Start AI assessment Software development is reinventing itself. The future no longer belongs exclusively to those who know syntax by heart, but to those who can explain domain problems simply, precisely, and with plenty of context.
We just proved it ourselves: we completed a full relaunch of our website without involving a single software developer.
The problem: The old way was a marathon
Previously – and by that we mean our very last web project – the process was classic, slow, and resource-intensive:
- The process: First conceptual planning, then content creation, then image sourcing, followed by separate frontend and backend development.
- The effort: Concept and content took months. The technical implementation tied up three people for nearly three months.
- The result: An almost year-long project that cost a lot of coordination and nerves.
The solution: “Vibe coding” in a weekend session
This time we flipped the script. Instead of writing tickets for developers, we used AI to build the software directly.
The result: within 24 hours we had a first, fully functional website including deployment. What used to be a year-long project became a small weekend session.
Our tools:
- Engine: We used Codex 5.3 in Opencode.
- Method: We followed the setup workflow by Peter Steinberger.
- Tech stack: The site is built on Astro and TailwindCSS v4, hosted via Cloudflare Workers – modern, fast, and static.
How does this work without developers? Context is architecture!
The key to success was not writing code, but managing context. Our developers were fully occupied with LIVOI development – so we had to give the AI the “memory” that normally lives inside a programmer’s head.
We followed one principle: build your project not for humans, but for a model that needs to navigate.
- Global memory: A
docs/folder and anAGENTS.MDserved as the instruction manual for the AI. - Structure over cleverness: We did not tell the AI how to code, but what the domain goal was and which rules applied.
- Scaling: The project now spans thousands of lines of code – thanks to clean context structure (docs updates instead of guessing), the AI stayed on track throughout.
Conclusion: A paradigm shift
This relaunch showed us: anyone who can describe a domain problem in detail is now able to build their own software solution to a certain degree – entirely without a traditional developer. Peter Steinberger, whose workflow we referenced above, takes this even further with his project OpenClaw – we explain why he stopped reading code altogether in our post on Agentic Engineering.
Of course we chose this path because it was a static website (where this approach is unbeatable in efficiency) and not a highly complex backend system. But for projects of this kind, the rules of the game are changing radically.
What do you think? Would you have expected the time advantage to be this extreme? Get in touch and share your experiences with AI-assisted development!