It is increasingly common to hear that anyone can code apps now with the support of artificial intelligence, even without any prior coding knowledge. We tested instead of doubting and guessing! The Codecool Research Lab checked what a beginner can do with AI and how the same technology can boost the work of professionals. We’ve also found the best AI tool – here are the results!
Why is it important to see clearly?
Have you heard of “vibe coding”? It’s a relatively new trend where people with no programming experience develop apps using AI. Until now, there has been a lot of pressure on developers to get to grips with AI, which has caused constant anxiety for many. And as the technology evolves, this is only getting worse.
At Codecool, we’re constantly testing the latest AI solutions because we train developers who use all new technologies wisely. We were curious to see which AI deserves its place in our curriculum and whether it is possible to create an application that works with AI alone, without any developer knowledge. So, we tested four promising agentic AI tools with our graduate students: Windsurf, Cursor AI, Google IDX, and Microsoft Copilot.
Each has the potential to be the best assistant for programmers.
But which one is the best?
How the research was conducted
- Some teams played the “naive” role (as if they didn’t understand programming), while others worked as “experts” (they could steer the AI in the right direction if it got stuck)
- They had a whole week to complete the task, but most finished much sooner
- Our mentor kept collecting students’ experiences with each tool
- Participants wrote detailed reports and uploaded the generated code to GitHub
What did AIs have to create?
The task was to develop a simple question-and-answer system. The goal was to allow users to:
- Log in
- Ask new questions
- Answer existing questions
- And of course store the data in a database
The technology stack was predefined: the backend was built on Spring Boot with a REST API and a relational database, while the frontend was React-based.
Results
All the tools tested (Windsurf, Cursor, Google IDX, Copilot) handled the simpler, standard tasks smoothly and quickly sketched both the backend and frontend projects.
But as soon as more complex security settings, custom architecture, or multi-stage error correction was required, the situation changed:
- Naive prompting (when we used AI as if we didn’t know how to code) often led to endless rounds of bug fixing, and the result was rarely fully functional.
- However, with expert prompting, each tool significantly accelerated development, even to the level of a working prototype.
It was interesting to see that the code completion features of some tools, although useful, could only suggest mostly template solutions for serious problems (e.g., dependency issues or version conflicts). In such cases, the developer had to take the lead.
What did we learn from the first Research Lab?
Our experiments have confirmed that AI assistants can indeed be a valuable aid to development. However, developer expertise and precise, multi-step communication are still essential for more complex tasks.
Of the tools we tested, we currently recommend that our students use the Windsurf Editor in our curriculum, as it seemed the easiest to use in our tests. Its trial version allowed experimentation, and its solutions were quite good. The situation may change soon, as these tools are developing very rapidly and new competitors are constantly entering the market. Have you tried using any AI tools for coding? If you want to learn how to use them effectively, sign up for our full-stack training!
Artificial intelligence is fundamentally changing the way we develop and work today.
Interested in IT but still looking for your place? Check our digital courses in Hungarian!