I have a confession to make. It is 2:47 am and I am sitting at my computer because I can’t sleep. Not that I couldn’t fall asleep, but I woke up and can’t get back to Dreamland because something is bothering me and my mind won’t shut off.

Each month, I am responsible for writing a blog that goes on my company’s website and gets posted to the socials. This month, I chose the topic of AI agent sprawl and how an Agentic Control Plane (ACP) like IBM watsonx Orchestrate is the perfect solution to cataloging and managing agents from different platforms and applications. With an ACP, it is possible to take your siloed agents and tie them together to initiate automations that cross org charts and departments for true enterprise AI. ACPs are a great story because they let you deploy agentic AI across your entire company with observability, security, and coordination.

I finished my writing and then did something I always do upon completion: I fed it into AI and asked it to check my grammar and spelling, as well as to identify rough paragraph transitions or sentences that are awkward. I took some of the suggestions and applied polish to my product. I was happy with what I had written.

As an AI Practice Leader at LRS, I am constantly testing AI platforms to understand what is possible. I have built a vacation travel planning application and another one that consolidates the content of my two favorite fantasy football podcasts (I planned on drafting a lot of Chicago Tight End Colston Loveland this year, and now I see others are also excited about him, as he is already an early season riser on multiple fantasy platforms. He’s going to get a thousand targets from Caleb Williams this year. If you play, he needs to be on your team. You heard it here first.).

Back to tech. Another thing I created was an AI content detector based on nine dimensions and thought it would be interesting to run my blog through it. This is where things got weird.

The detector rates content on a scale of 1-100. The higher the score, the more likely it was written by AI. I received a score of 43, which was way higher than I expected. AI concluded that there was “substantial human involvement, but AI-assisted.” It dinged me on having three evidentiary points in ascending sophistication to illustrate the risks associated with AI agent sprawl. Apparently, using bullet points is also something humans should no longer do in their writing.

I was angry. 43? “I’ll show you,” I thought to the AI, so I rewrote my blog making some structural changes to the things it had flagged me on in the first pass. When I submitted it to the detector again, it registered a 46. What. The. Hell. How could it be higher?

We all know AI slop when we read it. Even when people apply profile rules to AI output to make it sound human and pass AI detection tests, you can tell it is fake because it still reads strangely. It’s hollow. My blog wasn’t hollow.

Still frustrated at my score, I rewrote it a third time and the score did not change. I informed my AI friend that the content was 100% generated by me, a real person. The explanation I got back for my score was that my sentence lengths were too similar and that my framework had a “perfect structural completion.” Can I help it if that is how I write?

It was not until my current insomnia episode that I began to think about what I was doing. I had gone from creating an informative blog on an issue with which many of my partners are struggling to writing a blog that satisfies an AI framework that rated my content. I was no longer writing to bring attention to a solution for people, but for my desire to have AI give my work a better score. I am going to take a vacation from AI for a few days to reset myself and make sure I am motivated for the right reasons. Winning over the heart of AI should not be one of them.

It’s ironic that I started out writing about how companies can lose control of their environments through agentic sprawl, and I lost control of my own creative process by deferring to an algorithm designed to tell me whether my work was good enough.

While I take a personal break from AI to reset my brain, don’t let your organization’s AI run wild. LRS is still here to help you with automation. If your org has more than a handful of agents running in silos, you already have a sprawl problem. Let’s map it and fix it.