I recently started evaluating Cursor AI to increase my productivity when developing applications. This tool allows to build applications using natural language (Vibe Coding) and AI (LLMs and Agents). To my surprise, I was able to adapt rapidly and developed a few applications within hours.
To really evaluate the potential of Vibe Coding and Cursor AI, I decided to build a mobile application with React Native, which I did not have any prior knowledge using this framework. After adding the React Native documentation to Cursor AI, I was able to guide the agent to successfully developed an end-to-end mobile AI application. I was even able to include a subscription payment and some nice transitions between screens.
Since then, I try to use Cursor AI as much as I can, which has allowed me to build applications 10x faster. However, there is a big controversy around using these tools, because it can lead to serious issues during production. Mainly, because the developer doesn’t complete understanding of the entire code base.
To better describe Vibe Coding in 2025, I have the following analogy:
Let’s say you want to move from point A to point B. You can get there by walking. It will take some time, but you can safety make it. if you know how to drive a car and have enough experience, you can reach the destination 10x faster. However, if someone doesn’t know how to drive or has very little experience, the chances of crashing are high.
If you already have some experience developing applications and want to start using Cursor AI or any other Agentic IDE, I have some initial tips that have helped me avoid issues and move faster:
Commit and push often
When asking Cursor AI to implement a new feature or refactor code, it can make a huge change in your code base and potentially introduce bugs or break your solution entirely.
Once your application is in a working state, commit the changes to your Github branch. With this approach, you can revert changes if a new feature introduces bugs.
Test new changes before accepting them
When working on a new feature, you can see the changes that Cursor AI has made. You can test the changes before accepting them. Make sure everything works as expected and then accept the changes.
After that, apply the first tip which is commit and push to your Github branch.
Update your dependencies
One of the biggest issues I’ve seen with Cursor AI is that it will include old dependency versions to your package.json or requirements.txt
Because of this, the code may not work as expected. Try to instruct Cursor AI to include the latest version for each dependency or you can add them manually.
Most of the bugs I encountered during development were due to outdated dependencies that caused compatibility issues.
Clean your code often
Instruct Code AI to refactor and clean your code as often as you can. This approach will keep a readable, maintainable and clean code base.
Switch to Pro models for complex tasks
I encountered few occasions where the default models were not able to accomplish a correct solution. Once I switched to pro models, for example Gemini 2.5, Cursor AI was able to come up with a correct code to solve more complex features.
Provide official documentation
Give Cursor AI as much context as you can. You can provide official documentation to APIs, SDKs, or frameworks.
Conclusion
To conclude, Agentic IDEs or Vibe Coding are here to stay. As engineers, we must embrace the usage of AI responsibly. AI will not replace engineers, AI is here to empower us.
As we all know, technology evolves rapidly. It’s our job to adapt quickly to new tools and ways of development. In the near future, I envision all developers having their own AI assistant to be 10x more productive. The best developers will be the ones embracing and integrating AI into their daily work.
