You’ve probably heard it from another business owner by now. They built their own CRM over a weekend. They replaced the software they were paying for with something they made themselves. No developer. No agency. Just a prompt and a few hours.
Here’s my take: vibe coding, the way it’s being sold, is bullshit.
AI really can build software. I build with it every week: websites, workflows, narrow tools that each do one job, documentation, even a company intranet. They’re real, they work, and I’m not going back.
But building it was never the hard part. Owning it is.
You Already Know This. You Own a Business.
If you own a business, you’ve met the person who thinks they could do what you do. All it takes is some money and a few people who know the work, right? Rent the space, hire the staff, open the doors.
You know better. You’ve been through the years of actually running it. You know about everything that never shows up from the outside: the slow months, the staff who quit, the supplier who’s late, the thing that breaks on your busiest day.
Michael Gerber wrote about this decades ago in The E-Myth Revisited. He called it the Fatal Assumption: “if you understand the technical work of a business, you understand a business that does that technical work.” A great hairdresser opens a salon and finds out that cutting hair was the easy part.
Software is the same. Owning a piece of software is like owning a business of its own. Building it is the start. Keeping it running, fixing it and growing it over the long haul is the real job.
Vibe coding sells the haircut. It leaves out the salon.
It Looks Like a CRM. That Doesn’t Make It One.
Some of the people saying “I built my own CRM” really did. A lot of them built something that looks and feels like a CRM. That’s not the same thing.
A CRM isn’t the screen with your customer list on it. It’s everything behind that screen:
- Where the data lives, and whether it’s backed up when something goes wrong.
- Who can get in, and what each person is allowed to see.
- What happens as it grows, from 50 customers to 5,000, or when two people edit the same record at the same time.
- What it’s connected to, like your email, your calendar or your payments, and what happens when one of those changes how it works.
- Who fixes it when it breaks, and whether anyone besides the person who built it can understand how it works.
If you don’t know those pieces exist, you don’t ask for them. The AI won’t insist. So you end up with something that works fine on day one and quietly piles up problems underneath.
Developers call that technical debt: shortcuts that cost you later, with interest. With a CRM, it’s also a liability. That’s your customers’ names, emails, phone numbers and history. If it leaks, or just disappears one day, that’s your problem, not the AI’s.
And here’s the part the weekend builders skip. AI can build you a CRM that mechanically works. It isn’t a team of real people watching it, catching problems early, fixing what breaks and supporting the people who use it. The right people using AI can get it to do a lot of that. Most business owners can’t.
The Weekend Is the Easy Part. I’ve Done It.
I’m not saying this from the sidelines.
When I relaunched my own site, the first version went live 14 minutes after my first prompt. That was 12:56 in the morning. The last fix went in at 5:09 AM.
That sounds like the vibe coding dream. It isn’t. Two weeks before that, I’d spent two late-night hours on nothing but the thinking: what the site needed to say, who it was for, three design directions and five rounds on the homepage. Behind that was 15-plus years of building sites. When I sat down with the AI, I wasn’t asking it to figure out a website. I was telling it exactly what to build.
The AI did the typing. The decisions were already made.
Then there’s the after. Here’s what my last couple of weeks looked like:
- A few of the websites I host got hacked through a flaw in one add-on. Someone had quietly given themselves the keys. Everything looked normal from the outside. I found it, cleaned the sites up, fixed the hole and locked them down.
- A security flaw meant older sites needed an upgrade. Three live client sites broke the same afternoon, and I put them back the way they were.
- A fourth was broken for five days and nobody knew, including me, because visitors were being shown a saved copy of the page.
None of those sites were vibe-coded. That’s the point. Every piece of software needs this: updates, security, backups, the thing that breaks at 2 AM. Nobody selling “build it this weekend” mentions any of it, and it’s most of the job.
Why the Things I Build Actually Work
Everything I build with AI works for one of two reasons. Either I can maintain it, because I understand how it’s built. Or I scoped it tight on purpose: a minimum viable product, the smallest version that does the job well, with very little that can break.
That’s not luck. It’s experience.
I’m not a programmer who writes code fluently. But I don’t just understand how the pieces fit together: the database, what you see on the screen, the code running behind it, the hosting, the domain settings. I’ve had my hands in all of them. I’ve been knee-deep in databases. I’ve dug through PHP code, made edits by hand and learned the code myself. I know HTML and CSS almost fluently. I’ve built on PHP, MySQL and WordPress, and I’ve worked on app development projects.
I’m not deep in every one of those, and I’m deeper in some than others. But I’ve touched enough of it to know what it all looks like and what it all does, because I’ve worked with it with my own hands. I’ve been in the trenches when shit breaks, and I’ve been the one who had to fix it. Sometimes it was small. Sometimes it was massive, with a lot riding on me solving the problem, for big businesses and small ones.
I’m a Swiss Army knife, not a scalpel. AI fills the gaps where I’m not deep. And because I understand every piece, I can direct it like a good field general instead of hoping it figures things out.
Everyone has access to the same AI. Not everyone has the experience to aim it. That shows up in four places:
- Knowing what to ask for. When I asked AI to build this blog, I didn’t say “make me a blog.” I listed the behind-the-scenes pieces it needed so Google and AI tools can find it, read it and show it properly. The AI wouldn’t have insisted.
- Catching what it gets wrong. AI is confidently wrong in small ways all the time. It looks fine to the AI every time. You only catch it if you know what right looks like.
- Knowing when to say no. AI doesn’t click around inside ad accounts or anything that handles money. Nothing gets set up that can bill me later. Nothing goes public without a person saying yes.
- Knowing what “done” means. Done isn’t “it works on my screen.” It’s checking it on a phone, in Google, in a shared link and a week later. That broken site hiding behind a saved copy? A plain check said it was fine.
The Bottom Line
None of this is an argument against AI. I use it for everything, and it’s going to keep getting better. In the right hands, it’s close to a superpower.
But in business, in life and in nature, a shortcut is never a replacement for the process. The process is where you learn what can go wrong. It’s what makes you the person who knows what to do when something breaks.
So if you built your own tool, or you’re about to, ask yourself the questions you’d ask about any business you were buying:
If you can’t answer them, you don’t have a CRM yet. You have a demo.
And if you own a business, the website or tool you just launched is day one, not the finish line. Before you ask who can build it, ask who’s handling it for the next three years.
If you want someone who knows the whole job behind your website or your tools, not just the weekend, let's talk.