Introduction
The day after tomorrow (03.06.2026.) we are celebrating 10 years in business 🎉🎂, and oh boy, in these 10 years the product code has grown a lot. It is a time to make a pause, a break, and clean it up. Technology has changed a lot in these 10 years. Not only Java got a lot of new features but AI is now more and more involved in development. Without adopting and accepting new technologies, we will not be able to compete. Thus, we started to test and experiment with AI as a coding companion. Mostly for code analysis, refactoring and boilerplate source generation as a first step.
AI code refactoring
For a very long time, there were tools such as static code analysis used to analyze source code to find issues. Very helpful, but not perfect. Then AI came as a development support. So, we tried it for the last couple of months.
Initially, we used AI primarily for generating boilerplate code or to refactor POJO and Beans into Java Records classes. We did not make such a transition before as it is time consuming hard manual work. There was always something else, more urgent to do. AI was of great help here. Very precise source refactoring that saved us a lot of time.
Next step was to try code analysis, which brings another set of problems…. intellectual property protection. We did not want all parts of the code to be exposed to the public cloud based AI services, so we were limited only to some general libraries and modules.
Was it helpful? Well…sort of, yes. The main issue we found out is that no matter how AI helped us in optimization and refactoring, we still needed to do a code review. Before the AI era, we would write boilerplate code manually, or we would do a lot of (even unnecessary) testing. Now, we have to review what AI prepared for us.
Sometimes, the code is very good, sometimes, there are mistakes. And logical mistakes are hardest to spot. The good part is that we can always ask the AI for an explanation why he did it in a specific way. And, yes, sometimes AI surprises us, we even learn something. But, still, it turns out, it is still consuming the same amount of time, if not more in some cases. The bottom line, it is a helpful tool but still with some limitations.
How it saves us time in classical programming, AI brings another level of “time wasting”. It’s a trade of.
Another problem mentioned is intellectual property protection. To fully utilize AI for full source code, we have to switch to offline local AI. Well, it turns out that a single machine which can process the data costs in a range 6000€-12.0000€ depending on the processing power and size of an AI model. The bigger the model, the smarter it gets.
The general idea is being able to load a full AI model into a memory. And this is not RAM. Don't get confused or mislead with the statements as "next generation processor with integrated AI". The most of today processors with integrated AI are for small helper tasks. For real AI processing, we need some serious power, and dedicated hardware, either NVIDIA RTX or Apple M5 Max with unified memory. The topic itself is too broad to explain everything here.
The bottom line, we came to a conclusion that investment is not justified right now, technology is not yet fully matured. We will have to be satisfied with cloud AI and parts of the code we can share to the public AI.
The result of using AI in project analysis and code optimization in general was very helpful. AI managed to detect some issues standard static code analysis will never be able to detect. AI optimized some parts of generic code related to stream processing, and fully managed to refactor Java POJO into Java Record classes. The result is more stable, more performant and cleaner code.
And where does this lead us with our product...?
What is coming in GS-2026.03
Future releases from GSv5-2026.Q3 and GSv6-2026.Q3 and onward will have security updates and performance optimizations made with help of AI.
Some new incoming features made with the help of AI will bring fully integrated a brand new Firewall Engine (replacement for IP ACL engine) and a brand new API Key ACL Both of them deeply integrated into a server runtime giving users very detailed access control on interface level (firewall) and resource level (API Keys).
Now, both new modules will be independent from the product itself and will operate as independent services with its own REST services and management API.
The Firewall engine will differentiate between PROXY channels and regular ones, to have separate IP rules. While with API Key ACL, any URI service can be protected in a generic way.
Another optimization is improved startup and server runtime auto-configuration on first install. More robust with better reliability. In future updates post GS-2026.Q3 we will add REST services for managing server runtime with Firewall IP rules and API Key access control.
Another planned feature is to add support for email reporting and integration with the web2spool engine for automatic spool emailing. This feature is planned for 2026.Q4.
Among all mentioned, we will continue with internal refactoring and improvements with AI help.