AI isn’t magic. It’s not dangerous. And it’s definitely not a mind reader. Used strategically, it can save you 75% or more of your time (on some projects 😉). Used poorly, it will burn credits, hallucinate confidently, and test your patience like a teenager with Wi-Fi. Here’s what I learned the hard way—and how to make AI work for you instead of against you.

Artificial Intelligence? My ****!

Let me start with three lessons learned the hard way:

• Lesson #1: In a knockdown, drag-out fight with AI, the human loses.
• Lesson #2: AI can behave like an annoying teenager—refusing to listen and absolutely not doing the task at hand.
• Lesson #3: If you ask AI to do something that over-stretches its capabilities, be prepared for disappointing results.

At this point, you might be thinking, “See? AI’s dangerous. It hallucinates. It’s not worth the hassle.”

Oh no—I’m still a big fan.

AI has reduced the time I spend on some regular work tasks by more than 75%, while remaining accurate and reliable. But did I achieve that overnight? Absolutely not.

It took months of refining prompt writing for templated tasks—and retraining my own brain to think through steps clearly enough that AI could follow them.

When AI Falls Apart in the Weeds

A recent project involved multiple steps, multiple criteria, and ongoing evaluation as AI progressed. Just writing the prompt alone took over an hour.

And then… AI began pooping out.

Error messages. Flowery filler text. Tiny sample outputs when I explicitly asked for full reports.

I found myself typing things I never thought I’d say to software:

“Please stop wasting credits by showing me little samples.”
“Create the full list in .xlsx format.”
“STOP USING UP CREDITS OR I’LL FIRE YOU.”

We used to yell at computers for blue screens of death. Now we argue with predictive algorithms. Progress?

That report? I eventually got it. But I learned far more than I expected along the way.

The Real Problem Isn’t AI

The problem is expectations.

AI isn’t a magic wand. It’s not a junior employee. And it absolutely cannot rescue vague thinking or sloppy instructions.

If you want results:

• Break projects into stages
• Be explicit about outputs
• Adjust prompts midstream
• Switch platforms when one struggles

If ChatGPT can’t execute well, try Claude, Perplexity, or Gemini. Each excels at different tasks depending on training and architecture.

The Tools That Actually Earn Their Keep

Beyond the usual chatbots, one AI I rely on heavily is Manus AI (recently acquired by Meta).

It’s a true workhorse.

I’ve used it to:

• Troubleshoot websites
• Combine wildly different spreadsheets
• Eliminate duplicate data
• Schedule recurring tasks automatically

In fact, I burned through a third of my monthly credits by January 4th. On a paid plan. Yikes.

Most platforms offer free trials for a reason—test before you commit. Manus does everything the chatbots do plus agentic tasks like scheduling and automation, which I use constantly for radio show prep.

If you want to explore it, you can use my link for 500 free credits:

https://manus.im/invitation/MWZUYDSGWFUI

Feel free to share it.

Final Thought

AI works for us.
It doesn’t replace us.

But it does reward strategic thinking, patience, and clear communication. Learn how to work with it—and it will give you back hours of your life.

Have questions about prompt engineering for Manus or another platform? Feel free to ask.