A cautionary tale of cutting corners, eating humble pie, and finally finding the right tool for the job
Let me start with an apology. In my previous article, I called my AI helpers “employees.” Turns out, they prefer “AI assistants.” Who knew digital beings had HR departments? Consider this my official workplace sensitivity training completion certificate.
The Great Helm-OS Disaster of Last Month
Armed with Copilot and ChatGPT 5.1, I embarked on building Helm-OS—version 1 of my digital marine solution. I had visions of grandeur. I had ambition. I had absolutely no idea what I was doing.
Three weeks. THREE WEEKS of trying to make these AI assistants do things they were never designed to do. It’s like asking your toaster to file your taxes—technically both involve electricity, but that’s where the similarities end. I cut corners like a NASCAR driver, pushed limits like a teenager with Dad’s car, and learned more about frustration than I ever wanted to know.
Was it rewarding? Sure, in the same way touching a hot stove teaches you about thermal conductivity.
Enter the Specialist (Thanks, Son)
Last Thursday, after what I can only describe as a “consultation” with my son (he used words like “completely wrong approach” and “what were you thinking?”), I discovered Claude Code. Not because I’m smart, but because I was desperate.
Here’s what shocked me: Claude Code doesn’t hold your hand. No friendly GUI. No casual conversation. It’s like the difference between texting and writing a formal letter—you need to actually think about what you’re saying and how you’re saying it.
Turns out, this is a feature, not a bug.
Five Days of Actually Getting Stuff Done
Claude Code demanded something revolutionary: documentation, strategy, planning, and—gasp—actually understanding what I was trying to build. You know, all those boring project management things real developers do.
Day 1: Created D3kOS (Deck OS, for those keeping track of my terrible naming conventions) and set up a GitHub repository. One day. Not three weeks.
Day 2: Connected all the relevant systems together and started producing actual, working code. Together. Like a team. A team where one member never sleeps and the other survives on coffee.
End of Monday: Beta prototype ready for distribution. I’m about to “release the hounds”—which is my dramatic way of saying I’ll post about it on social media.
I also built the hardware component, affectionately nicknamed “Jellyfish” (official designation: D3-K1, because if R2-D2 can have a cool name, so can my marine gadget). It’s a DIY system that’s actually repeatable, which is more than I can say for my previous attempts.
The Klingon Documentation Incident
Oh, and remember when I mentioned needing to create user manuals for marine equipment that was apparently written in Klingon? Yeah, knocked that out too. All within five days.
The Verdict
Do I have a complete product? Absolutely not. Do I have a working prototype that makes me say “wow” instead of words I can’t print on a family-friendly boating blog? You bet.
The lesson? Sometimes you don’t need an AI “employee” who’ll do whatever you ask. Sometimes you need an AI assistant who makes you do your homework first.
Now, if you’ll excuse me, I have beta testers to recruit and a Jellyfish to deploy.
Stay tuned for the next installment: “How Many Beta Testers Does It Take to Break My Marine Software?” (Spoiler: Probably fewer than I hope.)