Daniel Krauss, co-founder of Flix (Flixbus), offers rare insights into the real-world AI practices of one of Europe's most ...
What Daniel Krauss said
Written from the verified transcript and checked against it. Every figure links to the moment it was said.
Daniel Krauss, co-founder, CIO and CHRO of Flix SE, discussed the company's approach to agentic coding AI. He said Flix is a fast follower, not a trailblazer, to avoid wasting money, and built its own Flix GPT on AWS Bedrock but only started paying for AI tools like Claude and Copilot about half a year ago. Krauss dismissed claims of 50% productivity gains as 'bullshit' and warned about exploding costs and dependency on US tech monopolies. He emphasized that AI's real value is in supporting the long tail, automating support work, and enabling new revenue, not just cutting costs. He stressed the importance of a clean tech core for scaling, and said vibe coding is short-sighted; instead, people need context, business understanding, critical thinking, and creativity. Krauss argued that focusing only on cost savings leads to dependency on oligopolies, while the real lever is delivering what was previously impossible.
Key takeaways
Flix built its own Flix GPT on AWS Bedrock but only started paying for AI tools like Claude and Copilot about half a year ago.
Claims of 50% productivity gains from AI are 'bullshit'; costs are exploding.
Focusing only on cost cutting leads to dependency on US oligopolies that will eventually raise prices.
So, I'm incredibly pleased that Daniel is here today, one of the co-founders of Flixbus. A real IT professional, especially when it comes to production systems, and always at the forefront. And today on the Soft Factory Insider Podcast, it's actually about what is already possible in practice, and I'm very excited to learn some insights from your universe of Flixbus or Flix. Thank you very much for taking the time, Daniel. I'll just jump right in. I think you don't need a big introduction. How is it for you regarding the whole topic of agentic coding AI? Where do you already stand in practice?
Hello Chris, thanks for the invitation and the kind words. As a tech-enabled travel platform, we're better through technology, but we still transport people from A to B. For us, fast following is better than trailblazing. Our hybrid model means we're not at risk of a SaaS apocalypse. Being first risks wasting money. As a fast follower, we can capture real value. We played with ChatGPT and built our own Flix GPT on AWS Bedrock, but only started paying for and using it for real value creation about half a year ago with Claude and Copilot. Claims of 50% productivity gains are bullshit. Costs are exploding, but we focus on using AI to support the long tail and generate more revenue, not just switch costs from employees to tech monopolies.
Yes, I find it super exciting. You can go deep into the tech, but one fascinating thing from Anthropic is the horizontal approach – enabling changes in Excel, everything computer work. I'd like to ask: is the programmatic depth nice but not decisive? If you look at how much time developers spent on processes like Jira, agent-based support can dramatically shorten the entire development cycle. The coding part is given, but then there's support, documentation, the processual part until something goes live. That's much cooler and it was the annoying part. And the translation at the business interface – despite good Agile and product owners, there are always difficulties articulating ideas. So it's great to go into prepping first instead of running a whole team.
If you look at how much time developers spent on processes like Jira, agent-based support can dramatically shorten the entire development cycle. The coding part is already there, but then you have support, documentation, and the processual part until something goes live. That's much cooler and was the annoying part. At the business interface, despite good Agile and product owners, there are translation difficulties. So it's great to go into prepping first. Individual tools like Labeling are good, but for ordinary business people, Claude is now super because it updates everything. The Excel and PowerPoint stuff is helpful, but what's really helpful is when the copilot has access to data. Instead of laboriously building Power BI or bothering colleagues, you just ask and get instant answers. That automates support work so humans can focus on critical thinking and creativity. That will remain with humans for a while, which is why I want good people.
Super. And one thing that might be interesting – I've always seen Flix as building a solid tech infrastructure from the start, not as an afterthought. You said earlier, if your house is messy, it's hard. So for you, AI must be a layer on top of what you already have clean. You know how many companies don't have that.
I believe we have done that cleanly. That's important when scaling and internationalizing quickly. With a clean core, you can secure it better and focus on APIs and surface layers like apps and websites, scaling arbitrarily with AI. If everything is messy, it won't work. We've done our homework, so we can benefit greatly. Human masterminds focus on core and strategy, while features can be done much faster with the machine.
6 more exchanges in this transcript
Sign in free to read the rest of this interview. No card required.
Krauss, D. (2026, May 5). Software Factory Insider with Flixbus founder Daniel Krauss [Interview transcript]. Chris Strobl - Software Factory Guy. CEOInterviews.AI. https://ceointerviews.ai/interview/1006622/
MLA
Daniel Krauss. "Software Factory Insider with Flixbus founder Daniel Krauss." Chris Strobl - Software Factory Guy, 5 May. 2026. Transcript, CEOInterviews.AI, https://ceointerviews.ai/interview/1006622/.
BibTeX
@misc{krauss2026_1006622,
author = {Daniel Krauss},
title = {Software Factory Insider with Flixbus founder Daniel Krauss},
howpublished = {Interview transcript, Chris Strobl - Software Factory Guy. CEOInterviews.AI},
year = {2026},
month = {may},
url = {https://ceointerviews.ai/interview/1006622/},
note = {Speaker-attributed transcript with timestamps}
}
