How AI Shows Up in Our Daily Work
AI tools are not an experiment in our work anymore. They are part of the normal workflow, for both of us: Sebastian on the technology side, Christian on the strategy side.
What we actually use:
- Claude Code as an interactive development assistant: writing, refactoring, debugging, and explaining code. Not a replacement for thinking, but a fast second pair of eyes.
- Skills (reusable prompt workflows) for tasks that follow the same shape every time. This blog post, for instance, came out of exactly that kind of skill. Once it is defined cleanly, it runs reliably.
- Structured prompts for documentation, commit messages, and test cases.
The result: we ship cleaner code in less time. Not because the AI thinks for us, but because we spend less time on boilerplate.
What's Behind It
How you use AI tools is what matters. Treat a language model like an intern you hand a vague task to, and you get mixed results. Treat it like a well-prepared tool, with clear context, a defined task, and clear rules, and you get reliably good results.
For us that means: every skill has clear rules, a defined input, and a defined output. That is what makes the results reproducible.
Who Benefits
This way of working is not just an internal detail. It carries over directly to client projects:
- Teams that need to ship software: AI-assisted code review and automated documentation save real hours every week.
- Startups at the PoC stage: faster iteration without losing quality is exactly what matters in the early phase.
- Companies with internal tools: even non-technical processes speed up with the right prompt design.
Frequently Asked Questions
Does AI replace real developers at ex-nihilo? No. AI makes us faster as developers, but it does not replace thinking, understanding, or architectural decisions. We are still responsible for every line of code that ends up in production.
Which tools do you actually use? Mostly Claude (Anthropic) through Claude Code as a CLI tool. For certain tasks, other models too. The tool choice is secondary; the workflow and the quality of the prompts make the difference.
Can this be integrated into an existing team? Yes, and it is often easier than in a greenfield project. The biggest hurdle is not the technology, it is the discipline to define prompts cleanly instead of just "quickly asking" for something.
How do you make sure the generated code is correct? Code review stays mandatory. AI makes the first draft faster, but whether it is correct, maintainable, and performant is something we check ourselves. Automated tests help enormously with that.
Do you use AI for client projects too? Yes, wherever it makes sense. Not as a sales pitch, but because it makes the results better. We are transparent about it with our clients.