What I’ve Learned from Using AI

AI seems to be everywhere, and its use is growing rapidly. Over the last couple of years, I’ve incorporated various AI tools into both my business and personal life. Some have been great—practical, time-saving, and surprisingly inexpensive. In fact, of all the tools and services I pay for in my business, a couple of AI-based services may offer the best bang for my buck.

At the same time, there are plenty of strong and mixed opinions about AI and its current and future impact. I’m not going to get into the economic, environmental, or philosophical debates here. Instead, I thought I’d share a few recent experiences that gave me some perspective on AI’s usefulness—and its limitations—as a tool.

AI has been extremely helpful to me in a number of ways. But I’ve also had some recent experiences that reminded me that, as “intelligent” as AI can seem, it isn’t the same as human intelligence. Its limitations—and its mistakes—can sometimes be surprisingly significant.

I’ll share two personal examples.

The Backyard Project

We’re currently working on a backyard project. After years of battling rabbits in our lawn, I’ve finally surrendered and we’re installing artificial turf. We’re also replacing an older, cracked brick patio.

My wife had a particular look in mind for the new patio and consulted AI several times about how to achieve it, including uploading photos and asking questions in different ways. It never really gave her the answer she was looking for.

Eventually, we went to a local brickyard and explained what we were trying to accomplish. Within about 10 seconds, the person helping us understood the issue and gave us the answer AI had repeatedly failed to provide.

I also used AI to generate images of what the finished backyard might look like. At first, it was impressive. The initial rendering gave me high hopes.

Then I tried to get it to accurately represent the layout of our yard.

Our backyard is essentially a rectangle, with three distinct sections running from left to right. It seemed like a pretty basic concept, but AI simply could not get the orientation right.

I was using the voice feature, so I was essentially having a conversation with it—which itself takes some getting used to because it can sound so natural and personable. What made the experience particularly interesting was that AI seemed to understand exactly what I was telling it.

At one point, our conversation went something like this:

AI: “You got it. So, looking out from the house toward the view, turf on the left, brick patio in the middle, and the diamond pavers on the right, correct?”

Me: “Correct.”

It then generated an image that was even less accurate than the previous one. After several more attempts, I finally said:

Me: “You’re having trouble, and I don’t think you can figure this out.”

AI confidently responded:

AI: “Give me one more try. I get where I went wrong. Let me walk it back quickly. Looking out from the house, left is all turf, middle is herringbone brick running all the way to the fence, and right is the diamond pavers, also running house to fence. Just three side-by-side zones.”

That description was exactly right.

The next image wasn’t.

Eventually, I gave up. What struck me wasn’t simply that AI made a mistake. It was that it could accurately describe what I wanted, confidently tell me it understood the problem, and then repeatedly fail to execute it.

The College Search

We’ve also been using AI to help create a list of potential colleges for my daughter, Avery, who is a junior this year.

As we’ve had an ongoing “dialogue” with AI about colleges, it seems to get to “know” Avery—her interests, preferences, and what she might be looking for in a school. On the surface, it does a reasonably good job of suggesting schools that might fit.

But this process has revealed some limitations as well.

At one point, we asked AI to put together its top five schools for her based on everything it “knew” about her. It generated a list.

We had recently learned about another school that we hadn’t previously discussed, so we asked AI what it thought about that one. Its response was essentially, “That’s actually a great fit and should have been in the top five,” and it promptly generated a new list.

My immediate thought was: What other schools did you miss that should have been in the top five?

That’s an important distinction. AI can be very good at evaluating something you put in front of it. That doesn’t necessarily mean it has identified all of the best options in the first place.

A Tool, Not an Authority

These are simple examples, but they’ve been good reminders for me. As useful and impressive as AI can be, it is still a long way from being a complete solution—even for some tasks that seem like they should be relatively straightforward.

I’ve become even more cautious as I’ve read about what researchers call AI sycophancy—the tendency of AI systems to agree with users, reinforce their assumptions, or provide responses they are likely to appreciate rather than challenge them when appropriate.

That matters because AI can sound remarkably confident, friendly, and authoritative. The more natural these systems become, the easier it may be to give their answers more credibility than they deserve.

Of course, we all know AI isn’t a person. It’s a machine. It may have access to more information than any individual person could ever know, and it can process that information at incredible speed. But information isn’t the same thing as judgment or wisdom.

AI doesn’t have life experience. It doesn’t have values of its own. It doesn’t understand our families, businesses, or priorities the way another person can. And no matter how conversational it may sound, there isn’t actually a person on the other side of the conversation.

That doesn’t make AI less valuable. I expect I’ll continue using it more, not less, because for the right tasks it is an incredibly useful tool.

But that’s the key: it’s a tool.

Used that way, I think AI can be tremendously valuable. But when we start treating it as an authority—or assuming that a confident answer must be a correct or complete one—we’re likely to be disappointed. And I don’t think I need to spell out the implications of what bad or misinformation could do as it relates to planning your finances. In situations that really matter, the consequences could be more significant than a backyard rendering with the patio in the wrong place.

I used AI to help edit this article, a perfect use case in my opinion. 😊