Why every product manager must try vibe coding?

Background

Over the past few weeks, I have spent time learning about the strength of generative AI and even built some fun prototypes. However, this time I wanted to challenge myself, and I have built an end-to-end application called Link Spaghetti. Building the app wasn’t as easy as I had hoped, and I learned a lot about Vibe Coding. Link Spaghetti is built on React, Vite, Python Flask, MongoDB, and AWS S3 for storage, and is fully integrated with Stripe.

Although I come from a development background, I stopped writing code when frameworks like React emerged. With Vibe coding, I refreshed my skills and even learned a few new concepts. In today’s blog, I will discuss the pros and cons of vibe coding and outline a few reasons why, as a product manager, you should start building today.

Pros

Improves technical understanding

Vibe coding can help you gain a deeper understanding of coding. The moment you press Enter with your first prompt, you are asked to make several decisions. I am currently using Copilot on VS Code, and getting started with a new app was overwhelming, but also really educational. Copilot asked me what frameworks I prefer and also how I want to structure my code. Reading through the reasoning and thinking on my copilot chat, I was clearly able to understand the packages required to build Link Spaghetti. With my education benefits, I have access to premium models, and I specifically prefer the Claude Sonnet 4.0 for its explanation of how each step is evaluated and executed. By carefully reading through the chat, I was able to understand what tokenization is, how the front end interacts with my Flask API, and catch some errors when Claude went into loops trying to fix a bug.

Sharpens Product Focus

Vibe Coding helped me realize the importance of providing valuable business use cases. While it can be easy to simply get lost in a rabbit hole of writing code with all the features you desire, it is essential to understand the value these features bring to the current state of your product. While building Link Spaghetti, I had to pause and park some excellent ideas to build a solid MVP and focus on the use cases of each feature I was adding. By examining the business side, I asked Claude to provide a better UI/UX and keep the product simple.

It saves you serious money

Vibe coding is a money saver; this one is self-explanatory. Especially if you are a student, take advantage of the GitHub Education benefit to save yourself some money and time. It would cost me anywhere from $7,000 to $10,000 if I were to hire a developer or an agency to build Link Spaghetti. But with Copilot, it was peanuts. Vibe coding has a certain charm to it when you compare the money you can save. Building new products has never been easier, especially if financial constraints are holding you back.

Cons

Debugging can be challenging

If you don’t understand the basics of how web apps work, security, or programming, it can be difficult. You can really get lost when fixing a simple bug. While building out Link Spaghetti, I encountered many instances where the same error occurred, and even Claude would struggle to understand my prompt, which would result in the app adding random features. Fortunately, I used GitHub with good commit messages, so I can keep track of the changes. Although Generative AI can write good code, it can sometimes start to hallucinate and give you irrelevant code. To avoid situations like this, continually test your app and maintain version control, allowing you to roll back whenever needed.

Code Efficiency & Scalability Are Limited

While vibe coding can feel smooth and easy, there is no explanation of how your code handles searching or complex logic. The apps built by Vibe Coding are not scalable. The code generates a lot of logs, and the reasoning/explanation is not particularly focused on explaining its choices for solving a problem. Currently, it is only suitable for building prototypes. In hindsight, it is essential to remind your Generative AI to also prioritize efficiency.

Disorganized File Structure

This could be a difficulty I faced because I would ask my generative AI to document every change it made. Claude would store all the files in the root directory, making it challenging to understand the code structure. Sometimes, the Generative AI would not even test its code; it would only end by stating the changes it had made. Only after pointing out the bug would the AI fix it. Additionally, it is easy for Generative AI to get lost in circles trying to fix bugs. The leading cause was that it would write the same piece of code in multiple places, which really goes against the concept of “Don’t Repeat Yourself” in coding.

Conclusion

To test the limits of Generative AI, I built a production-ready application using React, Flask, AWS, MongoDB, and Stripe. The entire process took me about three weeks of work, one hour a day, and it was very easy and generally smooth. With vibe coding, you get to brush up your development skills, improve product focus and save a ton of money. However, it is easy to get lost in loops, fixing bugs and building a slow-performing product. In general, as a product manager, I was able to better empathize with the work done by a development team, clearly communicate business requirements, outline the trade-offs for a proxy, and improve the testing and validation of a product. If you haven't tried vibe coding as a product manager, you should do it today; you are only a prompt away.

If you enjoyed this post, here are a few ways to go deeper:

Check out my Prompt Engineering Guide: Prompt Engineering Playbook

Connect with me on LinkedIn