This is where I collect my AI videos: practical tool tests, tutorials, explainers and series on AI agents, automation, open source and new workflows. Each relevant video gets a summary, chapters, context and the current transcript status.
If you are not looking for one specific video, start here. These hubs group series, tool tests and practical examples by search intent.
Agents are not magic employees. They are systems with a goal, tools and a loop. Once you understand that, many demos become less impressive and a few real use cases become much more interesting.
5 videosAn app is not good because AI built it. It is good when it solves a real problem, stays inspectable and does not fall apart at the first edge case.
8 videosWith AI audio, the first impression is often stronger than the result after ten minutes of real use. Sound, control, rights, cost and workflow belong together.
7 videosSelf-hosted sounds like control. Often it is. But control does not automatically mean safety. Local AI tools need clear thinking about permissions, data, updates and boundaries.
6 videosNot which one is smarter, but what happens when you actually put them to work: on images, on strategy, on design, on debugging. And where it gets uncomfortable.
A free tool built on the open-source ACE-Step 1.5 model. One track came out genuinely good. Instrumentals, lyrics control and remix did not hold up.
From missing paper score sheet to offline HTML app: a compact example of when AI can help build useful small internal tools.
I cloned my own voice twice with the two methods ElevenLabs offers. One sounded nothing like me, the other one landed. Here is what separates them, and what the 22 dollar plan actually buys.
Voicebox part 4: Claude Code and Codex connect through MCP, diagnose the issue and finally make local voice output work.
Local dictation instead of a Wispr Flow subscription: Captures, Whisper Base vs Turbo, German accuracy, LLM refinement and the honest Mac hotkey verdict.
Qwen, Chatterbox and ElevenLabs in a direct listening test: live samples, German and French, laugh tag, crash and an honest local voice verdict.
The Voicebox series setup: installation, models, Chatterbox, Qwen, Captures and why the app challenges ElevenLabs and Wispr Flow at once.
The StickerLog longform build: Plan Mode, Codex as a second opinion, Claude Design, a local-first app, backup and the real lesson behind vibe coding.
An AI premortem on real YouTube data: why views are not value and how Claude surfaced hidden strategy problems.
Emergence World in context: model behavior, agent ecosystems and why safety is not only a property of a single agent.
A structured test of ChatGPT Images 2: what improved, what gets dangerously close to fake-news visuals and where the result can mislead.
Anthropic's Claude Design in context: prototypes, presentations, brand systems and why this matters for non-designers.
Nine common Claude mistakes: long chats, bad context, active connectors, output cost and why Projects or RAG can be better.
The architecture for tasks that are structurally too large for a single agent, and the expensive misunderstanding about what adding agents actually buys you.
The hard limit behind every AI conversation. Why it is a constraint rather than a bug, and why a bigger window is not uniformly better.
Every model has a knowledge cutoff. RAG is how you get around it without retraining anything, and why retrieval quality decides your answer quality.
The origin story of this website: AI browser control struggled inside Wix, while Claude Code could build a faster, cheaper and more controllable static site locally.
The capability that separates a language model from a system that can act in the world, and the misunderstanding that matters for your security model.
The instruction layer that runs before every conversation you have with an AI, and why most people never realise it is there.
The Model Context Protocol explained in four minutes: why agents get stuck in demos, how MCP connects tools and why it changes reach, not intelligence.
What an agent actually is, how it differs from a chatbot in practice, and the one mistake that breaks most agent projects before they start.
The practical walkthrough: choose a server, secure a VPS, install OpenClaw, connect Anthropic, pair Telegram, troubleshoot errors and run the security audit.
An open-source assistant with real file access and memory that spans weeks, reachable in WhatsApp or Telegram. Including the security risks nobody puts in the thumbnail.
Everything you can do with a track after generating it: personas, the timeline editor, section replacement, uploads and splitting a song into 12 stems.
A complete walkthrough from signup to finished song: what the free tier really gives you, how the plans differ, and how commercial use rights actually work.