We consciously chose to forgo quick applause in favor of developing a tool that not only passes the practical test, but excels it.

The tech world has been buzzing about artificial intelligence for quite some time, with many vendors moving quickly to bring features to market. Zammad, however, took its time. What drove that decision?

Martin: We made a deliberate decision not to just ride the wave of hype. The reason is simple. Artificial intelligence should never be an experiment conducted at the expense of our customers.

Instead, we worked quietly behind the scenes for a long time. We built prototypes, pushed them to their limits, learned from the results, and refined them from the ground up. We only put these features into the hands of our beta testers once we were fully convinced. It wasn't until their feedback confirmed that these tools provided tangible value that we officially launched them.

Zammad 7.0 is a milestone—and not just because of the artificial intelligence. What else makes this version special?

Martin: It is the foundation for everything to come. With version 7.0, we’ve accomplished two things that really move the needle for us. First, we’ve developed a technical architecture that seamlessly integrates AI features and is flexible enough to connect to various large language models.

Second, we’ve developed significant internal expertise. This will allow us to roll out new features much faster in the future. This is crucial because artificial intelligence isn't a "one-and-done" feature that you develop and then check off a list. It’s an ongoing, evolving process.

Companies shouldn’t have to give up their valuable data just to access artificial intelligence.

Rewriting subject lines, sorting tickets, and generating summaries sound like small tasks. Why do they make such a big difference for support teams?

Martin: Small tasks add up quickly. Categorizing or skimming a single ticket may seem trivial. But when you’re dealing with high ticket volumes, it creates a massive time sink.

This is exactly where AI steps in: it reduces repetitive tasks and takes a large share of routine work off the team’s plate. The result is not only time saved, but also a noticeable reduction in cognitive load.

What was your top priority from the very beginning when developing these AI features, and what role does the freedom to choose your own Language Model play in that?

Martin: Transparency and data sovereignty. Today, companies are faced with an unfair choice: embrace modern AI or uphold data privacy. This is especially problematic in the support department, where critical information such as customer data, internal workflows, and in-depth product knowledge is stored.

When an AI provider gains access to this data, there is always the risk that it will be used to train models. It's an economic nightmare because the hard-earned knowledge you’ve spent years developing could theoretically become available to your competitors overnight.

We are breaking this cycle. We empower our users to decide which path fits their requirements, whether it's public cloud, private cloud, or a completely local instance. This freedom of choice isn't just a technical detail for us; it’s a fundamental principle. We want to give businesses control back over their most valuable asset.

Algorithms should never have the final say when it comes to fairness and empathy toward the person on the other end.

Where should AI step back and the human take over to ensure service quality doesn’t suffer?

Martin: There is no one-size-fits-all answer to that. We are currently witnessing a completely new technology evolving at breakneck speed. What seems impossible today might be the industry standard tomorrow.

However, we draw a very clear line at manipulation and systematic discrimination. A critical issue here is AI bias. Algorithms learn from historical data and can adopt the prejudices embedded within it.

Take linguistic bias, for example: many AI models are trained on perfect, formal German. When it comes to regional dialects or non-native phrasing, their precision often drops. If the artificial intelligence then systematically flags inquiries from non-native speakers as "less urgent," the principle of fairness has been abandoned.

Another example is sentiment bias: if the AI prioritizes only the "loudest" customers while calm, factual descriptions get stuck in the queue, that’s a problem. In cases like these, you will always need a human being to make decisions rooted in responsibility and empathy.

We are demonstrating that there is no place for the “black box” of AI in the world of open source.

When we look back on the launch of Zammad 7.0 in two years, what influence will AI have had on Zammad’s identity?

Martin: Hopefully, it's the other way around! :) Our DNA remains the same: open source, transparency, and control. With Release 7.0, however, we are demonstrating that these values also work in the world of AI. Zammad will become an even stronger lighthouse project in the open-source space.

While many AI systems today are "black boxes," we are proving that open source and artificial intelligence are not mutually exclusive. I believe that is our unique selling point.

Our toolkit is much more powerful now, yet we remain the reliable, open system our users trust.