AI Is in My Way Now

I should probably not be the guy writing this. I run an IT and cybersecurity company that sells AI services. I run a marketing agency. I just finished writing a book about small businesses using AI, and I'm in the middle of editing it. I have told more business owners than I can count to stop being scared of this stuff and go use it.

So it's a little awkward to admit that lately, AI is in my way.

Not the technology. I still think it's the best thing to happen to small businesses in a long time. What's in my way is what happens when people start using it without anybody ever deciding how it should be used, and I've been watching that play out in my own company for a while now. I've also been reading enough about it to know we're not special. This is going on in a lot of places.

The way I've come to think about it, it goes in four stages. Here's how it went for us.

Stage one: the emails

Somebody on the team gets really into AI. A couple weeks later, you start noticing that everything they send has been through it first. And you do notice. Everybody notices. The messages get longer. They have this shine to them that the person never had before. They open by restating whatever you just said and they wrap up with some line about being happy to help further, like you're a customer and not the person who signs their paycheck.

People who used to answer me in one sentence are now sending four paragraphs, and I promise you three of those paragraphs are filler. I bring up something I'm concerned about and what I get back is not their take on it. It's a chatbot's take on it with their name at the bottom.

It's not even that it's annoying, although it is. It's that it costs everybody time. Ten seconds to generate, three minutes for me to read, and I'm not the only one getting these. Whatever time got saved on one end got pushed onto the rest of us. And on top of that, when I ask somebody what they think, the whole point is I want to know what they think. If I wanted a chatbot's opinion, I've got several of them on my phone.

Turns out there's research on this. Stanford and a company called BetterUp did a study last year and gave it a name, workslop, which is AI-generated work that looks finished but doesn't actually accomplish anything. Forty percent of the people they surveyed had gotten some in the previous month. The number I keep coming back to is that 42 percent of the people who received it said they trusted the sender less afterward. Which is the part nobody who does this seems to understand. You think you look sharper. The people around you are quietly downgrading you.

Stage two: the trust goes away

And nobody tells you. That's what makes it bad. Nobody sits down and says I've stopped trusting your work. They just start skimming your emails. They double-check your numbers. They stop pulling you into stuff that matters. You don't find out, so you keep going, and the gap gets wider, and the whole thing started with some long emails that nobody wanted to be petty enough to bring up. I didn't bring it up either, for the record.

Stage three: it gets into your systems

This is where it stopped being a pet peeve for us.

The same people who are excited enough to run their emails through AI are excited enough to build apps with it, and honestly I'm fine with that part. I want people building things. But then the apps start talking to the real systems. Somebody hooks up an API. Somebody sets up a sync into the database the whole company lives in. Nobody reviewed any of it because there was no process for reviewing it. They just had access. And a while later company names are wrong in the CRM and records don't line up and data that other people rely on is quietly messed up because of something one person built in an afternoon with an AI coding assistant.

I have spent thirty years telling businesses not to let this happen with their servers. Don't let one person be the only one who knows how it works. Don't push anything to production without another set of eyes on it. Every one of those rules applies here, and almost nobody applies them, because this doesn't look like an IT problem. It looks like somebody being productive.

Stage four: dependency

This is the one that actually got me to sit down and write this. Eventually somebody tells you they can't do part of their job today because their app is broken.

We did that job for years before there was an app. The job did not need the app. The app was supposed to make the job easier. But that's not what got built. What got built was a person and their homemade tool standing between the company and something the company already knew how to do, and when the tool broke, the work just stopped. Nobody went back to the old way, because in their head there wasn't one anymore.

So that's the pattern I've watched. Long emails, then nobody trusts each other, then your data's wrong, then one day AI is the reason a sale didn't happen. And the first place you could see the habit was in the emails.

I want to be careful here because somebody is going to read this and decide I'm an old guy yelling at the future. I'm not. The book I wrote tells small businesses to use more AI, not less. The best people I know use it constantly and most of the time I can't tell, because they use it to think and then they write to me in their own words.

In the write-up of that study, the researchers had a way of putting it that I liked. Pilots and passengers. A pilot is steering the thing and adding their own judgment. A passenger is letting it carry them and skipping the hard part. Your coworkers can tell which one you are. I've never had a problem with the tool. I have a problem with passengers.

And if you're one of them, here's what I'd want you to think about. What are you actually getting paid for? It isn't typing. It's that you know the client, you know which of those four paragraphs matters, you know when the tool is wrong. Nobody can subscribe to that. But when everything you send has been run through AI, nobody can see it either. You didn't lose it, you just buried it. If I can't find you in your work, I've learned nothing about you, and honestly I trust you a little less than I did last month. The tool doesn't pay for that. You do. The people getting more valuable right now are using AI to make their judgment more obvious. The people getting less valuable are using it to hide.

This is what I actually mean when I say rogue AI, which is what I'm calling the book. People hear that and picture something out of a movie. It's not that. Rogue AI is AI that got into your business through the side door, one excited employee at a time, with no plan and nobody deciding how much of it belongs where. Because the right amount of AI in a sales job is not the right amount in bookkeeping, and neither one is the right amount for the person who touches client data. Somebody has to actually make those calls, set some guardrails, and then check that they're holding. That's what an AI initiative is. It's a real project with somebody's name on it, not a Google Chat message that says go play with ChatGPT. We're doing it ourselves right now at Directive and BriteAgency, in our own operations and our own marketing, and it's what most businesses I know of have not gotten around to.

For what it's worth, the rules I've landed on are pretty simple.

• Use AI to think and use yourself to talk. Argue with it all you want, then close it and write the message yourself.
• You own everything you send. There's no such thing as "the AI wrote that."
• Nothing you build touches a shared system until somebody else has looked at it. Build in a sandbox all day long, but the second it writes to data other people depend on, it gets reviewed.
• Every automation needs a fallback. If your tool breaks and the work stops, you didn't build a tool, you built a single point of failure with a chatbot in front of it.
• I don't care if you used AI or not. I care whether what you handed me was right, and clear, and actually yours.

Looking back, I should have said something the first week the emails got long. It felt like a dumb thing to bring up. It wasn't.

If people on your team have started sounding like a press release, that's the beginning. The rest is already on its way. Have the conversation while it's still about writing style and not about your database.