BärGPT Launches: A New AI Assistant for the Berlin State Administration. While we have received very positive feedback from employees, public discussion has also raised critical questions: Why develop an in-house solution when other German states already offer similar tools? An analysis by Ingo Hinterding, Head of Development at CityLAB Berlin.
The development of AI applications for public administration is booming – and that’s a good thing. Whether generative text assistants, specialized analytical tools, or automated document processing, the possibilities are vast, and many dedicated teams are working to make public administration more digital and efficient. This innovation pressure is a sign of a disruptive technology that is fundamentally transforming workflows.
At the same time, the debate around reusing software in public administration is not new – and it is an important one. The principle of “one for all” saves resources, avoids duplicated work, and allows organizations to benefit from the experience of others. In a federal country like Germany, where 16 states and thousands of municipalities maintain their own administrative structures, the potential is enormous. No one should have to reinvent the wheel when functioning solutions already exist.
The Reality is more complex
In practice, however, reuse only works if the available solutions are accessible, high-quality, and adaptable. And this is exactly where it gets complicated. When we began evaluating AI solutions for the Berlin administration in spring 2025, we studied existing offerings from other states extensively. We tested these systems, conducted user tests with Berlin administration employees, and analyzed the results.
The conclusion was sobering: the solutions we tested did not meet our requirements. They were either functionally limited, cumbersome to use, or technically not on par with our own initial prototypes. Feedback from user tests was consistently negative – hardly a solid foundation for a tool intended for daily use by tens of thousands of employees.
Practical hurdles also arose: many offerings were simply not available – either because they were still in early development stages or because cross-state collaboration processes were missing. Some systems were proprietary and not open source; others were far beyond what we could reasonably fund. And even where there was genuine interest, adopting an existing product was far from straightforward and could take months or even years.
Digital Sovereignty Means the Ability to Shape
There is much discussion today about digital sovereignty, but opinions differ on what that actually means. For us, sovereignty is more than the question of whether an OpenAI server is located in the EU or the USA. It means the ability to design, operate, and further develop technology independently and to respond quickly and directly to new needs and requirements. Relying solely on external solutions creates dependency on conditions beyond one’s control: pricing, development priorities, and availability that may not align with one’s own needs.
After discussions with other states, it became clear that reuse would at best have made us a second-class user. Development would have been guided by the needs of the originating state, and our specific requirements would have been secondary. This is not a viable basis for a pilot project that must operate in an agile, user-centered way.
Competition Drives Innovation
It is worth approaching the debate around reuse more nuancedly. Of course, it makes sense to examine and adopt existing solutions where appropriate. But it is short-sighted to categorically demand that no new offerings should be developed – especially in a rapidly evolving field like artificial intelligence.
A look at other areas shows that competition fosters variety, innovation, and better products. No one would seriously argue that we need only one operating system, one browser, or one office suite. iOS and Android, Windows, macOS and Linux, Chrome, Firefox and Safari – all coexist and drive each other to improve. Users benefit from this variety through more choices, fair prices, and continuous development.
Why should AI assistants be any different? Especially in a phase of rapid development, it is crucial that multiple approaches are tested, compared, and refined. What starts as a prototype today may become a standard tomorrow – or disappear from the market because another solution performs better. This is a normal and healthy process.
Specialization Instead of a One-Size-Fits-All Solution
Another key point: we must let go of the idea that all AI-related applications need to converge into a single, consolidated product. A tool that does everything usually does nothing particularly well. Just as you cannot edit videos in Microsoft Excel or create PowerPoint presentations in Outlook, we should think similarly about AI applications.
Just because a large language model is used does not mean that every use case should be served with the same software. Artificial intelligence is rightly a hyped topic, but ultimately it is just another component in the development of user-centered applications. A product uses AI; AI is not the product. We need to focus on software requirements, not just on which technical components are employed.
The future lies in specialization: moving away from general-purpose chatbots toward solutions that address highly specific AI use cases. An AI assistant for legal document analysis requires different functionality than a tool for automating citizen inquiries. Dedicated offerings are needed for this – not a single product that tries to do a bit of everything.
The Best of Both Worlds
The solution is not either/or, but constructive collaboration. Reuse makes sense when products are easily accessible, high-quality, and adaptable to specific needs. In the case of BärGPT, we concluded after careful consideration that developing our own solution would yield a faster, cheaper, and higher-quality result than any reuse option – and hindsight shows this assessment was correct.
The openness and interoperability of our products remain a priority. Like all our developments, BärGPT is open source and thoroughly documented so that others can benefit from our work – and we from theirs.
Conclusion
Reusing an existing solution can often be a good option to reduce effort and avoid unnecessary duplication. However, especially in a young field like AI, it should not stifle innovation at its root. The development of administrative AI is still in its early stages, and there is much to learn. In such a phase, multiple solutions emerging in parallel is not a problem – it is a sign of healthy competition that fosters innovation and improves quality. You may not always need to reinvent the wheel, but sometimes it’s worth turning it further.
