<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"><channel><title>rubeen.dev</title><description>A blog about software engineering, AI systems and technology</description><link>https://rubeen.dev/</link><language>en</language><item><title>Chat deleted. So why does the agent still know?</title><link>https://rubeen.dev/en/blog/memory-datenarchitektur/</link><guid isPermaLink="true">https://rubeen.dev/en/blog/memory-datenarchitektur/</guid><description>Agent memory makes copies of a single message, and “delete chat” often reaches only the chat. A small lab shows where a diagnosis survives deletion and which deletion path removes it.

Made with AI: The text and images of this article were created with generative AI. This version was translated from the German original with generative AI. Editorial responsibility: Ruben Vitt.</description><pubDate>Sun, 08 Nov 2026 00:00:00 GMT</pubDate></item><item><title>Who Grades the Grader?</title><link>https://rubeen.dev/en/blog/wer-prueft-den-grader/</link><guid isPermaLink="true">https://rubeen.dev/en/blog/wer-prueft-den-grader/</guid><description>A grader is a system in its own right, and it needs testing. I had deliberate errors planted in correct release notes and set eleven graders on them. In the end, even the label was wrong.

Made with AI: The text and images of this article were created with generative AI. This version was translated from the German original with generative AI. Editorial responsibility: Ruben Vitt.</description><pubDate>Sun, 01 Nov 2026 00:00:00 GMT</pubDate></item><item><title>Building Was the Easy Part: What Half a Year of Running Volunteer Software Taught Me</title><link>https://rubeen.dev/en/blog/ehrenamt-software-betreiben/</link><guid isPermaLink="true">https://rubeen.dev/en/blog/ehrenamt-software-betreiben/</guid><description>Three evenings, three tools - and then what? How nine small apps for our volunteer unit became one, why a green health check noticed nothing, and what I do when I&apos;m on vacation.

Made with AI: The text and images of this article were created with generative AI. This version was translated from the German original with generative AI. Editorial responsibility: Ruben Vitt.</description><pubDate>Sat, 24 Oct 2026 00:00:00 GMT</pubDate></item><item><title>More than a lockfile: the Agent Contract for LLM features</title><link>https://rubeen.dev/en/blog/agent-contract/</link><guid isPermaLink="true">https://rubeen.dev/en/blog/agent-contract/</guid><description>An LLM feature depends on five things: output schema, prompt, model, tools and eval set. If you only version the prompt, you don&apos;t know what you deployed.

Made with AI: The text and images of this article were created with generative AI. This version was translated from the German original with generative AI. Editorial responsibility: Ruben Vitt.</description><pubDate>Sat, 10 Oct 2026 00:00:00 GMT</pubDate></item><item><title>A Prompt in the Wiki Is Not Evidence: Audit-Proof Logging and the Agent Contract in Regulated Environments</title><link>https://rubeen.dev/en/blog/revisionssicheres-logging-agent-contract/</link><guid isPermaLink="true">https://rubeen.dev/en/blog/revisionssicheres-logging-agent-contract/</guid><description>The Agent Contract is the intended state, the audit-proof log is the proof of what actually happened. Without the second, the first is just a claim. What DORA, GoBD and the upcoming MaRisk demand right now.

Made with AI: The text and images of this article were created with generative AI. This version was translated from the German original with generative AI. Editorial responsibility: Ruben Vitt.</description><pubDate>Mon, 06 Jul 2026 00:00:00 GMT</pubDate></item><item><title>The Best Answer Is the Wrong Target</title><link>https://rubeen.dev/en/blog/die-beste-antwort-ist-keine-zielgroesse/</link><guid isPermaLink="true">https://rubeen.dev/en/blog/die-beste-antwort-ist-keine-zielgroesse/</guid><description>An eval score that only measures quality rewards brute force: more sampling, a bigger model, more retries. Why the right basis for a decision is a cost-quality-latency frontier, a curve instead of a point, and how to pick the operating point in dollars instead of percent.

Made with AI: The text and images of this article were created with generative AI. This version was translated from the German original with generative AI. Editorial responsibility: Ruben Vitt.</description><pubDate>Sat, 27 Jun 2026 00:00:00 GMT</pubDate></item><item><title>Chat is the interface that came first - not the one that works</title><link>https://rubeen.dev/en/blog/die-bessere-ui-ist-kein-chat/</link><guid isPermaLink="true">https://rubeen.dev/en/blog/die-bessere-ui-ist-kein-chat/</guid><description>As soon as an agent does something with consequences, chat is rarely the best interface. A look at the editor, diff, approval panel and timeline - and when a text box is still the right tool.

Made with AI: The text and images of this article were created with generative AI. This version was translated from the German original with generative AI. Editorial responsibility: Ruben Vitt.</description><pubDate>Thu, 11 Jun 2026 00:00:00 GMT</pubDate></item><item><title>The Model Isn&apos;t the Risk: Permissions, Tenant Isolation and the Access Surface of Agents</title><link>https://rubeen.dev/en/blog/nicht-das-modell-ist-das-risiko/</link><guid isPermaLink="true">https://rubeen.dev/en/blog/nicht-das-modell-ist-das-risiko/</guid><description>Permissions, tenant isolation, overly broad tools, sloppy context access: why an agent&apos;s reach determines the damage - and how to contain it.

Made with AI: The text and images of this article were created with generative AI. This version was translated from the German original with generative AI. Editorial responsibility: Ruben Vitt.</description><pubDate>Wed, 10 Jun 2026 00:00:00 GMT</pubDate></item><item><title>Give the AI the Decision, Not the Control: Why Free-Form Agent Loops Are Often Too Open</title><link>https://rubeen.dev/en/blog/agent-loops-entscheidungspunkte/</link><guid isPermaLink="true">https://rubeen.dev/en/blog/agent-loops-entscheidungspunkte/</guid><description>Autonomy is a dial, not a switch. Why most production AI systems get better when the model only sits at clearly bounded decision points - instead of steering an open agent loop.

Made with AI: The text and images of this article were created with generative AI. This version was translated from the German original with generative AI. Editorial responsibility: Ruben Vitt.</description><pubDate>Fri, 29 May 2026 00:00:00 GMT</pubDate></item><item><title>Phoenix Progression: Sprints Need Ashes</title><link>https://rubeen.dev/en/blog/phoenix-progression/</link><guid isPermaLink="true">https://rubeen.dev/en/blog/phoenix-progression/</guid><description>My own productivity system built from the 12 Week Year, the 7 Habits and a cycles framework - with a real ash week instead of just a buffer break. And an honest look at where it breaks.

Made with AI: The text and images of this article were created with generative AI. This version was translated from the German original with generative AI. Editorial responsibility: Ruben Vitt.</description><pubDate>Fri, 22 May 2026 00:00:00 GMT</pubDate></item><item><title>From Incident to Regression Test: How Production Failures Make Their Way into Your Eval Set</title><link>https://rubeen.dev/en/blog/incident-zu-regressionstest/</link><guid isPermaLink="true">https://rubeen.dev/en/blog/incident-zu-regressionstest/</guid><description>An eval suite that doesn&apos;t grow with your incidents shrinks relative to reality. What the postmortem-to-eval loop looks like in practice, from the trace ID to the required field in the postmortem template.

Made with AI: The text and images of this article were created with generative AI. This version was translated from the German original with generative AI. Editorial responsibility: Ruben Vitt.</description><pubDate>Wed, 13 May 2026 00:00:00 GMT</pubDate></item><item><title>Release Discipline for AI Features: Canary, Shadow Runs and Automatic Rollback</title><link>https://rubeen.dev/en/blog/ai-rollout-canary-shadow/</link><guid isPermaLink="true">https://rubeen.dev/en/blog/ai-rollout-canary-shadow/</guid><description>Green eval in the PR, red numbers in production two weeks later - without a new deploy. Why AI features need their own gates between merge and 100% traffic.

Made with AI: The text and images of this article were created with generative AI. This version was translated from the German original with generative AI. Editorial responsibility: Ruben Vitt.</description><pubDate>Tue, 28 Apr 2026 00:00:00 GMT</pubDate></item><item><title>Observability for AI features: which spans, events and IDs you actually need</title><link>https://rubeen.dev/en/blog/observability-ai-features/</link><guid isPermaLink="true">https://rubeen.dev/en/blog/observability-ai-features/</guid><description>Without traces, model and prompt versions, tool spans and cost metrics, every incident stays guesswork. A guide for teams running AI features in production.

Made with AI: The text and images of this article were created with generative AI. This version was translated from the German original with generative AI. Editorial responsibility: Ruben Vitt.</description><pubDate>Tue, 21 Apr 2026 00:00:00 GMT</pubDate></item><item><title>Collecting Knowledge Was Never the Problem: Mycel, Thinking Tools and the New Division of Labor in Thinking</title><link>https://rubeen.dev/en/blog/wissen-sammeln-war-nie-das-problem/</link><guid isPermaLink="true">https://rubeen.dev/en/blog/wissen-sammeln-war-nie-das-problem/</guid><description>Classic PKM fails at maintenance. An LLM-maintained wiki fixes that - and turns thinking tools from a book on the shelf into a skill on demand.

Made with AI: The text and images of this article were created with generative AI. This version was translated from the German original with generative AI. Editorial responsibility: Ruben Vitt.</description><pubDate>Wed, 15 Apr 2026 00:00:00 GMT</pubDate></item><item><title>Three Evenings, Three Tools: When Software for Volunteers Suddenly Becomes Doable</title><link>https://rubeen.dev/en/blog/drei-abende-drei-tools/</link><guid isPermaLink="true">https://rubeen.dev/en/blog/drei-abende-drei-tools/</guid><description>Volunteer organizations need their own software, but they could never afford it. How AI is changing that right now, and why deployment knowledge matters more than knowing how to code.

Made with AI: The text and images of this article were created with generative AI. This version was translated from the German original with generative AI. Editorial responsibility: Ruben Vitt.</description><pubDate>Sun, 12 Apr 2026 00:00:00 GMT</pubDate></item><item><title>Approval Isn&apos;t a UX Bug: Why Agent Systems Need to Know When to Ask</title><link>https://rubeen.dev/en/blog/human-in-the-loop/</link><guid isPermaLink="true">https://rubeen.dev/en/blog/human-in-the-loop/</guid><description>Human in the loop isn&apos;t a weakness, it&apos;s security architecture. Why pause/resume for critical agent actions may be the most important feature of all.

Made with AI: The text and images of this article were created with generative AI. This version was translated from the German original with generative AI. Editorial responsibility: Ruben Vitt.</description><pubDate>Mon, 06 Apr 2026 00:00:00 GMT</pubDate></item><item><title>Paper Doesn&apos;t Fly: How a PWA Replaces the Drone Checklist in Civil Protection</title><link>https://rubeen.dev/en/blog/papier-fliegt-nicht/</link><guid isPermaLink="true">https://rubeen.dev/en/blog/papier-fliegt-nicht/</guid><description>How I used AI to build an offline PWA for drone operations in civil protection - and why domain knowledge matters more than prompts.

Made with AI: The text and images of this article were created with generative AI. This version was translated from the German original with generative AI. Editorial responsibility: Ruben Vitt.</description><pubDate>Tue, 31 Mar 2026 00:00:00 GMT</pubDate></item><item><title>AI Tools for My Own Blog: Podcast Pipeline, Copy Editing and the Skills Behind Them</title><link>https://rubeen.dev/en/blog/ai-werkzeuge-eigener-blog/</link><guid isPermaLink="true">https://rubeen.dev/en/blog/ai-werkzeuge-eigener-blog/</guid><description>How my blog applies the same AI principles I described in my last three posts - with automated podcast generation and AI-assisted copy editing.

Made with AI: The text and images of this article were created with generative AI. This version was translated from the German original with generative AI. Editorial responsibility: Ruben Vitt.</description><pubDate>Sun, 29 Mar 2026 00:00:00 GMT</pubDate></item><item><title>Tools Aren&apos;t Prompts: Why Agent Actions Need Idempotency, Auth, and Audit</title><link>https://rubeen.dev/en/blog/tools-sind-keine-prompts/</link><guid isPermaLink="true">https://rubeen.dev/en/blog/tools-sind-keine-prompts/</guid><description>A tool call can move money and change accounts. Why idempotency, auth, and audit in agent architectures aren&apos;t a nice extra but a hard requirement.

Made with AI: The text and images of this article were created with generative AI. This version was translated from the German original with generative AI. Editorial responsibility: Ruben Vitt.</description><pubDate>Fri, 27 Mar 2026 00:00:00 GMT</pubDate></item><item><title>Welcome to My New Blog</title><link>https://rubeen.dev/en/blog/willkommen/</link><guid isPermaLink="true">https://rubeen.dev/en/blog/willkommen/</guid><description>Why I moved from Ghost to a self-built Astro blog, and what this blog can do.

Made with AI: The text and images of this article were created with generative AI. This version was translated from the German original with generative AI. Editorial responsibility: Ruben Vitt.</description><pubDate>Wed, 25 Mar 2026 00:00:00 GMT</pubDate></item><item><title>Evals in CI: When AI features stop being prompts</title><link>https://rubeen.dev/en/blog/ai-evals-ci-pipeline/</link><guid isPermaLink="true">https://rubeen.dev/en/blog/ai-evals-ci-pipeline/</guid><description>... and start being software - with all the testing obligations that come with it

Made with AI: The text and images of this article were created with generative AI. This version was translated from the German original with generative AI. Editorial responsibility: Ruben Vitt.</description><pubDate>Mon, 23 Mar 2026 00:00:00 GMT</pubDate></item><item><title>AI Systems Architecture Starts Where Model Output Becomes Part of the Product Logic</title><link>https://rubeen.dev/en/blog/ai-systems-architecture/</link><guid isPermaLink="true">https://rubeen.dev/en/blog/ai-systems-architecture/</guid><description>How I turn a GitHub issue into an implementation plan that&apos;s testable and observable

Made with AI: The text and images of this article were created with generative AI. This version was translated from the German original with generative AI. Editorial responsibility: Ruben Vitt.</description><pubDate>Thu, 19 Mar 2026 00:00:00 GMT</pubDate></item></channel></rss>