I Vibe Coded a WooCommerce Plugin Because Developers Kept Failing Me

I used to think I needed a developer for complex functionality beyond the default options paid plugins give you. Then I spent months managing developers who couldn’t finish the job. So I built it myself. And now I’m wondering if that’s the best or worst thing that’s ever happened to my business.

I needed WooCommerce and NetSuite to talk to each other.

Stock and pricing. That’s it. NetSuite is the source of truth. WooCommerce is where the customers shop. The two need to mirror each other exactly or you end up with oversold products, wrong prices, and a whole lot of angry emails.

Seems simple. It’s not.

The developer merry-go-round

I started where everyone starts. I wrote a scope of work. A detailed one. I’ve been in WordPress for 15 years so I know what needs to be specified and what can be left to the developer’s judgement.

I started on Upwork at the client’s request. The responses came in fast but the quality wasn’t there. After the Upwork failures, I moved to Codeable.io — a more credible platform with higher-level coders. But even there, most wanted a five-hour discovery fee before they’d even look at the project. No guarantee the work would actually get done. Just five hours of paid “discovery” and then we’d see.

We paid it. More than once.

The developers got close. Seventy, maybe eighty percent of what I wanted. But they never delivered something that lasted. Something I could rely on long-term. A plugin that wouldn’t break the first time WooCommerce updated itself, which it does constantly because that’s WordPress for you.

It was demoralising. I spent hours writing scopes, reviewing proposals, managing timelines, testing broken code, going back and forth. And at the end of all that, I still didn’t have a working integration.

I should be clear about something

Those developer costs weren’t mine. They were on-costs to a client. The work started before AI agents existed — before anyone had a real handle on what AI could do.

There’s a timeline. Before AI agents and after AI agents. This project sits smack bang in the middle of the crossover. Halfway through, we realised AI could handle this. So I cut the developer and started using AI instead. The alternative was throwing good money after bad.

The client’s money. My responsibility. I take that on the chin.

When I take on a client project, I carry it. I don’t let things go. If there’s a problem causing operational issues, I’ll keep working until it’s fixed. That’s not optional for me. That’s just how I’m wired.

But I also have to ask myself — were those the developer’s shortfalls, or were they mine? Writing the scope of work was the most tedious and difficult part of the project. Maybe I didn’t give them enough to work with. Maybe the instructions weren’t clear enough. Maybe the failure was mine, not theirs.

That’s part of the frustration. Not just that the work didn’t get done. That the client could have saved that money if we’d known then what we know now. I’ve had to start again. But at least this time there’s real progress.

Holding onto the apron strings of developers was no longer viable. We also had to question their integrity. Now that I know how easily code can be written and troubleshot with AI — particularly with the later models — I wonder how much of what they charged for was actually hard work.

The default plugin problem

There are off-the-shelf WooCommerce-NetSuite connectors on WooCommerce.com. I tried those too. They get you so far. Basic stock sync. Basic pricing sync. Then you hit a wall and need custom code.

The problem with custom code in WordPress plugins is simple. As soon as the default plugin updates — and it will update — your customisations break. Your site goes down. You’re on the phone to the developer. More dollars. More downtime.

Unless you’re a major company with an IT developer sitting by your side 24/7, this isn’t a solution. It’s a recurring problem dressed up as one.

So I stopped looking outward and started mucking in.

Mucking in

I know WooCommerce. I’ve been building in it for 15 years. I don’t know NetSuite, but I know enough about data structures to understand what needs to go where.

I fired up Claude and started asking questions. Feeding it error messages. Letting it diagnose issues. Writing command lines in WordPress to retrieve and test code. Writing rules for the weird edge cases that default plugins never handle.

Like what happens when a NetSuite product has no price? Most plugins would either skip it or, worse, set it to zero. I wrote a rule that puts those products into draft status instead of deleting them. The commerce site mirrors NetSuite exactly, and nothing accidentally goes live with the wrong price or no price at all.

That’s the kind of detail a developer would charge you for. Each one. Separately. With a discovery fee.

The cross-check

I don’t trust one AI to write code. I use both Claude and Codex.

Claude writes. Codex checks. Or Codex writes and Claude checks. Either way, the other one is looking over the first one’s shoulder. If Claude makes an assumption, Codex catches it. If Codex takes a shortcut, Claude flags it.

It’s not perfect. But it’s a lot more reliable than trusting one AI blindly, and it’s a lot cheaper than trusting a developer who’s already moved on to their next client.

The mistakes are mine

When things go wrong, it’s usually my fault.

Not Claude’s. Mine.

I write vague prompts. Claude makes assumptions. Those assumptions are wrong because I didn’t give it enough to work with. I race ahead, stacking new code on top of code I haven’t fully tested yet. I get excited about progress and skip the methodical, step-by-step approach that would actually save me time in the long run.

Learning to slow down has been the hardest part. Writing a clear prompt takes longer than writing a vague one. Testing each step before building on it feels tedious. But every time I skip those steps, I pay for it later in troubleshooting hours.

The eye-rollers

I can hear the developers rolling their eyes from here.

They think people like me — non-coders with AI tools — will come a cropper eventually. That we’ll hit a wall we can’t get past and come crawling back, begging for a real developer to clean up the mess.

I disagree.

If you’re willing to go on a heavy learning curve, have patience, bring a certain level of human intelligence, and can craft a project plan with clear objectives, you can vibe code anything. You don’t need a computer science degree. You need discipline and the willingness to sit with problems until you understand them.

But the uncomfortable truth is this.

Everyone’s too smart for school

There’s a strange mood right now. Everyone feels like they’re too smart for school. Like AI handed them a shortcut past the hard bits and now they’re the expert.

I get it. I felt it too. The first time Claude wrote a working script for me in 30 seconds, I thought I was unstoppable. Then it broke and I had no idea why, because I didn’t actually understand what it had written.

Confidence without understanding is just arrogance with better tools.

The real winners

The winners in this AI revolution aren’t going to be my generation. They’re going to be the 20-somethings just starting their careers.

My daughter and her bunch of creative friends are more cautious with AI than anyone I work with. They use it, sure. But they’re careful. They make sure their original thought stays theirs. They don’t let the machine do the thinking for them.

They’ve grown up watching adults panic about AI taking jobs. So they’re not panicking. They’re paying attention. And they’re protecting the thing that actually matters — their own ideas.

The truth nobody wants to hear

I’m about to say something politically incorrect. Whatever.

Smart people will get faster. Dumb people will get dumber.

AI amplifies whatever you bring to it. If you can think clearly, ask good questions, and recognise good answers, you’ll move at a speed that wasn’t possible before. If you can’t, you’ll just produce more bad work, faster, and think you’re doing great because the machine made it look polished.

The tool doesn’t make you smart. It makes you more of whatever you already are.

The client mirror

I’ve noticed something else lately. Clients I introduced to AI tools are now doing things themselves. They don’t ask me to do it anymore.

I taught them to fish. Now they’re fishing. And I’m standing on the shore wondering if the fishing lesson was a career move or a career-ender.

So do I save myself from irrelevancy by being less helpful? By hoarding knowledge? No. That’s not me. I can’t be that way.

I genuinely love watching clients learn stuff they never thought they could understand. There’s a moment when it clicks and you can see the confidence shift. That’s the best part of this job. I’m not giving that up to protect myself.

The bullshit detector

There’s another side to this. I work with counterparts who overcharge clients for work AI can do in minutes. They wax lyrical in proposals, making simple things sound complicated. And I can tell their own communications are written by AI. Their emails, their proposals, their “expert analysis.” They’re not experts. They’re artificially intelligent instead of naturally intelligent.

Then there are the client managers who suddenly pretend they know more than you because they’ve got AI on their side now. When I get responses that sound like the old Little Britain sketch — “computer says no” — I know they’re out of their depth. They’re not thinking. They’re just forwarding the machine’s output and hoping nobody notices.

I know this because I’m on the other side of the curtain now. I can see the seams.

The biggest lesson

The biggest lesson AI is showing us is that there’s no way around the grunt work and original thought. We must not let go of original thinking. AI is a tool for ideation and for freeing up the creative process. Not a replacement for the thing that makes your work yours.

The gut punch

I used to think if I can vibe code a WooCommerce-NetSuite integration, anyone can.

But that’s not quite right. I brought 15 years of WordPress experience. I know data structures. I can specify business rules. I know how to test. I have enough technical literacy to interrogate what Claude and Codex produce and know when they’re wrong.

AI didn’t suddenly make a non-technical person a developer. It let someone with deep domain knowledge cross a technical boundary that previously required another person. That distinction matters. Because it means the people who’ll thrive aren’t starting from zero. They’re starting from knowing their business inside out and finally having tools that let them execute.

Am I irrelevant? Possibly. But not because anyone can do what I do. Because the gap between knowing what needs to happen and making it happen just closed.

The answer

This is what I do now. I ask clients if they need help with a process. I build them a tool. Then I show them how it was built. It’s up to them after that. Some keep asking me to build things. Some take the knowledge and do it themselves. Both are fine.

That’s the business model now. Not “pay me because you can’t do this.” It’s “pay me because I can do it faster, or take what you’ve learned and run with it.”

I’d rather be generous and slightly terrified than guarded and safe.

Phase 1 is live. Stock and pricing are syncing. The commerce site mirrors NetSuite exactly. Hiccups are guaranteed, but they’re manageable now because I understand the system. I built it. I’m confident. I’m still wondering if I’m relevant. And I’m still building.

Written by a human. Affiliate links may apply. Nuffin’s for free.