Dr. Kai Stalmann

Writing

Seven pieces from the first quarter of 2026, written while building fusion. They read best in the order of their numbers, from 01 up: two on how Claude Code works and how to extend it, one on why agile breaks when AI writes the code, the model that replaces it, its conceptual foundation, an essay on what the tools do to the people who use them, and a report from the implementation. New in November 2026: Apply Claude, a guide to using Claude, also available as a PDF.

  1. Apply Claude12 November 2026

    Fundamentals, fields of application, and working models. A comprehensive guide to using Claude: the basics, areas of application, and ways of working, from the chat interface to Cowork and Claude Code. For readers with prior knowledge and for beginners. · Download PDF

  2. Implementing Fusion: A Process Layer for AI-Driven Development10 March 2026

    What the tool does, why memory belongs in the filesystem, and what building a process layer on top of Claude Code taught me.

  3. The Maker's Paradox24 February 2026

    The tool reshapes the toolmaker long before the toolmaker notices. On the productivity trap of tireless machines, and why the people with the most craft are the most susceptible. · German version

  4. Fusion: A Process Model for AI-Driven Software Development15 February 2026

    Projects revolve around three poles: Goal, Decisions, Artifact. A Cycle moves from Goal to Closure through Turns of human-AI work, Coherence Review and Rebalance. Progress is measured by decision growth, not completed tasks.

  5. Foundation: Konzeptionelle Grundlage für AI-gestützte Projektarbeit17 February 2026 · Deutsch

    Das Modell hinter Fusion, ausgehend vom hermeneutischen Zirkel: Grounding, Directive und Discourse als Grundstruktur jeder AI-gestützten Projektarbeit, spezialisierbar in Profilen für Software, Forschung und Schreiben.

  6. Beyond Agile: Project Management for AI-Driven Development6 February 2026

    Agile was built for human-written code. When AI implements features in minutes, implementation stops being the bottleneck and decisions become it. Story points and velocity lose their meaning; goals, decisions and sessions take their place.

  7. Extending Claude Code29 January 2026

    Five extensibility mechanisms, with instructions: hooks on lifecycle events, plugins as distributable bundles, skills Claude activates on its own, slash commands, and agents, plus MCP for external services.

  8. Understanding Claude Code29 January 2026

    Claude Code's internal architecture, inferred from the traces it leaves on disk: stateless context replay, session files, compaction, on-demand discovery, and a multi-agent pattern that delegates almost half the work to cheaper models.