What Inspired Us
The Spark
In early 2026, OpenClaw exploded onto the scene – an open-source AI agent built by a single developer that collected over 100,000 GitHub stars within weeks. An autonomous agent that improves itself, writes its own skills and can be controlled via messaging platforms.
Our first reaction: Impressive. Our second: Why don't we build something truly useful – not a toy, but a production tool for real software development?
Nate B. Jones – The Catalyst
If you want to understand where AI-driven development is heading, you can't ignore Nate B. Jones. His YouTube channel is one of the best sources for what's actually happening – no hype pieces, just well-founded first-hand analysis. From Anthropic's development velocity to practical agent workflows to the strategic implications for the entire industry.
Nate popularised the concept of the five autonomy levels – from "spicy autocomplete" to the fully autonomous dark factory. This framework helped us determine where we stand and understand where we need to go.
Thank you, Nate. Without your content, we probably would never have started OpsFabrik.
Claude Code Company-Wide
In early February 2026, we made a radical move at Provitex: complete switch to Claude Code with human-in-the-loop development. No "let's try it out", no pilot project – all developers, all projects, immediately.
The workflow: Claude Code writes and edits code in one terminal window. In a second window, DDEV, Docker and infrastructure tools run. The human coordinates, reviews and decides. We call this the two-window workflow internally.
The results were immediate. Not because the AI writes perfect code – it doesn't – but because the feedback loop between human and machine is so tight that errors are caught quickly and decisions are made faster.
The Provitex Brain
In parallel, we introduced the provitex-brain – a company-wide, Git-versioned knowledge base that serves as the operational memory for all projects.
The core idea: every project starts with the accumulated knowledge of all previous projects. At the end of each project, new knowledge is extracted and fed back. A cycle that compounds with every project.
The brain contains:
- Standards – mandatory company rules (code style, security, hosting)
- Patterns – proven solution patterns from real projects
- Anti-Patterns – documented mistakes you only need to make once
- Blueprints – complete project starters with agent definitions
- Skills – slash commands for recurring tasks (
/commit,/new-project,/brain-inject)
Claude Code automatically receives the relevant context from the brain via CLAUDE.md files. Every agent knows our standards, our patterns and our conventions – from the very first prompt.
The compound effect is real: every project makes the next one faster and better.
The Dark Factory Video
Then we watched this video. The concept of the dark factory – a software factory where no human writes, reads or reviews code. Specifications in, tested software out.
Not as science fiction, but as a working prototype.
That was the moment inspiration became a plan. We already had Claude Code in production, a functioning brain system and the know-how from years of CI/CD work. What was missing was the final step: an agent that runs the entire loop autonomously.
That's how the idea for OpsFabrik was born – and for Darkfab, the AI agent at its centre.