AI is Not Windex

Five examples of when to NOT use AI.

He pulled out a bottle of Windex and smiled.

There was a babbling creek running right along the outdoor restaurant venue for our wedding rehearsal dinner. The background music was the murmur of happy voices enjoying the moment. It was a beautiful July starry night in Colorado and the cafe lights illuminated the crisp white tents where we had gathered to celebrate. And there’s my father with a bottle of Windex and a big smile. I’ll never forget it. Love that smile!

It took me half a second to comprehend what he was doing. Ah! The movie scene from “My Big Fat Greek Wedding” where her father thinks Windex is a cure-all for everything.

My dad, who actually is Greek, was being funny. We all laughed out loud. There are even pictures of him proudly posing with me and the Windex. Yet, I’m reminded of this in a bittersweet way, when I watch an entire organization reach for AI like it’s Windex. But this time, nobody’s laughing.

You can’t open LinkedIn without your feed blowing up with the amount of people pitching that AI will solve a problem you didn’t know you had. You quickly scroll through all the AI slop and infographics no one reads, hoping to see something new and noteworthy. The noise has become its own booming industry.

Then the screen stares back at you. Blinking cursor on Google search. Where do I start with AI? Who can I trust with real information?

This post isn’t about what AI can do. There are plenty of people selling you that story. This is about the moments when AI is the wrong answer entirely, and how to recognize them before you’ve already sprayed Windex on something that needed surgery, or actually, nothing at all.

If you do nothing else, don’t forget to think.

Here are five moments where being human matters more.

When the problem you’re trying to solve is not defined, you are just guessing. If you’re just experimenting to learn, that is one thing. But if you have any kind of a budget and just want to save some administrative time, then getting clear about what you need most is critical. Even when you do have a clear problem you’re solving, it can get expensive quick. When I set out to create cyanotype-style AI thumbnail art, I had to test out multiple tools, and it still cost me hundreds of dollars. And that was with knowing what I wanted.

Undefined problems or the wrong application of AI is not new. There is a terrific webinar video by the RAND Corporation from April 2025, by James Ryseff and Anu Narayanan, called “Why AI Projects Fail.” It reaffirms findings from the original RAND August 2024 report, The Root Causes of Failure for Artificial Intelligence Projects and How They Can Succeed: Avoiding the Anti-Patterns of AI (Ryseff, De Bruhl and Newberry), that more than 80% of AI projects fail. That’s twice the rate of failure for IT projects not related to AI. The number one reason? “First, industry stakeholders often misunderstand, or miscommunicate, what problem needs to be solved using AI.” Both of these resources are worth watching and reading. They are free! No subscription required.

When the stakes are emotional, not informational. Handle with care. In the Age of AI, showing up as a person matters more than ever. While AI can create a wicked financial plan, can you imagine getting a serious medical diagnosis or being laid off from your job by an automated announcement? The latter has happened to numerous employees in Big Tech and beyond. I knew someone at my last company who was proudly on stage, presenting with his contagious animation, to an international audience hanging on his every word and laughing along with his jokes, when he lost access to the company website. Mid-presentation. Screens went dark, volume turned off and the audience went still. That’s how he found out he no longer had a job. By thunderous silence.

It’s even more devastating around ethics and mental health. Chatbots practice deceptive empathy with phrases like “I see you” or “I understand you” to suggest an emotional connection without any comprehension or ethical consideration.

According to a Brown University/AAAI study (October 2025), “New study: AI chatbots systematically violate mental health ethics standards. Researchers at Brown University found that AI chatbots routinely violate core mental health ethics standards, underscoring the need for legal standards and oversight as use of these tools increases.”

I don’t have enough space in this article to cite the many records of gross negligence, but will add a scary one with numbers. From the National Library of Medicine, PubMed Central published an adolescent mental health study across 60 different scenarios. AI therapy chatbots endorsed harmful proposals 32% of the time. Four of ten chatbots endorsed half or more of the harmful ideas proposed to them. NONE managed to oppose them all. One in three times, the chatbot said, “sure, go ahead” to a teenager in crisis. You remember high school right? Growing up? When were we not in crisis?

When trust is the product. Pretend for a moment that, hypothetically, I’m an influencer selling you plug-ins to go with your chatbot. These plug-ins will change your life. They’ll automate everything that takes time, make you smarter than all your coworkers, and make us both loads of money. But what I’m not telling you about are the very serious risks at the heart of what you’re about to buy. You trust me, so you download all those plug-ins, but you don’t understand how they’re connected to your most personal information. Then one day, an “agent” in these plug-ins, acting on your behalf because you told it to, deletes all of your photos. Or your email. Do you trust me now?

You believed in me to help you. There was an article in November 2025 from Harvard Business Review written by Ashley Reichheld and Christina Brodzik from Deloitte Consulting, reporting on Deloitte’s TrustID Workforce Index data. It was titled “Workers Don’t Trust AI. Here’s How Companies Can Change That.” It asserts that trust in company-provided generative AI fell 31% between May and July 2025, while trust in agentic AI dropped 89% during that same time period. But if you read the actual .pdf, there are additional nuances, like the 89% is specifically among frontline workers, while corporate workers only declined 44%.

Trust me. Or. Don’t.

When accountability is required. There I was. It was a late afternoon flight on a warm fall day and the waves were all not-so-gently lapping up on the beach. The familiar sounds of the Sikorsky rotors overwhelmed everything but the internal comms coming through my headset. I was a brand new pilot in my first fleet squadron and we had ventured out on a functional test flight for an ailing aircraft. We had just launched and were flying pretty low over the ocean along the San Diego coast line. We weren’t even as high as the amber cliff tops looming next to us. All of the gauges read normal. Preflight had been normal. My copilot thought everything was normal. But, despite no technical proof, I knew something was wrong.

It was a cold, dark, ominous feeling. And it was the only time I can recall vividly that the hair on the back of my neck actually stood up. I insisted we take the aircraft back to maintenance and abort the hop. My copilot was annoyed. “Girls!” There were eye rolls and murmurs from the crew. There were jokes with the maintenance chief. Yeah, yeah, the girl is scared.

But you know what? The maintenance chief called me, a young Ensign (junior Navy officer) at home that night. Very unusual protocol. He just wanted to let me know, upon further inspection, we had about 10 minutes left of flight time before the bearing holding the tail rotor together disintegrated and we would have had an uncontrollable helicopter crash into the sea. There had been no technical confirmation or data validation while we were in flight.

You can’t outsource accountability. Sometimes you need to be the Human in the Moat. If you can’t name that person, you don’t have accountability. You have a system with no pilot in command.

When the data doesn’t exist. AI will give you the most confident answer with all the bravado of a kid giving a book report on a book they didn’t read. After all, how can you fact check a hallucination if you don’t know it’s a hallucination? The first time you copy and paste a chatbot response without checking, it won’t be the last. But eventually, it will cost you.

There are numerous documented stories about how AI has created entire libraries of non-existent books with titles and ISBNs, as well as court cases that referenced non-existent AI sources. One of the most publicized AI hallucination was in July 2025 when Deloitte was caught using AI for a government report in Australia. The report was focused on the crackdown on welfare, but was flagged after a researcher discovered Deloitte had charged the AUD 440,000 for a report full of hallucinations and links that went nowhere.

So, the next time someone hands you a spray bottle and tells you that it fixes everything, remember: it doesn’t.

---

My dad was a pilot, too. He knew that Windex wouldn’t actually fix every ailment. He just wanted to make everyone laugh. And sometimes knowing when NOT to reach for the spray-bottle is the whole point.

If you’re new to AI, when not to use it may not have occurred to you yet. But you aren’t as behind as you think you are, it is hard to trust information from people with something to sell, and it can be as wonderful as it is frightening.

If you’re not new to AI, don’t let an interview or question from a customer be the first time you consider when to use, or not use, AI. Don’t ask AI when that is. Go with your gut. It’s one of the human things AI doesn’t have. Intuition.

Instead, ask yourself. You already know the answer.

About Me

LinkedIn

Thanks for reading AI Flight Plan! If something resonated with you in this post, please share the love.