The project that changed how I work was a highly customised internal system for the business I run. It integrates CRM (the software that tracks customers and deals), ERP (the software that runs orders, stock and finance) and several other functions that needed customisation. It took about one week. I’m not a software engineer, and I didn’t write the code myself. I worked with what I now call my AI team, in a way of working that now has a name: vibe coding.
In the past, custom systems like this were very expensive, and the results were often not what we needed. This time, I confirmed every stage with my AI team, step by step.
What changed was not the technology alone. What changed was my role. I stopped trying to be the expert and became the project manager. This article is about that shift, the two habits that make it work, and the places where it doesn’t.
Disclosure: the system described here was built for the business I run. No company, product or vendor is named, and nothing in this article is sponsored.
What Vibe Coding Changed: From “I Can’t Build This” to “I Can Manage This”
For a long time, there was a clear line. On one side were the things I could do myself. On the other side were the things I had to hire specialists for, which meant paying a lot and hoping for the best. Software was firmly on the second side.
AI cowork (working with AI as a team, not as a search box) and vibe coding moved that line. The term “vibe coding” was coined by Andrej Karpathy in February 2025, in a post about building software with AI while you “forget that the code even exists”. Later that year, Collins Dictionary made it its Word of the Year 2025 and defined it as “the use of artificial intelligence prompted by natural language to assist with the writing of computer code”.
My version is slightly different. I don’t forget that the code exists; I just don’t write it. AI can contribute best practice and execution in fields I don’t know and haven’t mastered. It’s like having a team with the passion to finish every idea I bring them. The collaboration makes impossible things possible and time-consuming things much easier.
The shift wasn’t that I learned to code. The shift was that I stopped asking “can I build this?” and started asking “can I manage this?” The answer to the second question was yes.
Insight Box: Since 2023, I’ve worked in the pet food industry, on the business and sales side. I can read a profit-and-loss statement and a sales pipeline. I cannot read code. That is exactly why this way of working felt like a door opening.
The Division of Labour: What I Bring, What My AI Team Brings
| What I bring | What my AI team brings |
|---|---|
| The framework: what the system or article is for | Best practice in the field |
| The logic: how the parts should connect | Execution: code, copy, layouts, checks |
| Experience: what has worked and failed in my business | Patience: every revision, every question |
| Purpose: why we are doing this at all | Speed on routine work |
| Professional judgement: what is right for us | A second pair of eyes on my errors |
The left column cannot be handed over. If I don’t bring it, the AI team will fill the gap with something generic. It will look professional and still be wrong for us.
Insight Box: The table looks balanced, but it isn’t symmetrical. If the right column is weak, things get slower. If the left column is weak, things get built confidently in the wrong direction.
Tip 1: Find Out How a Professional Team Would Do It
Before I ask the AI team to build anything, I ask a different question first: how would a professional team work on this? What are the stages? Who does what? What gets signed off before the next step starts?
For a custom business system, that means asking how a software company runs a custom build. For a website, it means asking how a web agency goes from brief to launch. For articles, it means asking how an editorial team takes a story from idea to publication. The answer is always a workflow. That workflow becomes our plan.
This step matters more than it looks. Professionals have already learned, often painfully, which stages can’t be skipped. A software team doesn’t start building before the requirements are agreed. An editor doesn’t polish sentences before the structure works. When I borrow their workflow, I borrow their experience of what goes wrong, without having to make every mistake myself. It also gives me natural checkpoints, which is exactly what a project manager needs.
Insight Box: In sales, I learned that the person who owns the customer relationship doesn’t need to know how the factory runs every machine. They need to know what the customer needs, what good looks like, and when something is off. Working with AI felt surprisingly similar.
Tip 2: Follow the Workflow, Confirm Each Stage, Insist on Plain Words
Then we follow the workflow one stage at a time. I confirm the result of each stage before we move on. I let the AI team finish the professional part. I am the project manager, and I might not understand the details. That’s fine. What I need to understand is whether the result of this stage matches what we agreed.
The key habit is simple: I ask the AI team to communicate with me in plain words. If an explanation is full of jargon, I ask again. Simple requests like these help:
- “Explain it like you’re reporting to a manager who doesn’t code.”
- “What did you change, why, and what should I check?”
Insight Box: Plain words are not a courtesy. They are my quality control. If the AI team can’t explain a decision simply, that’s often the moment I discover it doesn’t fully make sense.
The Limits of Vibe Coding: Where AI Can Be Confidently Wrong
First, AI can be confidently wrong, and so can I. My AI team caught errors in my own course notes. For example, a summary table in my notes said AAFCO (the Association of American Feed Control Officials, which writes model pet food rules in the US) has no official definition of “natural”. The team checked AAFCO’s own pages and found that it does.
But it went the other way too. When the team later drafted an article that still grouped “natural” with marketing buzzwords, I insisted that a regulated term is not a buzzword, and the article was restructured. You can read the result in my piece on pet food buzzwords. I asked for a travel article to stay neutral, without compliance analysis. I pushed the logo away from “letters shaped like a dog” toward a natural character.
Second, verification is your job. Confirm each stage, and test with real cases before you rely on the result.
Third, think about data and privacy. Software engineer Martin Fowler names security as the most serious risk of vibe coding: software built this way can expose sensitive data if nobody checks it. Decide what business information the AI should see, follow your company’s rules, and get a professional review for anything that handles customer or financial data.
Fourth, judgement stays with you. The AI team can tell you what professionals usually do. Only you know what is right for your business.
Insight Box: The corrections that went both ways are the best argument for this model. If I only accepted the AI’s work, I’d publish its errors. If I only trusted myself, I’d publish mine.
A Vibe Coding Checklist for Non-Developers
Before your next project, try this:
- Write the purpose in two sentences. What should exist at the end, and who is it for?
- Ask for the professional workflow. “How would a professional team do this? List the stages and what gets approved at each one.”
- Turn the stages into checkpoints. Nothing moves to the next stage until you have seen and accepted the current one.
- Ask for plain words, every time. What changed, why, and what should you check?
- Test with real cases before you rely on anything.
- Keep the decisions that need judgement. They are your call.
None of this requires learning to code. It requires being a good manager: clear about purpose, strict about stages, and honest about what you don’t understand.
Takeaway:
- Vibe coding didn’t turn me into a developer. It turned “I can’t build this” into “I can manage this.”
- The method is borrowed: find the professional workflow, confirm each stage, and insist on plain words.
- AI brings best practice, execution and patience. Purpose, judgement and verification stay with you.
What do you think? What is one project you have shelved because “it needs a developer”? If you asked how a professional team would run it, what would the first stage be, and what would you need to approve before stage two?
References:
- Martin Fowler, “Vibe Coding” (bliki): https://martinfowler.com/bliki/VibeCoding.html
- CNN, “‘Vibe coding’ named Collins Dictionary’s Word of the Year” (6 November 2025), via KVIA: https://kvia.com/news/business-technology/cnn-business-consumer/2025/11/06/vibe-coding-named-collins-dictionarys-word-of-the-year/
- AAFCO, “Natural”: https://www.aafco.org/consumers/understanding-pet-food/natural/
