„Wenn es ab morgen kein Claude Code mehr gäbe, mach ich den Laptop zu und geh nach Hause"

Six weeks. That's how long Marco has been working with Obsidian and Claude Code—and he says: It's the best phase of his life. In this episode, Marco and Manuel speak openly about the new pace (a factor of 10? More like more), about genuine dependency, and about the one realization that changes everything: It's no longer about the best output, but about the best input. In this episode: ✅ United Domains is now selling AI telephony assistants with a single click—is that the end of the story? ✅ Why off-the-shelf AI telephony rarely works without process consulting ✅ "What if someone pulls the plug tomorrow?" — How dependent we really are ✅ The 6-week Obsidian journey: from best output to best input ✅ Iteratively improving skills and orchestrating them via master skills ✅ The AI ​​paradox: Why two AIs get worse, not better, in continuous dialogue ✅ Building two worlds: what's universally applicable — and what's only for individual clients 🎙️ This is the AI ​​snack - less talk, more action. Quote from the episode: "It's no longer about getting the best output — but about using that output to generate the best input, so your skillset keeps improving." AI tip: Give back your final result. If you change just three words in an AI caption, feed that exact corrected version back into your skill ("this is my final result, learn from it"). This way, the skillset becomes unique and better with each iteration. And: Context beats quantity — it has to be good, precise, and not too much. What you'll take away from this episode: Marco and Manuel demonstrate what productive work with Claude Code and Obsidian as a second brain feels like when you not only use skills but continuously improve them. You'll learn why productivity with AI agents goes far beyond the often-cited "factor of 10," how to automatically create reels, newsletters, and captions from podcast transcripts, and why your own iteratively trained AI skills are a real competitive advantage. They also assess what AI telephony can truly offer small businesses today—and where the limitations of off-the-shelf solutions lie. Frequently Asked Questions: Is AI telephony worthwhile for small businesses? There are good use cases for inbound scenarios such as secretarial and appointment scheduling, especially when staff is scarce. Outbound and cold calling are hardly feasible with AI. Process consulting is almost always crucial—one-click solutions often fail due to individual workflows. What does "making the best input from the output" mean? Instead of being satisfied with the first AI result, you feed your revised final version back into the skill. The AI ​​learns from your final result, and your skillset becomes more precise and unique with each iteration. Do AI models improve when you let two AIs talk to each other? No. If AI trains primarily with AI-generated data, the quality usually decreases—a well-known problem ("model collapse"). That's why high-quality, human sources and precise context are so valuable. How do you use Obsidian and Claude Code for content production? You can automatically generate YouTube descriptions, newsletters, Reels, and social media captions from a podcast transcript. The trick is to neatly package each step into a reusable skill once and then continuously improve it with feedback. About us: The AI ​​Snack is the daily podcast from OctopusGenius GmbH in Cologne/Pulheim. Marco and Manuel discuss artificial intelligence, AI automation, AI tools, and digital transformation in medium-sized businesses. OctopusGenius develops customized AI solutions.