of hearing about AI agents. Talk about beating a dead horse! But if everyone’s talking about it, there must be something I’m missing. Why am I not getting it?

Then I thought I’d found the answer. A real use case, finally—clean up my name and contact info scattered across a dozen data broker sites. Perfect job for an agent: repetitive, rule-based, tedious enough that I’d happily hand it off.

I didn’t have the energy to do it myself. Turns out the agent didn’t either.

What actually happened

The idea was simple: point an AI agent at Spokeo, Whitepages, BeenVerified, and a dozen other sites, let it fill out the opt-out forms, done. First site, it hit a CAPTCHA and stopped cold. Fine, I thought, minor glitch. Second site wanted a photo ID upload to “verify” the removal request—which meant a human had to step in anyway. Third site’s form had changed since the last time anyone looked at it, so whatever the agent thought it knew was already wrong.

Twenty minutes in, I wasn’t managing an agent. I was doing the same manual work I’d wanted to avoid, except now I also had to figure out why the automation broke.

Why this keeps happening

Data broker sites don’t want to be automated. That’s the whole design. So the idea of “let an AI agent handle your privacy cleanup” was broken before I even started—I was asking a robot to sneak past defenses built specifically to keep robots out. The photo ID requirement was the punchline: uploading my driver’s license to a site I was trying to erase myself from. I made the problem worse before I made it better.

Every site’s form is different, and half of them had changed since the last time anyone bothered to check. My agent had a plan. The plan was garbage by site three. This is the part nobody says out loud when they hype these tools: most of what gets called an “AI agent” is a bet that the internet will hold still long enough for the automation to work. It won’t. It never has.

It’s not unique to data brokers. Same story everywhere in IT. Legacy systems, undocumented exceptions, interfaces that were never built to cooperate. Every demo you’ve seen runs on a sanitized environment. Nothing you actually manage looks like that.

The cost nobody mentions

Here’s the part that was mind-boggling. This wasn’t some expensive tool—it was “free.” And it still somehow cost me more than doing the work myself. Twenty minutes to set it up. An hour watching it fail, guessing why, and finishing the job by hand anyway. Net time saved: negative. “Free” isn’t free when it eats your afternoon and hands you back a slightly dumber version of yourself.

You could ask it to check back in three months. But that’s the tell right there: the pitch was “set it and forget it,” and now you’re the one setting a reminder to babysit a robot. If it needs supervision, it’s not an agent. It’s a chore wearing a name tag that says AGENT.

If the tool can’t survive a broken form or an unexpected CAPTCHA, it’s not an agent. It’s a script in a trench coat, standing on another script’s shoulders, only working when nothing goes wrong—which is to say, never.

So here’s what’s actually useful

Not full autonomy. An agent that flags “I’m stuck, here’s why” is worth something. An agent that quietly proceeds anyway, or silently fails, is worth less than doing the task yourself. Ask what happens when it hits a wall. If nobody has a good answer, you already know what you’re getting.

Before you hand anything off, ask a simpler question first: is this task boring and predictable enough for a robot to handle, or does it just look that way from a distance? Most of real IT work is the second kind. That’s the whole lesson. Not “avoid AI agents”—just stop trusting the demo.

Have ideas, questions, or thoughts to share? I’d love to hear from you—feel free to reach out anytime at g.*******@***et.com. And if you want more insights like this, subscribe on our website to get regular tips on cybersecurity, IT management, AI tools,  and more delivered to your inbox weekly.