Avoiding Personal AI Catastrophes

Cautionary tales and lessons of when too much AI automation is a bad thing

This is a very, very painful AI Lesson to write. You see, this edition is the result of an epic failure. I’ve had a vision for how I could automate a large portion of my work and build one of those billion-dollar, one-person companies that Sam Altman has famously predicted. I’ve been trying to engineer something that simply isn’t possible yet. I wanted to dictate my daily newsletter while I walked my dog, and then have AI agents handle the copyediting, formatting, publishing, and yada, yada, yada.

You saw some fancy layouts over the last week, and the feedback was…mixed at best. And then I realized that my fancy layouts were not impressive, it’s not why you read this newsletter. It’s because I really do try to figure out how to apply AI to real business problems.

But that wasn’t the reality. It could do the pieces well. But it just couldn’t put them all together, and I still needed to be present in the process. And I got so fancy, AI couldn’t keep up with my ideas, and even if it could have, it probably shouldn’t have.

So today’s edition was written the way I started, with a blank page, and Grammarly telling me that my grammar is atrocious and I am wildly inconsistent. True and true.

I have a heuristic for sorting AI tasks, which I call the S.M.A.R.T framework. It stands for Sort, Match, Automate, Refine, Take Control. Here are the lessons learned for everything I did wrong, so you don’t have to make my mistakes when you build your workflows.

// Sort - I am good at this part. For the newsletters, I’ve been doing this for over 3 years, and I know every part of the publishing process—the research, the writing (I love this part), the revision, the copyediting, and my nemesis, the layout. I also didn’t use AI to tell me what it could and couldn’t do well. I was surprised when I asked it to critique its ability to handle the workflow I gave it. It actually pointed out where I was a much better than I was.

// Match - Here’s where I messed up. I started to think that the tools were better than humans. AI can’t tell my stories, it can’t think my thoughts, or know my opinions. It can create really cool pictures better than me, though (thanks to Google Nano Banana and ChatGPT Images 2.0).

// Automate - So now I have automated a process with the wrong tools and became too overconfident in my beautifully authored skills. It turns out they were too complex; they had thousands of lines of markdown when it seems like they should be limited to a max of 500 lines. I should have gotten more rigorous in creating my skill.md files following best practices instead of pushing the envelope. I literally spent over 80 hours creating four skills that I thought could run my process. All I did was compound human mistakes at machine speed.

// Refine - The system worked exactly as designed — that was the problem. I had all sorts of subagents, creating review protocols that superseded my judgment and my taste. I should have gone through my thought process, asked people for feedback, and not delegated. I was enamored by my new design maven, Claude Design.

// Take Control - It’s one thing to use AI as a thought partner; it’s fine to automate well-defined tasks, but when it comes to doing something truly exceptional, I don’t think we can lean on AI for anything above a B. For repetitive tasks, yes. Draft that polite email to decline an invitation, or share your scheduling preferences, but don’t use it for replying to a tense situation or one that requires empathy.

People often mistake me for an AI expert, but I am not. I just get mistaken for one on the Internet. I am well-versed, but I am trying to keep up with the technology just like you. I am building stuff, but not models and GPUs. I am building websites, presentations, and conference software for All Things AI. Pretty much the same kinds of things you are.

I am also trying to keep up with the tools while running a couple of businesses and providing training and upskilling to people in my community. Just like you are doing your day job as an executive, a marketer, a developer, an engineer…insert your role here. Then you want to get home to your family and friends, enjoy your hobbies, and your dog, and not be tethered to a 60-hour workweek.

That’s my dream too.

So, I am going to rethink my strategy for the AIE newsletter. I think I am going to try to do less automation, less fancy layouts, and keep things simple. Give you the benefits of my learning—make it relevant, actionable, and shorter.

Oh, and I am still going to use the hell out of the em dash, I always liked them even before AI became a thing.

Your AI Sherpa,