Jeder LLM-API-Call ist ein Clean Slate — 8 Memory-Architekturen im Vergleich: Context Window, RAG, MemGPT, Mem0, Zep, LangGraph und File-basierte Ansätze.
Why the "Desk Tax" is a myth for power users — and what Claude Code actually needs instead of mobile access.
Every LLM API call is a clean slate — 8 memory architectures compared: context windows, RAG, MemGPT, Mem0, Zep, LangGraph, and file-based approaches.
Why Claude Code's idle capacity problem only applies to one user segment — and the security risks nobody's talking about
What changed since January: session-mining, browser feedback loops, parallel instances. Why deeper integration beats adding more tools.
Browser DevTools MCPs, ContextMine, Session-Mining und parallele Instanzen — was sich seit Januar verändert hat und warum tiefere Integration mehr bringt als mehr Tools.
651 commits, 233 documents, zero production code. Three lessons: language-agnostic specs, hybrid validation, and why "done" is a myth.
153 commits, 102 specs, 4 phases — and a realization on day 5 that I'd accidentally invented Spec-Driven Development. An authentic journey.
AI agents generate code fast — but without structured specs, projects drown in tech debt. Here's how Spec-Driven Development fixes that.
Nach drei Monaten und $6.000 API-Kosten — was bleibt: Handover-Patterns, Quality Gates und der Übergang zu Claude Code.
Over-Engineering, Mode Drift, Context-Overflow — echte Probleme und Lösungen aus der Entwicklung eines Multi-Agent-Systems mit 15 KI-Modi.
Wie Handover-Patterns, Quality Gates und eine Rule Hierarchy 15 spezialisierte AI-Modi zu einem orchestrierten Agile Development Team koordinieren.
From 118 commits in three days to a pragmatic pivot — the final chapter of building an AI development team.
Erzwungener Wechsel von JetBrains zu VSCode wurde zum Wendepunkt — 15 spezialisierte AI-Modi, 118 Commits in 3 Tagen, ein selbst-organisierendes Entwicklungsteam.
Over-engineering, mode drift, context overflow — real problems and solutions from building a 15-mode AI development team.
Language Server, Superpowers, Context7 und v2.1.0 Security-Fix — meine essentiellen Tools nach 3 Monaten intensiver Nutzung.
Language servers, Superpowers, Context7, and the v2.1.0 security fix — my essential tools after 3 months of intensive use.
How I built an AI development team that coordinates like a real agile squad — with zero information loss between 15 specialized modes.
A forced VSCode migration sparked the creation of a complete AI development team with 15 specialized modes — 118 commits in 3 days, a self-organizing system.
How metacognition transforms reactive chatbots into adaptive, self-improving AI systems
Warum GPT-3.5 den Bat-Ball-Test versagt, GPT-4 mit Chain-of-Thought aber besteht — und wie TRAP, Reflexion und LATS KI-Agenten das Lernen aus Fehlern beibringen.
