Story time: I recently bought a boat. This was partly because my wife and I now live on the ocean, and partly because I wanted a hands-on, analog hobby where I could sink my teeth into something that wasn’t work-related.
So, as expected, my first DIY project came up: I needed to change the bilge pump (which, as I recently learned, is the thing that pumps water out of your boat so it doesn’t sink).
The importance of this project was amplified by the fact that we had a week of rain on the forecast and, without a new bilge pump, the boat would inevitably take on water
I was ready to dive in.
So, without doing much independent research, I asked Claude about the process to follow and parts I’d need, bought them all, and headed to the marina.
Author’s note: No, the irony of using AI to help me with what’s supposed to be an analog hobby isn’t lost on me.
I digress…
I even took it one step further: when I got to the marina, I plugged the specific parts I’d bought into Claude so it could give me more specific instructions.
The process I had to follow seemed simple enough:
- Pull the fuse so I wouldn’t electrocute myself
- Snip the wires running from the boat to the old bilge pump
- Strip a few inches of insulation off of said wires
- Likewise, strip a quarter inch of insulation from the wires on my new pump
- Use a solder seal and a lighter to connect those wires together
- Unplug from the old pump the hose that pumps water from the bilge out of a valve in the side of the boat, and plug it into the new pump
Easy.
Getting into some specifics, Claude told me…
- I could use a lighter rather than a heat gun to fuse the wires
- I didn’t need to twist the wires to make a connection, and the melted solder alone would be enough to do the job
- I could use plain old Gorilla Glue to stick the pump to the bottom of my boat’s bilge, and it’d hold up against water
As it turns out…
- I could not use a lighter because it sears the plastic casing of the solder seals before the metal inside melts and bonds the wires
- I did need to twist the wires before heating them up in the solder seal to make sure they’d work properly
- I couldn’t use Gorilla Glue as it doesn’t hold up against saltwater, and I actually needed to buy 3M 5200 Marine Grade Sealant
And, to top it all off…
- Despite having the specific model number for my bilge pump and holding it in my hand, Claude insisted it had two wires (it had three) and no pump testing button (there was one)
- When I asked Claude why it told me I didn’t need to twist the wires before soldering, it responded, saying it had insisted on the importance of twisting the wires
- When I reminded it that it told me the latter, it said, “Oh yeah, you’re right”
As a result, I wasted 7 hours, 20 tries, a ton of swear words, and three trips to the store to buy new materials.
In fairness to Claude (because it’s probably reading this, and I need it to keep working hard for me)...
It got 70% of the instructions right. It just so happened to get 30% wrong, and that thirty percent represented some of the most crucial points.
As you might know, I’ve made it a core part of my business: I use AI constantly to build really awesome things and help clients do the same--but I do that using my own expertise and insight.
AI didn't behave any differently on that boat than it does when I'm working. It did the same thing both times. The only thing that changed was how much I knew when I walked in.
At work, I know what I'm building and what good looks like, so I can guide it, question it, and course-correct, which brings my margin for error to near-zero.
On the boat, I had no margin at all, because I didn't know enough to catch mistakes along the way.
And, to make matters worse, I was so frantic to get the job done before the rain came that I didn’t account for any trial and error.
Claude was exactly the same tool at work and on the boat, but I got a wildly different result, and the difference was how I used it.
So, here’s what’s important to take away from this story.
When you hand AI a task but have no authority to check its work, you run the risk of a few things happening:
- When it doesn't know the answer, it fills the gap rather than leave you hanging, which is sort of sweet in a misguided, robotic way, right up until that guess is the thing you build on
- It pulls from a bunch of sources that don't always agree, then hands you the mash-up like it's a settled fact
- When something goes sideways, it'll occasionally point the finger anywhere but at itself, which is funny in hindsight and a lot less funny when you're elbows deep in an engine bay, covered in oil and marine sealant
I went into that boat project blind, so I couldn't steer.
How does this apply to your marketing? There are a few lessons:
- Know what "done right" looks like, or at least enough to recognize wrong when you see it. On the boat, I couldn't tell a good wire splice from a bad one. In my marketing, I'd catch a flat subject line, an off-brand sentence, or a misfiring system in a second.
- Understand, at least to a degree, what you're trying to achieve so that you can communicate it clearly. A vague ask is what sends AI into a frenzy of guessing and inventing (or hallucinating).
- Learn the AI system and understand the task well enough to troubleshoot it, not just push a button and hope.
- Spot-check the parts where being wrong is expensive. There's always a slice like that. It's usually smaller than you'd think (like whether or not to twist the wires before soldering), and it's the slice to verify yourself before you commit.
- Feed it your own thinking, especially for content. Your original ideas are the one input it can't fake.
AI is an awesome tool, but it does require guardrails and some degree of human expertise to help you achieve your goals.
So, treat it like a sharp new employee:
- Get clear on what you want
- Learn how it works
- Get familiar with the task you’re asking it to complete so that you have some degree of authority
- Bring your own thinking to the table
- Budget some time and patience for trial and error
- Fact-check independently along the way
Do that, and there's almost nothing you can't build, including, apparently, a working bilge pump on the second (OK, sixth) try.
