Why Do We Focus on Greenfield Projects with AI?
Greenfield projects provide a unique opportunity to integrate generative AI into the process right from the design phase. The key to AI’s effectiveness lies in how well it knows the technologies being used. If we choose tools and programming languages that the AI has been trained on with large amounts of data, it can generate not only faster but also higher-quality code.
This new approach allows us to produce even hundreds of lines of complex code in a single iteration. Although the generated code is not always perfect on the first try, after fine-tuning it adheres to clean code principles and meets the technical requirements of the target platform. Our experience shows that 90% of the generated code is immediately usable, and the remaining 10% can be easily integrated with minor adjustments.
Real Project Example: Development Across Three Platforms
One of our ongoing projects perfectly demonstrates how we apply generative AI in the development of various technologies. The frontend is built on React, while the backend uses the Java Spring framework with a Postgres database. The Android client is developed in Kotlin, and in each of these technologies we experience different benefits thanks to AI.
AI enables faster creation of complex components on the frontend, such as UI elements and state management logic. In backend development, it proves particularly effective in optimizing database queries—AI-generated native SQL queries can efficiently join even ten tables, resulting in significant performance improvements.
For the Android client, AI automates routine coding tasks such as assembling data models and handling API calls, freeing developers’ time and energy for solving more complex problems.
Challenges and Lessons Learned from AI Support
Although AI provides tremendous assistance, its use also comes with challenges. One such difficulty is inaccuracy and the so-called "hallucination," which occurs particularly in narrow domains where little training data is available. For this reason, generated code must always be thoroughly reviewed, just as if it had been written by a junior developer.
Another obstacle is that AI does not know our proprietary, internal libraries, meaning their integration requires manual intervention. In our initial projects, we consciously aimed to use widely available community standards to make collaboration with AI smoother.
In addition, learning to use AI also involves a learning curve for developers, as it requires a new mindset and toolset, which takes time for teams to adopt effectively.
Faster Development, Higher Quality – The Results of AI
The greatest advantage of AI-supported development lies in significantly faster workflows without compromising code quality. While complex and creative tasks still require experienced senior developers, AI accelerates routine tasks and helps solve complex technical problems.
AI-supported projects stand out not only for their speed but also for their results: optimized database queries, efficient algorithms, and clean code all contribute to delivering higher-quality outcomes for our clients.