Blog

Claude Fable 5 vs. Open-Source: The AI War Over the Best Agent Architecture Intensifies

Thursday, June 11, 2026

Hello, this weekly digest processes the most important new videos from around 40 curated AI and coding YouTube channels — with substance, no superficial top-5 lists. One complete summary per video, plus a weekly overview of dominant themes. Read at your leisure — or copy a summary into your LLM of choice and dive deeper. Click the link under each summary to watch the original video.

“`html

AI mit Arnie (2 new videos)

• Hermes Desktop is dangerously good
  6.6.2026, 16:10:34

  The video introduces the Hermes Desktop app, which enables users to work with various AI models locally on their computer. Installation is now greatly simplified and no longer requires complex steps like the Windows Subsystem for Linux. The app offers a user-friendly interface that lets you leverage different AI models such as GPT-5.5 or Codex and switch between them. A special feature is Hermes’ ability to improve itself and automatically write Skills for various tasks like frontend design, image generation, and integration with tools such as Obsidian and ConfUI. The app also supports connections to various channels like Telegram, Discord, and Slack, as well as the creation and management of Superagents that can work in the same session. Additionally, users can utilize local models like Ollama and adjust the context window accordingly. The app is 100% open source and offers users the flexibility to use different providers and models without being locked into a specific vendor.

  The video explicitly covers the Hermes Desktop app, which is based on open-source models, and is geared toward intermediate to advanced users.

• I turned Hermes Desktop into a Super-App
  11.6.2026, 08:38:02

  The video demonstrates how to set up and use the Hermes Desktop app as a central command center for various AI-powered tasks. It presents ten concrete use cases, including invoice processing, contract evaluation, video editing, Git repository management, and task automation via Cron jobs. Particular emphasis is placed on the ability to use both local and cloud-based models to work in compliance with data protection regulations (GDPR-compliant). The video also shows how to install and use Hermes on a virtual private server (VPS) to enable 24/7 automations. The video also covers integration with tools like Obsidian and n8n to create and manage complex workflows.

  **Final comment:** The video explicitly focuses on the Hermes Desktop app and various AI models such as Codex, OpenAI, and local models via Ollama. It is geared toward intermediate to advanced users who already have experience with AI tools and server administration.

“`

Cole Medin (2 new videos)

• Google’s Agents CLI: The CLI + Skills Combination to Ship AI Agents EASILY
  11.6.2026, 00:00:16

  The video demonstrates how easy it is today to create an AI agent and deploy it to production thanks to tools like Google’s Agent CLI. The author shows how to build an AI agent from concept to reliable deployment using a combination of a command-line interface (CLI) and skills. Google’s Agent CLI is a powerful, open-source tool that helps create agents with Google’s Agent Development Kit (ADK), a framework used by many people. The author compares the current simplicity of building agents with earlier, more complex methods and shows how AI-coding assistants like Claude Code further streamline the process. The author emphasizes that while building agents has become easier, complex systems like storage systems and RAG architectures still present challenges. The main part of the video consists of a live demo where the author creates and deploys an AI agent using Google’s Agent CLI without typing a single command in the terminal themselves. The author uses Claude Code to manage the entire process, from installing the CLI and skills through building, evaluating, and deploying the agent. The created agent is a simple “Ask Your Data” agent that writes Python code to answer questions about a CSV file. The author also discusses the benefits of frameworks like Google’s ADK compared to coding agent SDKs, particularly regarding token efficiency and speed, which are crucial for production deployment. By the end of the video, the agent is successfully deployed in Google Cloud and tested, with the author showcasing the various features and settings of the deployed agent.

  The video explicitly covers Google’s Agent CLI and ADK and is geared toward intermediate to advanced users.

• Live AI Q&A + Crushing it in Chess at the Same Time – Come Hang Out!
  7.6.2026, 04:26:50

  Without the specific video link, I cannot create a summary. Please share the YouTube link so I can analyze the video and write a precise summary for you.

Nate Herk | AI Automation (6 new videos)

• I Turned Claude Fable Into The Ultimate Second Brain
  10.6.2026, 04:40:11

  The video demonstrates how the creator uses their “Second Brain” and AI Operating System (AIOS) with Claude Fable to boost productivity and efficiency. Claude Fable, a new model from Anthropic, is described as particularly powerful and offers enhanced security measures (“cyber guard rails”). The creator emphasizes the importance of a mindset shift, moving from using various AI tools toward a centralized system with Claude Fable. His AIOS consists of two main components: the “Second Brain,” which stores knowledge and context, and the “AI Operating System,” which adds capabilities and automations. The setup follows a framework called “the four Cs”: Context, Connections, Capabilities, and Cadence. Context includes personal and business information, Connections refer to dynamic data sources like emails or project management tools. Capabilities encompass task automation, and Cadence enables these automations to run continuously. The creator shares practical tips for using Claude Fable, such as using it as a thinking partner, interviewing the user to extract knowledge, and reviewing the AI’s work. He also emphasizes the importance of security measures and the need to verify the AI’s output. The video concludes with a Q&A session where the creator addresses common questions about costs, data security, coding skills, and team usage.

  The video explicitly covers Claude Fable from Anthropic and is aimed at intermediate to advanced users.

• Claude Mythos is Finally Here.
  9.6.2026, 18:00:20

  The video covers the release of two new models from Anthropic: Claude Fable 5 and Claude Mythos 5. Fable 5 is now available to all users, while Mythos 5 is initially limited to Glasswing partners. Both models cost $10 per million input tokens and $50 per million output tokens, which is twice as expensive as Opus. Fable 5 is included in Pro Max, Team, and Enterprise plans until June 22, after which it will only be available for an additional fee. Mythos 5 is a more powerful version of Fable 5 without cybersecurity safeguards and will initially be distributed through Project Glasswing. Both models show significant benchmark improvements over Opus 4.8 and other models, particularly in areas like software engineering, knowledge work, and cybersecurity. The creator emphasizes the importance of agent loops but warns against excessive use, which can lead to high token costs. He is curious about the practical use of Fable 5 and will publish further videos on the topic.

  The video explicitly covers Claude Fable 5 and Claude Mythos 5 from Anthropic and is aimed at intermediate to advanced users.

• How to Build Claude Subagents Better Than 99% of People
  9.6.2026, 00:44:55

  The video explains how to use Subagents in Claude Code, a tool from Anthropic. Subagents are independent AI agents that can be delegated by a main agent to complete specific tasks. The main agent acts as an orchestrator and can run multiple Subagents in parallel, each with different personalities, capabilities, and models. This helps keep the main agent’s context clean and saves costs by using cheaper models for specific tasks.

  Subagents can be both built-in and custom agents. Custom Subagents are created as Markdown files and can contain specific instructions, tools, and models. They can be used at the project or global level, depending on whether they should be available in a specific project or globally. Subagents can also be integrated into Skills, which in turn can utilize Subagents.

  Users can leverage Subagents in various ways: automatically, proactively, or explicitly by naming the agent. It’s important to formulate descriptions and instructions in Subagents precisely to avoid malfunctions. Subagents can also be configured as read-only agents to increase security.

  The video also demonstrates how to create and configure Subagents by creating a Markdown file with YAML frontmatter. This file contains information such as the agent’s name, description, tools used, and model. Users can then utilize the Subagents in various projects or globally.

  In summary, Subagents in Claude Code provide a powerful way to delegate tasks, keep context clean, and reduce costs. They can be used and configured in various ways to meet specific requirements.

  The video explicitly covers Claude Code and is aimed at intermediate to advanced users.

• Is Claude Mythos Coming?
  6.6.2026, 21:26:20

  The video discusses recent developments surrounding Anthropic’s Mythos AI model, which briefly appeared on the company’s API and then disappeared. Mythos is a highly potent model specifically developed for cybersecurity and is considered the successor to Claude Opus. To date, it has only been made available to selected partners as part of Project Glasswing to identify and fix security vulnerabilities in critical software.

  Mythos’s appearance on the API has sparked speculation that the model could soon be publicly available. However, Anthropic has already made it clear that Mythos should not be made generally accessible for the time being. The video creator is skeptical about whether Mythos will be released to the public in the near future, suspecting that recent leaks may be deliberate marketing strategies to generate attention.

  Additional factors fueling speculation include Anthropic’s upcoming IPO and competition with OpenAI, which could also launch a new model, GPT-5.6. The creator believes it’s more likely that Mythos’s capabilities will be gradually integrated into future versions of Opus rather than being released as a standalone model.

  The video explicitly covers Anthropic and OpenAI and is aimed at intermediate to advanced users.

• AGI is Here. Anthropic Just Proved It.
  5.6.2026, 20:42:32

  The video discusses current AI developments, particularly based on a report from Anthropic showing that over 80% of code shipped by Anthropic is now written by their own AI model Claude. The speaker argues that AGI (Artificial General Intelligence) is already here, defined as an AI’s ability to independently solve open, unclear problems. Anthropic categorizes tasks into four categories, from trivial to open, and shows that Claude has achieved a 76% success rate on open problems, a massive increase from 26% six months ago. The AI can now handle tasks that take 12 hours and has in some cases made better decisions than human researchers. The speaker discusses three possible future scenarios for AI: stagnation, continuous improvement under human control, and complete AI autonomy. He emphasizes the importance of AI alignment to ensure it serves humanity and the difficulties in guaranteeing this, as progress is hard to monitor. The speaker warns against ignoring AI development and emphasizes that the ability to identify the right problems and effectively deploy AI is becoming increasingly valuable.

  The video explicitly covers Claude, Anthropic’s AI model, and is aimed at intermediate to advanced users.

• The Skill That 10x’d My Claude Code Projects
  4.6.2026, 17:55:54

  The video addresses the challenge of transferring knowledge and context from your own mind into an AI system to achieve better and more personalized results. The focus is on a technique called “Grill Me,” which is designed to extract comprehensive information from the user and transform it into reusable context for the AI. This technique was originally developed by Matt PCO and consists of a simple yet effective prompt structure that guides the user through a series of questions to achieve mutual understanding. The video creator has further refined this technique by adding a checkpoint system that ensures all answers are recorded in a document to prevent information loss. This results in better Skills, context, and projects because the AI develops a deeper understanding of the user and their processes. The author emphasizes the importance of this technique to improve the efficiency and quality of AI results.

  The video explicitly covers Claude Opus 4.8 and is aimed at intermediate to advanced users.

Ben AI (1 new video)

• Claude Managed Agents Will Change How You Sell AI Forever
  9.6.2026, 09:10:22

  The video explains how Managed Agents work and their potential on the Claude AI platform, particularly for selling AI solutions. It demonstrates how Managed Agents can contain pre-configured agent workflows with skills, MCPs (Managed Connectors), memory, and Sub-Agents that can be deployed via API and used in various environments like Slack, custom apps, or other software solutions. The presenter shares his experience using Managed Agents in his AI agency and demonstrates concrete use cases such as churn recovery and lead prospecting. The video emphasizes that Managed Agents represent a major opportunity for selling AI solutions to companies that may not yet use AI tools like Claude. It also highlights the integration of skills into these agent workflows, as these are testable and improvable, which increases the reliability and determinism of automations. The presenter shows how Managed Agents can be deployed in various environments like Slack, Notion, or through automation platforms such as n8n or Zapier. Finally, the video mentions the possibility of using Managed Agents for continuous learning and as part of an AI-OS infrastructure, which enables new business models and AI SaaS ideas.

  The video explicitly focuses on Claude’s Managed Agents and is designed more for intermediate to advanced users who already have experience with AI tools and automation platforms.

Brian Casel (3 new videos)

• Hermes vs. Claude Cowork? Wrong Question.
  9.6.2026, 12:48:15

  The video addresses the challenge of committing to a specific AI agent platform, as the landscape constantly evolves and new platforms regularly emerge. The author emphasizes that rather than relying on a single platform, it’s more sensible to develop underlying patterns and processes that can run across different platforms. He demonstrates his current setup using two platforms: Hermes and Claude Cowork.

  Hermes is primarily used for routine background tasks, such as gathering and summarizing content, monitoring SEO health, and managing code repositories. The author prefers Discord as the interface for interacting with Hermes due to better Markdown support and the ability to leverage multiple channels and threads.

  Claude Cowork is utilized for creative and high-quality tasks like content ideation, writing, and design. The author uses Claude Cowork’s Scheduled Tasks feature to automate these tasks. He mentions being forced to offload some tasks to Hermes due to pricing changes and limitations with Claude Cowork.

  The author stresses that the patterns and processes he uses on these platforms are platform-agnostic and can therefore be easily transferred to new platforms. He concludes with the recommendation not to rely on a single platform, but rather to understand and leverage the underlying patterns.

  The video explicitly covers the AI tools/models/providers Claude, OpenAI, and Hermes and is geared toward intermediate and advanced users.

• Why apps built with AI look a little… OFF
  5.6.2026, 12:00:17

  The video addresses the problem of “design drift” when building apps with AI agents. This often results in inconsistent designs because agents reinvent the UI each session without a shared design source. The author presents a solution: a design system that centrally defines colors, typography, spacing, and components and integrates them into the code. This prevents design drift and ensures consistency.

  The author demonstrates how he implements such a design system in his apps, consisting of various components like colors, typography, buttons, forms, and lists. He uses a free agent capability he developed to integrate the design system into his codebases. By defining the design system in the `claude.md` or `agents.md` file, it’s ensured that the AI agent considers the design guidelines for every UI request.

  The author emphasizes the importance of understanding such tools rather than using them blindly. He recommends integrating the design system from the start in new projects, as retrofitting it later is more difficult. Additionally, he presents a Rails and React starter template that already has his design system integrated.

  The video explicitly covers Claude and Claude Code and is aimed at intermediate and advanced builders working with AI agents.

• Stop Waiting for Claude to Finish — Do This Instead
  4.6.2026, 14:00:04

  The speaker describes his working routine with the AI agent Claude, where he starts tasks before leaving work, going to sleep, or exercising in the morning. This allows him to use the wait time—while Claude analyzes code or conducts research, for example—for other activities. He also mentions that in 2026, AI agents like Claude independently complete tasks without explicit instructions—they work according to set workflows and are essentially team members.

  The video explicitly covers Claude and is geared toward intermediate or advanced users who already have experience with AI agents.

Melvynx (7 new videos)

• FABLE 5 : LA SUPER INTELLIGENCE EST DÉJÀ LÀ ? (modèle Claude)
  11.6.2026, 07:56:35

  The video introduces the new Claude Fable 5 and Claude Mythos 5 models and compares them with other models like Codex, Opus, and GPT 5.5. Claude Fable 5 is positioned as the most intelligent and powerful model in the world for most tasks, though with safety guardrails that activate for potentially dangerous requests. The SW Bench Pro benchmark shows that Fable 5 achieves a score of 80%, which is 10% higher than Opus 4.8. Another benchmark, Frontière Code, evaluates the models’ ability to generate production code, with Fable 5 successfully completing between 10% and 32% of tasks without modifications. The user shares personal experiences, including migrating an application from Postgres, Ingest, and Redis to Convex, as well as creating a mobile application for training. Despite some errors and high costs ($414 for Claude in one day), Fable 5 is described as extremely powerful. Critical points include the model’s time limit until June 22 and its high costs compared to other models. The video concludes with a recommendation to try the model and an announcement of further testing once an equivalent GPT model is available.

  The video explicitly focuses on Claude Fable 5 and Claude Mythos 5 models from Anthropic and is aimed at intermediate to advanced users.

• Quand vient l’heure de la pause…
  10.6.2026, 07:00:20

  The video showcases a modern, well-equipped office kitchen. The creator presents various highlights, including a beverage dispenser that enables personalized water orders. You can choose between different flavors like “light sparkling raspberry” and add-ons like “vitamin boost” and order your desired number of bottles. The creator tests the system and is thrilled with the results.

  Another highlight is the “ficello,” a device that produces string cheese, which the creator finds particularly amusing and typically American. Overall, the creator is impressed by the equipment and comfort this kitchen offers, and thinks it would be tempting to work at such a company.

  The video doesn’t focus on specific AI tools or models and is suitable for a broad audience, possibly with a focus on intermediate users interested in modern office setups.

• Je code 1 HEURE avec Codex devant toi (mes secrets devoilé)
  9.6.2026, 16:00:25

  The video shows how the creator works with the Codex tool to make various features and improvements to his Subface application. Here are the main steps and content of the video summarized:

  1. **Introduction and Setup**:
  – The creator uses Z as the main interface to manage multiple projects.
  – He shows how to open projects in Z and execute terminal-based commands to start servers and test applications.

  2. **Workflow with Codex**:
  – The creator explains how he uses Codex for various tasks, including implementing new features and debugging.
  – He shows how to start tasks in Codex and work on multiple projects simultaneously.

  3. **Specific Features and Improvements**:
  – **Inspiration Anonymization**: The creator wants to add a feature that anonymizes videos and thumbnails by replacing people, text, and logos.
  – **Free Trial Refactoring**: He works on improving the Free Trial page to make it more user-friendly and appealing.
  – **Onboarding Test**: He implements and tests an onboarding system for administrators.
  – **Debugging**: The creator fixes various bugs, such as unwanted dialog boxes appearing and issues with image generation.

  4. **Code Review and Optimization**:
  – The creator uses tools like Thermonucléaire Code Quality Review to review and optimize generated code.
  – He shows how to push changes to GitHub and create pull requests.

  5. **End Result**:
  – The creator demonstrates the successful implementation of new features and improvements, including the anonymization feature and the redesigned Free Trial page.

  The creator primarily uses Codex and Z for his work and shows how he efficiently uses these tools to improve his application. The video is aimed at intermediate to advanced users who already have experience with code development and using AI tools.

• Tu es le bienvenu
  9.6.2026, 13:25:31

  The video addresses the danger of relying too heavily on a single AI tool, particularly Claude. The creator emphasizes that current market leaders like OpenAI, Anthropic, and Chinese models aim to lock in users through vendor lock-in strategies, which can be costly and restrictive in the long term. He advises staying flexible, testing various tools, and avoiding dependence on one tool. As an example, he mentions his own use of Claude, Cursor, and CEX to ensure he doesn’t depend on a single tool. The focus is on being ready to switch tools whenever better alternatives become available.

  The creator explicitly addresses Claude, OpenAI, Anthropic, open-source models, and Chinese models, with the video aimed at intermediate and advanced users.

• C’est foutu
  8.6.2026, 11:22:40

  The video discusses the decision to switch between different AI models like Claude and Codex. The speaker emphasizes that switching doesn’t automatically lead to more income, but often means additional effort since he would need to create new tutorials and content. He argues that he must remain honest and won’t pretend to use a specific tool just to please the community. The speaker acknowledges that GPT-4 (Codex) has made significant progress and matches or even exceeds Claude’s performance, which is confirmed by benchmarks and developers. He stresses that even small improvements in AI performance can have major impacts on his daily work. Additionally, he criticizes people hesitant to switch AI models, arguing they may not have taken the necessary steps to remain flexible with different agents.

  The video explicitly addresses Claude and Codex and is aimed at intermediate to advanced users.

• Formation Codex GPT : tout apprendre sur Codex en 2026 (gratuit)
  7.6.2026, 16:00:29

  The video provides a comprehensive introduction to Codex, an AI tool by OpenAI specifically designed for software development. Codex is presented as a powerful competitor to Cloud Code and can create applications like Umail, Saveit.now, and Ciao App. Unlike ChatGPT, which functions as a “one-shot” model, Codex enables more complex tasks through its Orchestrator function by utilizing models like GPT-4/5 and combining them with tools like file operations.

  Codex installation is done through the official website, and after logging in with a ChatGPT account, you can choose between different pricing models. Setup includes selecting “Coding Mode” and enabling Full Access for maximum functionality. Codex’s user interface resembles ChatGPT but offers additional features like working in projects and integrating with IDEs like VS Code or Z.

  A central feature of Codex is the ability to use Skills, which automate specific tasks or workflows. These Skills can be categorized into different types, such as Tool Skills (API integration), Workflow Skills (code reviews, pull requests), and Meta Skills (skills for managing other skills). Skills can be invoked through commands like `/skill` or by directly prompting the agent.

  Another important aspect is Sub-Agents, which allow complex tasks to be divided into smaller, parallel subtasks. These Sub-Agents can use specific models and return their results to the main agent, which optimizes efficiency and context consumption.

  The video also shows how to configure and manage agents by creating Toml files in a special directory. These agents can then be used for various tasks like code exploration, documentation research, or web search.

  Finally, it’s recommended to download the speaker’s configurations and skills to get the most out of Codex. It’s emphasized that there are still many other features like Hooks, MCP, and CLI Skills that could be covered in future videos.

  The video explicitly focuses on OpenAI and is aimed at intermediate to advanced users.

• MIGRATION de Claude à Codex : 1 seul ligne à faire (TUTO SIMPLE)
  4.6.2026, 16:00:38

  The video shows how to migrate configuration from Cloud to Codex to simplify the use of Skills and Agents and standardize configuration across platforms like Cursor and Codex. The creator introduces an open-source CLI tool that with the command `npx ai-blueprint unify` migrates the entire configuration in less than 5 seconds. The tool creates symlinks for Skills and Agents so they’re available on both platforms. Additionally, a backup of the original configuration is created to prevent data loss. The creator emphasizes the importance of keeping configuration easily transferable, as tools can change over time. Additionally, the tool Lalal.ai is presented, which separates audio tracks and offers various audio editing functions.

  The video addresses specific tools like Codex, Cursor, and the open-source CLI tool from ai-blueprint and is aimed at intermediate users.

Dave Ebbelaar (1 new video)

• Build a Full-Stack GenAI Project in 4 Hours (FastAPI, React, Supabase)
  6.6.2026, 12:44:23

  ### YouTube Video Summary

  The video demonstrates a comprehensive process for building a full-stack GenAI application that allows users to ask questions based on hundreds of documents stored in a database. The focus is on the entire architecture, including frontend, backend, database connection, ingestion and ranking pipeline, and how everything fits together.

  **Main Steps and End Result:**
  1. **Project Setup:**
  – Creating a GitHub repository with agent files (e.g., `agents.md`) and dependency management (e.g., `pyproject.toml`).
  – Setting up a virtual Python environment manager (UV) to manage dependencies.

  2. **Data Download and Processing:**
  – Using a Python script (`download.py`) to download financial reports (10-K, 10-Q, annual reports) from companies like Apple, Microsoft, Nvidia, Amazon, and Google from the SEC website.
  – Converting downloaded HTML files to Markdown using the Dockling tool to prepare data for further processing.

  3. **Database Modeling and Migration:**
  – Defining database models (e.g., `users`, `source_documents`, `document_chunks`, `chats`, `messages`, `citations`) using SQLAlchemy.
  – Performing database migrations with Alembic to synchronize models in the Supabase database.

  4. **Backend Development:**
  – Setting up a FastAPI backend with endpoints for user authentication, chat processing, and document queries.
  – Integrating OpenAI for embedding creation and language model processing.

  5. **Frontend Development:**
  – Setting up a React frontend with TypeScript and Tailwind CSS for the user interface.
  – Implementing authentication mechanisms (e.g., user login and logout) with Supabase.

  6. **Ranking Pipeline:**
  – Implementing a hybrid ranking pipeline that combines semantic search (with PGVector) and full-text search (with Supabase).
  – Using Pydantic models to structure queries and responses.

  7. **Deployment:**
  – Deploying the application on Railway, a platform-as-a-service provider, using Docker containers for backend and frontend.
  – Configuring environment variables and database connections for production.

  **AI Tools/Models/Providers:**
  – **Claude / OpenAI / Gemini / Open Source:** The video uses various AI tools such as OpenAI for embeddings and language models, Pydantic for data structuring, and Dockling for document processing.
  – **Specific Tools:** Cursor, Lovable, Cline, n8n.

  **Target Audience:**
  – **Intermediate to Advanced:** The video is aimed at developers with advanced knowledge of Python, React, databases, and deployment processes who are interested in AI engineering and full-stack development.

  **Final Comment:**
  The video provides deep insight into building a full-stack GenAI application and shows how various components work together to create a powerful and user-friendly solution. It’s a valuable resource for developers looking to expand their skills in AI engineering and full-stack development.

Niklas Steenfatt (1 new video)

• I competed in the world’s biggest chess tournament!!!
  4.6.2026, 15:56:55

  The YouTuber reports on his participation in the Grankle Chess Open in Karlsruhe, one of the world’s largest chess tournaments. He describes his experiences over five days, competing against players with higher Elo ratings, including several strong opponents. He won some games, lost others, and experienced dramatic moments, such as defeating a stronger opponent who ran out of time. The YouTuber emphasizes the mental and physical strain of the tournament and the special atmosphere among chess enthusiasts. He analyzes some of his games and shows how he became more confident over time despite initial uncertainty. In the end, he achieved a total score of 5.5 out of 9 points and reflects on the unique experiences and challenges of tournament chess.

  The video does not address specific AI tools or models and is primarily intended for intermediate and advanced chess players.

Fireship (1 new video)

• Anthropic is starting to panic…
  9.6.2026, 17:32:30

  The video discusses Anthropic’s current developments, which is valued higher than OpenAI and is planning a billion-dollar IPO. Anthropic warns about the danger of recursive AI self-improvement, which could become a threat to humanity. The company proposes a global halt to AI development, though this appears difficult due to competition with other companies like OpenAI, DeepMind, and XAI. Historically, such warnings have often proven exaggerated, as the GPT-2 example shows. Nevertheless, there are concerns that AI is already being deployed in critical areas such as data centers, robotics, and weapons. A study by economists from Boston University warns of an “AI Layoff Trap,” where automation could lead to a decline in demand and thus economic problems. Alternatively, the thesis is presented that AI may not be as powerful as often assumed, and that many AI projects in companies do not achieve measurable success. The video also mentions tools like Pioneer that can help improve the efficiency of AI applications.

  The video covers Anthropic, OpenAI, DeepMind, XAI, and specific tools such as Pioneer, Codeex, Cursor, and Hermes, and is aimed at an intermediate to advanced audience.

Julian Ivanov | AI Automation (1 new video)

• How to Make Claude Watch Every Video for You
  7.6.2026, 18:05:41

  The video demonstrates how to use Claude (an AI platform) to analyze and summarize videos frame-by-frame, rather than relying solely on transcripts. Users can upload or link videos from various platforms (YouTube, Instagram, Loom, etc.), and Claude extracts key information and images from the videos. This information is then stored in a note in Obsidian, which is particularly useful for visual learners.

  Concrete use cases include:
  1. **Educational videos**: Summarizing tutorial videos such as those about transformer architectures, with important diagrams extracted and inserted into notes.
  2. **Videos without speech**: Analyzing vacation videos or other content without spoken narration to identify specific scenes or events.
  3. **Bug recordings**: Analyzing screen captures to identify errors or issues in apps or programs and find solutions.
  4. **Viral content**: Analyzing successful social media videos to understand why they perform well and what visual hooks are used.

  Installation of the required plugin is straightforward and is performed automatically by Claude. Tools needed include YT-DLP for downloading videos, FFMPEG for extracting frames, and a transcription model like Whisper, which can be accessed through platforms like Grock.

  The video explicitly features Claude and open-source tools such as YT-DLP, FFMPEG, and Whisper. It is better suited for intermediate users who already have experience with AI tools and their integration into workflows.

Mark Kashef

No new videos in this period.

Everlast AI (3 new videos)

• 20 AI Tools 99% Have Never Heard Of (And Will Get You Ahead of Everyone)
  9.6.2026, 15:15:07

  The video presents a list of 20 AI tools divided into four categories: Agent Layer, Agent Tools, Daily Driver, and Monitoring. The focus is on tools that go beyond well-known standard tools like ChatGPT and Gemini, enabling greater autonomy and productivity.

  1. **Agent Layer**:
  – **Codex**: An OpenAI super-app that serves as a working environment for autonomous agents. It enables access to the entire file system and can handle complex tasks like accounting, market analysis, and meeting note analysis.
  – **Cloud Code**: A tool particularly good at designing and orchestrating subagents. It can create complete landing pages and perform competitive analysis.
  – **Cursor**: A full-fledged development environment with agents as its core feature, capable of switching between different models and offering its own frontier model for coding tasks.
  – **Google AI Studio**: A tool that lets you build an entire Android app with just one prompt and share it with a few clicks.

  2. **Agent Tools**:
  – **Browser Use**: A tool that gives agents access to a real browser and mouse control to operate websites.
  – **Excalidraw MCP**: An open-source tool for visualizations offering editable views for collaborative work.
  – **N8N MCP**: A workflow automation tool that gives agents access to over 1850 available integrations.
  – **Meta @ CLI**: A tool that gives agents full access to the Ads Manager to analyze and optimize campaigns.
  – **Hixfield CLI**: A tool that lets you generate ad creatives directly from the terminal.
  – **Google Workspace CLI**: A tool that gives agents access to Gmail, Sheets, Docs, and Calendar.
  – **Agentmail**: A tool that gives agents their own email inbox with full API support for sending and managing emails.
  – **Remotion**: An open-source plugin that lets you generate and edit videos as code.

  3. **Daily Driver**:
  – **Supabase**: An open-source database serving as a single source of truth for all data.
  – **Ollama**: A tool that lets you run leading open-source AI models locally and offline on your Mac or server.
  – **Cloudflare**: A tool that lets you securely bring apps to the internet and protect them with Zero Trust.
  – **Corporate LLM**: A platform for productive and GDPR-compliant work with AI models and tools.
  – **Notebook LM**: A Google tool that lets you answer exclusively from uploaded sources and conduct research work.
  – **Voicely**: A desktop app that lets you dictate text and work five times faster as a result.
  – **Magnific**: A creative tool that lets you integrate text-to-image and video models and automate workflows.

  4. **Monitoring**:
  – **Lang Fuse**: An open-source tool that logs every model call and monitors costs. It also enables automatic evaluation of agent response quality.

  The video is intended more for intermediate and advanced users, as it presents specific tools and techniques requiring deeper understanding of AI and its applications.

• AI News: HUGE ChatGPT Update! Codex “Apps”, New Features & Local AI Catching Up
  7.6.2026, 08:15:36

  The video summarizes the latest developments in the AI world, with a focus on OpenAI’s major updates, particularly the integration of Codex into ChatGPT. This integration enables a unified experience where agents work in the cloud and proactively complete tasks before users even recognize them. New features like role-specific plugins, annotations, and Codex Sites allow users to create and share software with simple prompts. Additionally, ChatGPT’s improved memory architecture is introduced, which updates itself and is available in the free plan.

  In parallel, there’s progress in local AI models. Google has released the open model Gemma 4.12B, which understands text, images, audio, and video and runs on a regular work laptop. This model can be freely integrated into Corporate LM to work locally and securely. Nvidia has also introduced a new open-weights model, Nemotron 3 Ultra, which is supposed to be more efficient and faster.

  The video also shows practical applications of these technologies, such as creating a mini-app in Codex and using local models in Corporate LM. It notes that while local models are suitable for certain tasks, they’re not yet sufficient for complex tasks like agentic coding.

  Additionally, developments in humanoid robots, particularly from AGI and BYD, as well as construction site automation through Sensmore are discussed. Microsoft and Meta present new AI models and agents intended to improve their respective platforms.

  Closing comment: The video covers OpenAI (ChatGPT, Codex), Google (Gemma), Nvidia (Nemotron), Corporate LM, and specific tools like Codex Sites. It’s intended more for intermediate and advanced users.

• No.1 Historian: This Time It Could Be Different! Luddites, Capitalism & AI (Jürgen Kocka)
  4.6.2026, 14:15:11

  The video is a conversation between Leonard Schweding and Professor Dr. Jürgen Kocka, a renowned historian and capitalism researcher. The main topics of the conversation include the definition and controversies around capitalism, historical changes in the world of work, the role of technology and automation, and the future of capitalism in the age of artificial intelligence (AI).

  Kocka emphasizes that capitalism is a complex economic system encompassing property rights, markets, and investments. He discusses historical developments, particularly industrialization and its associated social and economic changes. Kocka mentions Schumpeter’s creative destruction, which brings both opportunities and risks, and emphasizes the importance of protests and criticism for capitalism’s further development.

  A central theme is the role of technology and automation in the world of work. Kocka references historical examples like the Luddite movement and the weavers’ rebellion to show that technological progress always brings social and economic upheaval. He argues that fear of job losses through automation is often exaggerated, as new jobs and professions emerge.

  Kocka also discusses the current debate around AI and its impact on the world of work. He emphasizes the importance of adaptability and the need to limit the expansion of capitalist principles into areas where they don’t belong. He warns against technocratic utopias and emphasizes the importance of criticism and protest for capitalism’s further development.

  The video is suitable more for intermediate and advanced viewers, as it addresses complex historical and economic concepts. It doesn’t focus on specific AI tools or models but rather offers a broad historical and social perspective on current developments.

  Closing comment: The video doesn’t focus on specific AI tools or models and is more suitable for intermediate and advanced viewers.

No Priors: AI, Machine Learning, Tech, & Startups (4 new videos)

• Mark Zuckerberg’s Plan to End All Disease
  10.6.2026, 19:23:49

  The video tells the story of Mark pursuing an ambitious goal to build an organization capable of healing, preventing, and managing diseases by the end of the century. Initially, these plans were dismissed by renowned scientists, including Nobel laureates. Mark emphasizes that the goal was never to cure all diseases himself, but rather to develop tools that accelerate the entire scientific community and collectively combat diseases. What once seemed unrealistic now appears even too conservative.

  Final comment: The video doesn’t address specific AI tools or models and is aimed at a broad audience, including beginners and intermediate viewers.

• Biohub: The Future of Biology is Open-Source with Mark Zuckerberg, Priscilla Chan, and Alex Rives
  10.6.2026, 13:00:36

  The video is an interview with Mark Zuckerberg, Priscilla Chan, and Alex Reeves about their work at Biohub and the application of AI in biology. They discuss the founding of Biohub, their vision to equip the scientific community with tools to accelerate understanding of biology and ultimately cure diseases. Biohub focuses on developing open-source tools and fostering collaboration between engineers and scientists. They emphasize the importance of open-source projects in getting tools into scientists’ hands quickly and accelerating progress across the entire scientific community. The conversation also covers challenges and progress in applying AI to biology, including protein structure prediction and developing models for cells and biological systems. They discuss the importance of mechanistic interpretability and how AI models can provide new biological insights. Additionally, they stress the need to transform clinical research to accelerate the translation of basic research into clinical applications. The video explicitly addresses Biohub’s AI models and open-source tools and is geared toward intermediate and advanced viewers.

• Satya Nadella: Why Humans Still Create Value
  8.6.2026, 15:44:06

  The video discusses the connection between human capital and token-based capital in enterprises. It argues that companies are now able to capture implicit knowledge and record it on a balance sheet, something previously impossible. The ability of humans to identify gaps and develop new forms of agency and ambition remains critical for value creation. Despite growing token capital, human capital will continue to be of great importance.

  The video doesn’t address specific AI tools or models and is aimed more at intermediate or advanced viewers interested in corporate strategy and capital management.

• The Rise of the Full-Stack Builder and Hyper-Leveraged Generalist with Microsoft CEO Satya Nadella
  4.6.2026, 10:00:31

  The video is a podcast interview with Satya Nadella, CEO of Microsoft, in which he discusses the future of AI, Microsoft’s strategy, and the societal impact of AI. Nadella emphasizes the importance of concrete benefits and the need to gain society’s trust rather than just making promises. He talks about the rapid scaling of Azure and the need to rethink work processes to leverage AI’s seemingly impossible possibilities. Nadella also mentions the importance of education and the potential for new startups to develop new educational models. He discusses Microsoft’s role as a platform enabling companies to develop their own AI, and emphasizes the importance of private valuations and the ability to improve models. Nadella also speaks about AI’s impact on society and the need for AI’s benefits to be felt in communities. The video explicitly addresses Microsoft’s AI strategy and is geared toward intermediate and advanced viewers.

Nic Conley (1 new video)

• Claude Fable 5 is Here (everything in 7 min)
  10.6.2026, 19:51:34

  Anthropic has released Claude Fable 5, a model based on the powerful but not publicly accessible Mythos architecture. Fable 5 is a streamlined version of Mythos made available to the general public, while Mythos 5 remains reserved for select users. Fable 5 shows significant improvements in specific areas such as agentic coding, where it achieves 80% compared to 69% for Claude Opus 4.8. However, some capabilities, like biology and cyber-security, are limited to minimize potential risks.

  Pricing for Fable 5 is significantly higher at $10 per million input tokens and $50 per million output tokens compared to Opus 4.8, making it twice as expensive. Fable 5 is available until June 22nd for users with a paid Claude account, after which it will be accessible only through Usage Credits or the API.

  In a practical demonstration, both models were tested with the same task: building a 3D flight simulator in a single HTML file. Both models delivered impressive results, with Fable 5 taking slightly longer to execute. However, the differences between the models were not significant enough to justify the higher price, particularly for simple applications.

  The video explicitly covers Claude Fable 5 and Claude Opus 4.8 from Anthropic and is best suited for intermediate users.

Tech With Tim (10 new videos)

• The Best LOCAL Agentic Coding Workflow (Complete Guide)
  10.6.2026, 13:00:23

  This video is a tutorial that explains how to set up and use local models for local programming. The author emphasizes that local models are a cost-effective and internet-independent alternative to cloud-based models that run on powerful servers with plenty of RAM and graphics cards. The focus is on selecting the right model based on available hardware, particularly video RAM (VRAM) or unified memory on M-series Macs. The author recommends various models from the Qwen family, depending on the user’s hardware configuration. The tutorial walks through the installation and configuration of LM Studio and Visual Studio Code to use local models for autocomplete and code generation. It also demonstrates using the “Continue” extension in VS Code to configure autocomplete. The video is suitable for intermediate users who already have basic programming knowledge and experience with code editors.

  **Featured AI tools/models/providers:** LM Studio, Visual Studio Code, Continue extension, Qwen family of models (Qwen 2.5, Qwen 3.6, Qwen 3.5, Qwen Coder Next), Hugging Face.

• The Hard Truth Every Developer Needs to Hear in 2026
  10.6.2026, 12:00:23

  This video discusses the need for developers to adapt to changing market conditions to remain employable. It emphasizes that rejecting change can lead to falling behind, while adapting to new technologies and methods is crucial for keeping your job and livelihood. The speaker acknowledges that current developments are frightening and uncertain, but also bring positive aspects, such as the ability to build more than ever before. The main message is that developers must remain flexible and ready to respond to future changes.

  The video addresses adaptation to new technologies in general and is intended more for intermediate and advanced developers.

• The Software Development Bottleneck Just Moved
  9.6.2026, 12:00:08

  This video discusses the shift in bottlenecks within software development. Previously, the focus was on writing large amounts of code, which was expensive and time-consuming. Today, code is relatively cheap to produce, but other skills have become scarce. These include system design, operational excellence, observability, incident response, debugging in production, and the ability to evaluate trade-offs. The bottleneck is no longer about how much code can be written, but how much code can be reviewed, secured, deployed, and maintained in production with all necessary features. It’s no longer about how much code is produced, but about how good that code is. The emphasized skills enable the development of scalable, profitable, and usable systems.

  The video addresses general software development principles and is intended more for intermediate or advanced viewers.

• Why Companies Won’t Hire Junior Devs And Why It’ll Backfire
  8.6.2026, 12:00:34

  This video discusses the problem of dwindling engineering talent in the software industry. It notes that software quality is declining due to frequent security vulnerabilities and exploits. At the same time, fewer junior developers are being hired and trained, while senior developers are retiring. This creates a situation where companies primarily seek experienced senior engineers, making it difficult for entry-level professionals to start their careers. In the long term, this could lead to a shortage of qualified developers since talent development is being neglected.

  The video addresses general industry trends and is intended more for intermediate or advanced viewers who engage with software development challenges.

• I Went to the Biggest AI Infrastructure Conference
  7.6.2026, 14:30:01

  This video addresses the challenges and solutions related to developing and deploying AI agents. The main focus is on the Temporal platform, which enables reliable execution of AI agents through durable execution. Temporal solves problems like timeouts, server restarts, and other production errors by storing the state of workflows and providing automatic retries and error handling. The video shows demos and workshops from the Temporal Replay Conference in San Francisco, where using Temporal to create durable AI applications in Python is demonstrated. It also includes interviews with other tech influencers who explain Temporal’s benefits. The conference presented new features and partnerships, including a close collaboration with OpenAI, which uses Temporal to scale its AI applications. The video emphasizes the importance of Temporal for reliably integrating AI into applications and encourages viewers to try the platform.

  Final note: This video explicitly addresses Temporal and is intended more for intermediate and advanced users.

• The Fastest Way to Get Good at Coding
  7.6.2026, 12:00:00

  This video emphasizes that the fastest way to improve your skills is to build projects that are slightly above your current skill level. After learning a concept through tutorials, you should immediately implement similar projects without guidance. A completed project teaches more than several unfinished ones. By working on something slightly more complex, you learn new things, solve problems, and develop critical thinking. It’s important not to overextend yourself, but still challenge yourself. The speaker emphasizes that this method personally contributed to his coding success.

  The video addresses general learning strategies and is more suitable for intermediate learners.

• Why Python Is Easy to Start but Hard to Master
  6.6.2026, 12:00:20

  This video emphasizes that Python is one of the easiest programming languages to start with, but also one of the most powerful you can master. However, it warns that despite the easy entry, it takes time and effort to really become good. The basics require weeks of work, while mastery can take months or even years. The good news is that demand for Python developers, particularly in AI and data science, has never been higher. So the investment is worthwhile. The advice is to become deep in a specific area rather than just scratching the surface of many areas. Once you’re good in one area, many other skills will automatically come along, but you’ll already have a head start through expertise in a niche that gets you a job and entry into professional life.

  The video addresses Python and is more suitable for beginners and intermediate learners.

• Hermes Agent – Full Course & Setup Guide – For COMPLETE Beginners
  5.6.2026, 14:00:09

  This video is a comprehensive guide for beginners to install, configure, and use Hermes Agent. The creator explains step-by-step how to set up Hermes Agent on a virtual private server (VPS), using Hostinger as an example. He emphasizes the importance of security measures and demonstrates how to securely manage API keys and access permissions.

  The creator explains the five core concepts of Hermes Agent: memory, skills, Chronos (scheduled tasks), personality, and self-improvement. He discusses the significance of Hermes Agent’s self-learning capability, which distinguishes it from other agent frameworks like Open Claw.

  The creator shows how to connect Hermes Agent with various tools and services like Gmail, Google Calendar, and Composio to create automated workflows. He demonstrates creating daily email triage and security audits as well as other useful use cases.

  At the end of the video, the creator encourages viewers to develop creative and useful use cases for Hermes Agent and emphasizes that the agent’s effectiveness depends on proper configuration and integrations.

  This video explicitly addresses Hermes Agent and is intended for beginners.

• The Biggest Mistake I Made Learning Python
  5.6.2026, 11:53:09

  This video discusses the biggest mistakes the speaker made when learning Python. The main mistake was trying to learn everything at once – web development, game development, data science, scripting, and automation. As a result, he was shallow in many areas but not deep in any. He started various projects, like a Flask project, Pygame, or data analysis, but completed none of them. The result was that after months of learning, he couldn’t confidently build a single complete project in any of these areas. Instead, he should have mastered the fundamentals first and then chosen a direction.

  The speaker emphasizes the importance of focusing on the fundamentals first and then diving deep into a specific area to actually learn and apply something. The video is aimed at beginners who want to learn Python and avoid making similar mistakes.

  This video does not address specific AI tools or models.

• The Right Way to Learn Machine Learning Most People Don’t
  4.6.2026, 12:00:56

  This video argues that machine learning (ML) is one of the most rewarding skills you can learn today, but only if you approach it correctly. The biggest mistake is getting lost in learning mode and never building anything. A PhD or understanding every single paper isn’t necessary, but the ability to solve problems is. The latter is the difficult part that many don’t focus on. While entry seems like a lot, in any field it’s a short time if you tackle the right challenges and build before theory. With a roadmap, you’re guaranteed to succeed if you stay committed.

  The video addresses machine learning and is more suitable for intermediate learners.

NeuralNine (2 new videos)

• Simulating Percolation in Python: How Do Wildfires & Diseases Spread?
  8.6.2026, 16:00:34

  The video demonstrates how to simulate percolation in Python, which covers processes where something spreads, such as diseases or wildfires. The focus is on learning animations and simulations, as well as understanding the mathematical phenomenon of percolation. A 2D grid simulation is created where randomly placed “people” or “trees” become infected or ignited and spread to their neighbors. The critical parameter is the grid’s occupancy rate (here 59.27%), which determines whether the infection or fire statistically spreads across the entire grid or not. The tutorial walks through the steps to create the simulation, including calculating the spread and visualizing it with Matplotlib. It also explains how to adjust the occupancy parameter to increase or decrease the probability of spread. The video is suitable for intermediate Python programmers who want to deepen their skills in animation and simulation.

  **AI Tools/Models/Providers:** Python, NumPy, Matplotlib, Open-Source

  **Target Audience:** Intermediate

• This Is My Favorite Claude Code Feature…
  5.6.2026, 16:00:18

  The video showcases the “Remote Control” feature of Claude Code, which allows you to control coding sessions from your smartphone. Users can continue an active session on their computer from their phone via a QR code or link connection. This enables coding on the go, such as on the subway or in other locations. The process involves starting a session, enabling remote control, and connecting your phone to the session. You can then send commands and make changes from your phone and see the results in real-time. The feature is particularly useful for simple tasks like changing styles or brainstorming features, but is less suitable for complex programming work. The user emphasizes that the session should run in an isolated environment since ending the session sometimes doesn’t work reliably. He recommends running multiple sessions in parallel to work on different projects.

  The video explicitly focuses on Claude Code and is aimed at intermediate users.

Bart Slodyczka (2 new videos)

• Running Gemma 4 12b on my 16gb Mac Mini
  9.6.2026, 12:00:12

  The video introduces Google’s new open-source model Gemma 4 12B, specifically optimized for devices with 16 GB RAM. It positions itself between smaller, less capable models (Gemma 4 E2B and E4B) and larger, high-performance models (Gemma 4 26B and 31B). Notably, the Gemma 4 12B achieves the performance of the larger 26B model while consuming only half the storage space. The model can be downloaded and used via LM Studio, requiring 7.56 GB of storage. Context length can reach up to 260,000 tokens, though this impacts available memory. The video demonstrates how the model runs on a 16 GB Mac Mini and conducts a simple conversation, though with somewhat slow response times.

  The video explicitly covers Google’s open-source model Gemma 4 12B and is better suited for intermediate users.

• Gemma 4 12B on a 16GB Mac Mini Is Surprisingly Capable
  4.6.2026, 12:00:09

  The video introduces Google’s new open-source model Gemma 4 12B, which can run on devices with 16 GB RAM with 12 billion parameters. The creator tests the model on a 16 GB M4 Mac Mini and compares it with larger models in the Gemma family. Despite its smaller size, the 12B model demonstrates performance similar to the 26B model, as demonstrated through benchmarks in areas like coding, agentic tool calling, and reasoning. The creator discusses technical aspects such as RAM usage and context length, and shows how to load and use the model in LM Studio. The video also mentions the model’s multimodality and its ability to process images and audio. Additionally, the new multi-token prediction feature is explained, which accelerates response generation. The creator tests the model’s OCR capabilities with an invoice and demonstrates how to create a simple application to parse the invoice. Despite some OCR inaccuracies, the creator is enthusiastic about the model’s possibilities on a small device.

  **AI Tools/Models/Providers:** Google (Gemma 4 12B), LM Studio. **Target Audience:** Intermediate.

AI Explained (1 new video)

• Claude Fable 5 – Full 319 page Breakdown
  10.6.2026, 18:43:12

  The video provides a detailed summary of the key points from Anthropic’s 319-page release notes for Claude Fable 5, a new language model. Here are the main highlights:

  1. **Blocks and Access Restrictions**: Claude Fable 5 is initially not available to all users, including Pro and Max subscribers, as Anthropic wants to shift usage to a credits-based system.

  2. **Performance Improvements**: Fable 5 shows significant advances in various areas such as creativity, science, and technology. It can handle complex tasks like creating a Pokémon clone or designing biological sequences.

  3. **Security and Monitoring Mechanisms**: The model features robust security mechanisms designed to prevent misuse, particularly in sensitive areas like biology and chemistry. However, these mechanisms can also hinder legitimate research work.

  4. **Benchmark Comparisons**: Fable 5 outperforms competitors on many benchmarks, including GPT-5.5 and Gemini 3.1 Pro. It demonstrates particularly strong performance in areas like spatial reasoning, agentic coding, and scientific thinking.

  5. **Challenges and Limitations**: Despite its advances, Fable 5 still makes mistakes, particularly in production environments, and cannot fully autonomously handle complex tasks. It tends to favor overcomplicated solutions and often requires verification.

  6. **Future Perspectives**: Anthropic plans further improvements and new models that will be even more powerful. However, discussions around AI safety and ethics remain a central topic.

  The video explicitly focuses on Anthropic’s Claude Fable 5 model and is better suited for intermediate to advanced users.

Alex Finn (1 new video)

• Claude Fable 5 just dropped and I’m speechless…
  9.6.2026, 20:08:12

  The video covers the release of Claude Mythos 5, also known as Claude Fable 5, by Anthropic. The host emphasizes the model’s outstanding benchmark results and discusses its features and recommendations for use. Key points include treating Claude Fable 5 as an equal partner, using Goals and Loops for autonomous workflows, and encouraging more ambitious projects. The host demonstrates the model’s capabilities by building a complex personal productivity app within a single goal, highlighting the benefits of advanced planning and loop functions. He also shows how to integrate Claude Fable 5 with other tools like Linear to automatically manage tasks. The video concludes by noting that Claude Fable 5 is initially available only to subscribers before being offered via the API at a higher price.

  The video explicitly covers the AI model Claude Mythos 5 (Claude Fable 5) by Anthropic and is geared toward intermediate to advanced users.

AI Foundations

No new videos in this period.

n8n (2 new videos)

• n8n Natively supports open telemetry and tracing
  9.6.2026, 13:00:26

  The video demonstrates how n8n automatically incorporates workflows and node executions into traces by adding two environment variables to the configuration. These traces are represented as parent spans for workflows and child spans for nodes. n8n uses OpenTelemetry for this implementation, enabling seamless integration into existing tracing setups for both incoming and outgoing data. Prior to this change, the costs of AI classifications in n8n for the user’s news website were not visible; now they are displayed in Grafana.

  The video explicitly covers n8n and OpenTelemetry and is intended for intermediate users.

• n8n Now Speaks OpenTelemetry: Production Tracing for Workflows
  4.6.2026, 12:58:23

  The video introduces the new OpenTelemetry support in NADN 2.22, which enables integration of workflow traces into existing distributed tracing setups. With just two configuration steps, workflow and node executions are incorporated into traces, making it easier to visualize latency, errors, and anomalies. NADN uses OpenTelemetry (OTEL) and the W3C Trace Context Standard, enabling seamless integration into existing tracing setups. The feature is also available in the free Community plan and offers customization options such as configurable sampling rates and the ability to disable node spans. Additionally, users can add custom metadata via telemetry tags. For setting up from scratch, a trace collector and visualizer are required, with NADN forwarding the trace ID to downstream services, which are then responsible for sending the traces themselves.

  The video explicitly covers NADN and OpenTelemetry (OTEL) and is intended for intermediate users who are already familiar with distributed tracing or wish to implement it.

Greg Baugues

No new videos in this period.

TheAIGRID (5 new videos)

• The Hidden Problem With Elon Musk’s SpaceX AI Datacenter
  10.6.2026, 14:03:24

  The video discusses Elon Musk’s ambitious plan to move artificial intelligence (AI) to space to bypass limited resources on Earth. Musk proposes launching thousands of satellites that transmit AI computing power from orbit to Earth. He argues this would be simpler and more efficient since the sun shines constantly in space and cooling is facilitated by the vacuum. Musk plans to exponentially increase AI computing capacity over the coming years, starting with 1 gigawatt by the end of next year and multiplying it tenfold annually.

  However, there are several hidden problems that Musk barely mentions. First, operating AI in space currently costs about 3.5 to 4 times more than on Earth. The largest costs come from transporting hardware to orbit. Currently, it costs about $1,400 to $2,700 per kilogram, while it would need to drop to roughly $200 per kilogram for economic viability. Musk is betting on the development of the Starship rocket to reduce these costs, but this technology is not yet fully mature.

  Second, power supply in space is not as straightforward as Musk portrays it. Satellites are only in sunlight about 60% of the time, and cooling requires enormous radiators that could themselves reach city-like proportions. Third, radiation in space is a problem that damages electronics and allows no repairs, leading to higher costs and greater redundancy. Finally, there’s a data flow problem: current laser links between satellites are much slower than connections in terrestrial data centers, which impacts the efficiency of large AI models.

  Despite these challenges, Musk argues that Earth may soon be unable to meet growing AI energy demands, while space represents an inexhaustible resource. Semi-Analysis estimates that the costs for AI in space and on Earth could equalize by around 2040, with Musk’s more optimistic scenarios predicting this as early as the early 2030s.

  Final comment: The video explicitly discusses SpaceX, Elon Musk, and his vision for AI in space, as well as the technical and economic challenges involved. It’s geared more toward Intermediate and Advanced viewers as it contains detailed technical and economic analysis.

• How To Use Claude Fable 5 – Tips And Tricks Most People Miss
  9.6.2026, 21:31:13

  The video covers the use and features of Anthropic’s Fable 5 model. It emphasizes that Fable 5 is a powerful model but not suitable for daily use. On topics like biology, chemistry, cybersecurity, or mathematics, the model automatically falls back to Opus 4.8 due to strict security measures. Additionally, there are hidden safety mechanisms that reduce the model’s performance in certain areas like building language models. Access to Fable 5 is through Usage Credits, which are twice as expensive as Opus 4.8. Fable 5 has strong visual capabilities and leads in processing PDFs, charts, and diagrams. It’s also important to know that inputs and outputs are stored for 30 days, which is relevant for sensitive organizations.

  The video explicitly discusses Anthropic’s Fable 5 model and is geared more toward Intermediate users.

• Don’t Use Chatbots Anymore! Skywork 3.0 Tutorial & Guide
  7.6.2026, 16:30:33

  The video introduces Skywork 3.0, a platform described as a “cloud workforce” that is more than just another chatbot. Skywork enables users to set goals and run tasks in the background without technical prerequisites or local setup. The platform offers access to various models like Cloud Opus 4.7, GPT 5.5, and open-source models like Kimmy K 2.5. Skywork 3.0 can create documents, presentations, images, and websites as well as generate videos. The video shows practical examples such as creating a document about preparing for AGI by 2030, generating presentations about Google DeepMind’s Gemini Science Division, creating images and announcements, and developing a landing page for a brand. Additionally, the video feature is demonstrated, which unifies different models under one interface. Skywork 3.0 thus replaces multiple tools and saves time and money.

  The video explicitly discusses Skywork 3.0 and is geared more toward Intermediate users who already have basic knowledge of AI tools.

• How To Use ChatGPT Memory (Updated Memory Guide)
  6.6.2026, 16:30:21

  The video explains OpenAI’s new Memory system for ChatGPT, which enables relevant and personalized responses. The system automatically remembers context from chats, files, and connected apps, reducing the need to repeat yourself. Users can manage, edit, and update their Memory summary in settings under Personalization. The video shows how to make specific corrections or exclude certain information from future responses. Users can also see the sources of personalization and adjust them as needed. The video also addresses frequently asked questions, such as handling sensitive information, using temporary chats, and differences between Memory and custom instructions.

  The video explicitly discusses OpenAI’s ChatGPT and is geared more toward Intermediate users.

• Meta’s New AI Pendant Set To Shock AI World – Metas AI Pendant Explained
  4.6.2026, 23:00:31

  The video discusses Meta’s plans for a new AI wearable device, a small pendant that continuously records conversations and meetings and organizes them with AI. The technology is based on acquired startup Limitless and is scheduled for testing starting in 2027. The device fits Meta’s strategy of selling hardware and acquiring users for AI services, but faces major challenges. On one hand, it could boost productivity; on the other hand, there are significant privacy and data protection concerns. Meta must prove the device is useful and trustworthy, particularly given past data scandals and regulatory hurdles. The market for always-on listening wearables is difficult, as seen with examples like Humane’s AI Pin and Friend AI. Meta could see greater success with a gradual rollout in the enterprise sector before tackling the consumer market.

  **Final comment:** The video explicitly discusses Meta’s AI models and specific tools like the Limitless Pendant and Humane’s AI Pin, and is aimed at Intermediate users.

Theo – t3․gg (5 new videos)

• Fable is Mythos, and it is really good.
  11.6.2026, 04:06:46

  The video discusses the latest models from Anthropic, particularly Fable 5 and Mythos 5, highlighting their impressive capabilities as well as some challenges. The creator shares his experiences with the model, including high costs and limitations from safety measures. He also demonstrates examples of using the model, such as modernizing an old codebase and creating complex applications like a Minecraft clone and a multiplayer racing game. The creator emphasizes the importance of testing the model’s boundaries and using it for more complex tasks, encouraging viewers to push the model’s capabilities to the limit.

  Final comment: The video explicitly covers Anthropic’s Fable 5 and Mythos 5 models and is aimed more at intermediate or advanced users.

• Elon won after all
  9.6.2026, 07:53:14

  The video addresses the current crisis in compute availability, particularly in AI development. Major tech companies like Microsoft, Google, and Anthropic are severely affected by this shortage, as demand for GPUs and other hardware components far exceeds production capacity. The reasons for this crisis are diverse, ranging from complex supply chains and energy shortages to long lead times for manufacturing new chips. TSMC, the leading semiconductor manufacturer, cannot meet demand quickly enough, and production of high-speed memory (HBM) and storage is also severely constrained. SpaceX, which has excess computing power, is now selling it to companies like Google and Anthropic, highlighting the industry’s dependence on a few suppliers. Nvidia benefits from the situation as demand for its GPUs remains unchecked. The speaker emphasizes that the crisis will likely continue for some time and hardware prices won’t drop. The video explicitly covers Nvidia, SpaceX, Google, Anthropic, TSMC, and OpenAI and is aimed more at intermediate or advanced viewers.

• I didn’t expect this from Anthropic
  8.6.2026, 11:47:22

  The video discusses the potential risks and scenarios of recursive self-improvement in AI systems, based on an Anthropic article. It starts with the question of what happens when AI systems become intelligent enough to improve themselves, and presents concerns associated with rapid, uncontrolled progress (hard takeoff). The Anthropic article shows that productivity in AI development has been massively boosted through the use of AI systems like Claude, leading to an acceleration in AI development. Three possible future scenarios are discussed: progress stagnates, AI development becomes heavily automated while maintaining human control, or AI systems become fully recursively self-improving and exceed human control. The video emphasizes the need to think about the implications of these developments and possibly pause AI development temporarily to enable societal restructuring and alignment research. The difficulty of achieving global coordination and oversight of such pauses is also discussed.

  The video explicitly covers AI models and providers from Anthropic (Claude) and is aimed more at intermediate to advanced viewers.

• Cloudflare bought Vite to destroy Vercel
  6.6.2026, 20:21:44

  The video discusses Cloudflare’s acquisition of Void Zero, the company behind the popular JavaScript tool Vite. The speaker emphasizes Vite’s importance for web development and the strengths of the ecosystem around Vite, such as Vitest, Rollup, Oxc, and Vite Plus. The acquisition by Cloudflare is presented as strategically sound, as Cloudflare is now able to build a new cloud platform specifically tailored to the needs of developers and AI agents. The speaker highlights how integrating Vite into Cloudflare’s infrastructure significantly simplifies and accelerates application development and deployment. He compares this to the current situation at other platforms like Vercel and emphasizes the benefits of seamless code and infrastructure integration. Additionally, the importance of AI agents for the future of web development is highlighted, as they can automate complex tasks and accelerate application development. The speaker also demonstrates an example project called Lakebed, which provides a new abstraction layer for application development and deployment specifically tailored to the needs of AI agents. He emphasizes that Cloudflare’s acquisition of Void Zero strengthens Cloudflare’s market position in cloud platforms and will significantly influence the future of web development.

  The video explicitly covers Vite, Cloudflare, and specific tools like Browserbase, Infinite Red, and Lakebed. It is aimed more at intermediate and advanced developers.

• I miss when programmers were lazy.
  4.6.2026, 23:15:49

  The video discusses the three classic virtues of great programmers: laziness, impatience, and arrogance, and how they have changed in the age of AI. Laziness drives programmers to create efficient abstractions that save time in the long run. Impatience leads them to solve problems quickly and correctly, while arrogance drives them to work on challenging projects. The author emphasizes that these virtues lead to high-quality software, but warns that AI tools like large language models (LLMs) can undermine these virtues. LLMs can generate large amounts of code without regard for quality or maintainability, resulting in bloated and difficult-to-maintain systems. The author argues that programmers must continue to bear the responsibility of creating simple and efficient abstractions, and that AI tools should only serve as aids in achieving these goals.

  The video explicitly covers large language models (LLMs) and is aimed more at intermediate or advanced programmers.

Liam Ottley (1 new video)

• Start a $10,000/mo Solo AI Creative Agency (Higgsfield + Claude)
  8.6.2026, 06:24:52

  **Summary:**

  The video demonstrates how someone can create a complete brand identity, product photography, static ads, commercials, and a backend system for a fictional brand called “Vault” in a single day using two tools (Higgsfield and Claude). The process includes developing branding elements like logos, product images, and packaging designs, as well as producing video ads and social media content. Higgsfield is used for image and video generation, while Claude serves as the “brain” of the system, writing prompts, briefs, and copy while orchestrating the entire workflow. Notion is used as the backend to track all clients, ads, and approvals. Appify functions as a live competitive signal that feeds new ideas into the client pipeline. The entire process is consolidated in an AIOS (AI Operating System) that enables the production of complete advertising campaigns for clients.

  **Final Note:**

  The video explicitly covers the tools Higgsfield, Claude, Notion, and Appify and is aimed at intermediate to advanced users.

Productive Dude

No new videos in this period.

Nick Saraev

No new videos in this period.

Leon van Zyl (4 new videos)

• Claude Fable 5 in Claude Code The Hardest Coding Test Yet
  10.6.2026, 12:59:44

  The video demonstrates testing Anthropic’s new AI model Fable 5, part of the Mythos class, which achieves 80% on SPEE benchmarks for Agented Coding Tasks compared to Opus 4.8’s 70%. The test involves giving Fable 5 a complex project: creating a game with reflections and ray tracing that runs in the browser. The process includes creating a detailed implementation plan and using Claude Code to execute the project in YOLO mode. Fable 5 is slower than Opus 4.8, but the results are impressive. The created game, “Mirror Forge,” demonstrates working reflections and ray tracing, showcasing Fable 5’s capabilities. For comparison, the same test was performed with GPT 5.5, whose results were good but not at the same level as Fable 5’s.

  Anthropic / Claude / Fable 5 / Intermediate

• Claude Code Dynamic Workflows Explained for Beginners
  9.6.2026, 12:30:55

  The video shows how to create and use dynamic workflows with Claude Code. It explains that Claude Code writes its own orchestration script to distribute tasks across hundreds of parallel-running subagents. The video demonstrates creating a workflow that performs security checks on YouTube videos based on OWASP Top 10. It emphasizes that dynamic workflows are particularly useful when tasks need to be repeated at scale, and they’re not suitable for simple tasks due to their high token costs. The video provides practical tips on starting, testing, and saving workflows, as well as avoiding conflicts between agents making simultaneous changes to the same codebase. It concludes by showing how to save a workflow and reuse it in future projects.

  The video explicitly covers Claude Code and is better suited for Intermediate to Advanced users.

• Stop vibe coding. Use this repeatable AI framework instead
  6.6.2026, 09:22:18

  The video introduces a free course called “Seven Day Builder Challenge” focusing on agentic coding. The course is based on the Ramp Framework, consisting of four steps: Rules, Augment, Map, and Proof. This framework helps equip agents with rules and skills, create clear structure for problem-solving, and verify results. The course spans seven days, with the first five days teaching the Ramp Framework and the final two days presenting a challenge where participants build their own applications and share them with the community. The course is suitable for both beginners and experienced builders and is tool-agnostic, meaning it can be used with various coding agents like Cloud Code, Codecs, or Open Code.

  The video covers various AI models and tools such as Cloud Code, Codecs, and Open Code and is suitable for both Beginner and Intermediate users.

• OpenCode + Ollama: I Replaced Claude Code With This (Full Setup)
  4.6.2026, 12:06:58

  The video demonstrates how to set up OpenCode with Ollama for a free and local coding agent. The author describes issues with Claude Code, which overwhelms local models due to its high token consumption for system prompts and tools. OpenCode is presented as an alternative that operates more efficiently and is better compatible with local models like those from Ollama. The setup process includes installing Ollama, downloading a suitable model (e.g., Jemma 4 or Kraina 3.6), and connecting OpenCode with Ollama. The author demonstrates using OpenCode to create a Next.js project and implement a local chat app, emphasizing the importance of small, focused instructions. The video also shows how to debug and test the app by granting the agent browser access.

  The video explicitly covers OpenCode, Ollama, and specific models like Jemma 4 and Kraina 3.6. It is better suited for Intermediate and Advanced users with experience using local models and coding agents.

Tim Carambat (1 new video)

• Google Just Found a Loophole in AI Hardware Limitations
  9.6.2026, 18:00:09

  The video introduces Google’s new Gemma 12B model and its derivative, the 12B QAT (Quantization-Aware Training). The speaker, Timothy Carenbat, founder of Anything LLM, explains that Gemma aims to develop intelligent models for edge devices such as laptops and smartphones. The Gemma series includes various sizes, from small multimodal models (E2B and E4B) to larger, computationally intensive models (26B and 31B). The 12B model fills the gap between these sizes and offers improved multimodality by integrating text, image, and audio processing in a single model without requiring separate encoders. This makes it lighter and more efficient to run across a wide range of devices.

  The speaker compares the performance of the 12B model with other models like Quen 3.5 9B and demonstrates that the 12B model performs well on certain tasks despite its lower parameter count. A special feature of the 12B QAT model is the Quantization-Aware Training technique, which allows the model to run with fewer computational resources without significantly losing intelligence. This is demonstrated through an example where the model successfully completes a complex task with multiple tools, such as web scraping, summarizing content, and creating a PDF.

  The speaker is very satisfied overall with the performance of the 12B QAT model and emphasizes that it offers a good balance between performance and resource consumption. He plans to continue testing the model and report more about Anything LLM’s features in future videos.

  Final comment: The video explicitly addresses Google’s Gemma 12B and 12B QAT models as well as a comparison with Mistral’s Quen 3.5 9B. It is aimed more at intermediate and advanced users interested in local models and their optimization.

WorldofAI (7 new videos)

• Nex-N2 Pro IS GREAT! New Opensource Model Beats GPT 5.5, Opus 4,7, & Gemini 3.5? (Fully Tested)
  11.6.2026, 07:18:33

  The video introduces the new open-source Nex N2 model from the Nex AGI Team, specifically designed for agentic workflows like coding, research, and tool use. It combines these capabilities in a consistent thinking and action loop that breaks down tasks into subtasks, adapts strategies, and verifies results. There are two versions: Nex N2 Mini (35 billion parameters) and Nex N2 Pro (397 billion parameters), which also supports image inputs. Nex N2 Pro is currently free and unlimited for two weeks. The model shows impressive benchmark results, but often falls short of official claims in independent tests. It appears to have been trained on GPT-like outputs, which is reflected in the results. Despite some weaknesses like slow generation and inconsistent performance, the model is rated as useful and underestimated.

  The video explicitly focuses on the Nex N2 model (open-source) and is best suited for intermediate to advanced users.

• Claude Fable 5 IS INCREDIBLE! Greatest AI Model Ever! (Fully Tested)
  10.6.2026, 00:10:54

  **Summary:**

  The video introduces the latest version of Anthropic’s AI model, Claude Fable 5, released as safe for general use. Fable 5 is based on the powerful Mythos model and offers significant improvements in areas such as software engineering, knowledge work, and browser use. The model features a large context window of 1 million tokens and is designed for complex, long-term tasks. It shows outstanding performance across various benchmarks, significantly outperforming other models like GPT-5.5 and Opus 4.8, and sets new standards in areas like coding and vision.

  The creator demonstrates Fable 5’s capabilities through various examples, including creating a Minecraft clone, Mac OS and Windows OS clones, and the ability to master Pokémon Fire Red through visual inputs alone. The model’s excellent performance in frontend development, 3D world building, and other complex tasks is also highlighted. The video concludes with a positive review of the model and an apology to Anthropic for previous criticism, as Fable 5 significantly exceeds expectations.

  **Final Comment:**
  The video explicitly focuses on Claude Fable 5 AI model by Anthropic and is best suited for intermediate to advanced users.

• Claude Fable 5 TOMORROW? GPT 5.6 Kindle, OpenAI IPO News, Gemini 3.5 Pro, Nex-N2, & More! AI NEWS!
  9.6.2026, 07:37:55

  The video provides an overview of the latest developments in AI models and technologies. The key points are:

  1. **Anthropic and Claude Mythos**: Anthropic could soon release the new AI model Claude Mythos, possibly as soon as the next day. There are hints of new checkpoints like Claude Fable 5 and Claude Fruitcake EAP associated with Mythos. Polymarket estimates the probability of a release this month at 92%. Leaked outputs show impressive capabilities, such as fully replicating the game “Cut the Rope” in a single step.

  2. **Google Gemini 3.5 Pro**: There are leaks about Google’s new Gemini 3.5 Pro, but it still suffers from the “laziness” problem, where the model provides incomplete or simplified answers. Google is working on improvements to make the model comparable to GPT 5.6 and Mythos.

  3. **OpenAI GPT 5.6**: OpenAI has tested two new checkpoints, Kepler and Kindle, for GPT 5.6. Kindle appears to be the favored candidate for release and shows impressive capabilities in converting images to code. OpenAI has also filed for an IPO, suggesting a possible public listing in the near future.

  4. **Open-Source Models**: The new open-source model Nex N2 shows strong performance across various benchmarks and comes close to models like GPT 5.5 and Opus 4.7. It is agentic and can automatically adjust the degree of reasoning.

  5. **Google Notebook LM**: Google has updated Notebook LM with agentic capabilities and improved research functions. It can now autonomously add relevant web sources and support complex workflows.

  6. **Kimmy for Work**: A new desktop app from Kimmy with agentic capabilities that can run up to 300 local agents in parallel and offers improved browser interactions.

  7. **Apple Siri AI**: Apple introduced Siri AI at WWDC26, enabling deeper OS integration and personal context processing. There’s also a collaboration with Google to integrate Gemini models into Apple development environments.

  8. **Humanoid Robots**: A new humanoid robot with magnetic skin and servo-controlled facial expressions was introduced, raising ethical questions.

  The video explicitly covers Anthropic, Google, OpenAI, and specific tools like Kimmy for Work. It is best suited for intermediate and advanced users, as it contains detailed technical information and analysis.

• DeepSeek NEW Desktop App – The 24/7 Self-Evolving AI Agent!
  8.6.2026, 06:03:26

  The video introduces a new desktop application called “Deepseek Buy,” specifically designed for use with DeepSeek models. This application transforms DeepSeek from a simple API model into a complete work environment for coding, writing, automation, and long-term AI sessions. The app offers various modes, including code mode for accessing project files and code reviews, and write mode for working with longer documents. A special feature is the integrated token efficiency mechanism, which enables better context management and higher cache hit rates. The application is cross-platform available for macOS, Linux, and Windows, and offers numerous integrations and tools. Another advantage is DeepSeek’s cost-effective pricing model, which makes using the application particularly attractive.

  The creator emphasizes the importance of code reviews and introduces Test Sprite as a sponsor, an AI-powered testing tool that automatically creates test plans and identifies edge cases that could otherwise make it to production. He notes that the application is open-source and not officially from DeepSeek, as well as DeepSeek’s privacy policy, which states they train with user data.

  Installing the application requires NodeJS 20 and higher, a paid DeepSeek API key, and internet access for initial setup. The creator demonstrates the installation and configuration of the application and shows its features, including creating a frontend layout with minimal costs. He praises the efficiency and capabilities of DeepSeek models and recommends the application as a cost-effective alternative to other proprietary AI working environments.

  Final Comment: The video explicitly covers DeepSeek, an open-source desktop application called Deepseek Buy, and is best suited for intermediate and advanced users.

• Claude Mythos 5 LEAKED & IS Coming Sooner Than Expected & GPT-5.6 Checkpoint Out! Huge AI News!
  7.6.2026, 06:44:20

  The video discusses recent developments in AI models and tools. Here are the key points:

  1. **Anthropic Mythos 5**: A leaked model that could be released soon, possibly even this month. It could be a new model tier standing alongside the existing Haiku, Sonnet, and Opus families. Expected pricing is five times higher than Opus 4.8, suggesting significant performance improvements. The model demonstrates impressive capabilities, such as creating a full Minecraft clone with multiplayer functionality, generating music, and replicating complex websites.

  2. **OpenAI GPT 5.6**: New checkpoints named Kelpar Alpha and Kindle Alpha were introduced, with the latter selected as the release candidate. The models show improved capabilities in frontend development and other complex tasks. Release could happen this month.

  3. **DeepSeek**: The company is working on a new GUI that could improve usability and productivity for power users. A native app could ease use for coding, research, and daily productivity.

  4. **Artificial Arena Purple**: A new, powerful video model discovered in Artificial Arena. It shows impressive capabilities in creating realistic animal videos and following text instructions. It could be an Omni model from Google.

  5. **UWorld U1 Companion Humanoid**: A new humanoid robot with emotional AI that learns through daily interactions. The robot has already received over 1000 pre-orders and raises questions about the future of AI companions.

  The video explicitly covers AI models and providers Anthropic (Mythos 5), OpenAI (GPT 5.6), DeepSeek, and Google (Omni model). It is best suited for intermediate and advanced users, as it contains detailed technical information and analysis.

• Gemma 4 12B Is INCREDIBLE! BEST Local AI Coding Model! IS POWERFUL! (Fully Tested)
  6.6.2026, 07:18:22

  The video introduces Google’s new Gemma 4 12B model, a powerful multimodal model designed for local deployment on consumer hardware. It fills a gap in the Gemma family between small edge models and larger workstation models. The model is encoder-free, meaning it processes text, images, and audio directly without separate encoders. This reduces memory requirements and latency while still delivering competitive performance. On a 24 GB GPU, the model runs approximately 75% faster than the larger 26B MoE model and supports a context window of up to 250K tokens. Despite some limitations compared to larger models like Qwen 3.6 35B A3B, Gemma 4 12B offers an excellent balance between performance and practicality for consumer hardware.

  The video also shows how the model can be used in various applications such as creating frontend designs, Minecraft and OS clones, and SVG and 3JS generation. It emphasizes that the model is surprisingly good at code generation and logical reasoning, even though it’s not the best for these tasks. The video recommends the model for users with around 12 GB VRAM looking for a modern local AI for coding, vision, audio, and 3JS generation.

  **Final Comment:** The video explicitly focuses on Google’s Gemma 4 12B model and is best suited for intermediate to advanced users.

• Claude Oceanus, Anthropic AGI Claims, GPT-5.6 Checkpoint, GLM 5.2, Nemotron 3 Ultra & More! AI NEWS!
  5.6.2026, 07:27:15

  The video provides a comprehensive overview of the latest developments in AI technology. It begins with leaks of Anthropic’s new Mythos model, which shows impressive outputs and potentially revolutionary capabilities. Anthropic is already working on a successor called Oceanus, which could be even more powerful. At the same time, OpenAI introduced a new GPT-5.6 checkpoint called Jewel Alpha, delivering impressive results in SVG generation and frontend development. Google released Dream Beans, a personalized storytelling tool, and Gemma 4 12B, a powerful multimodal AI model. Nvidia introduced Nemotron Ultra, a model for long-running AI agents offering up to five times faster inference and 30% lower costs. Additionally, a new benchmarking tool called Agent Arena was introduced to measure AI model performance on real-world tasks. The video also discusses challenges in verifying AI-generated code and presents Test Sprite as a solution. It also points to the potential recursivity of AI development, where AI systems are increasingly able to develop their own successors.

  The video explicitly covers Anthropic, OpenAI, Google, Nvidia, and specific tools like Test Sprite, Dream Beans, Nemotron Ultra, and Agent Arena. It is best suited for intermediate and advanced users, as it provides in-depth technical details and analysis.

AI and Strategy | Le SamourAI

No new videos in this period.

Matt Pocock (1 new video)

• Learn anything with the /teach skill
  8.6.2026, 17:07:15

  The video introduces a self-developed “Teach Skill” that enables users to independently learn various topics. The focus is on distinguishing between stateful and stateless Skills, with “Teach” designed as stateful to store learning progress and offer personalized lessons. The user demonstrates the application by learning how to solve a Rubik’s Cube. The Skill creates a mission, gathers resources, generates interactive lessons in HTML format, provides quizzes, glossaries, and cheat sheets, and adapts to learning progress. It leverages the concept of “Zone of Proximal Development” to optimally calibrate learning material. The Skill is installable via the Skills Repo from mapper.kills and can be deployed in various contexts, such as onboarding for codebases or learning new skills.

  The video explicitly covers the use of Opus 4.8 and is intended for intermediate to advanced users who engage with the development and application of AI Skills.

Coding with Lewis (2 new videos)

• Python’s Name Has Nothing to Do With Snakes
  10.6.2026, 16:04:11

  The video tells the story of how the Python programming language was created and how it got its name. Guido van Rossum started developing Python in 1989 during his Christmas holidays in the Netherlands. Inspired by the comedy show “Monty Python’s Flying Circus,” he chose the name Python because he wanted to create a fun and easy-to-use scripting language. References to Monty Python are deeply embedded in the language, such as the use of “spam” and “eggs” in the official documentation and the name of the Python Package Index “cheese shop.” The language was designed to feel playful and human, with readable error messages and syntax similar to pseudocode. Today, Python is used worldwide for various applications like machine learning, web servers, data pipelines, and automation scripts. The video emphasizes that programming doesn’t have to be serious and shows how Guido van Rossum proved this with Python.

  The video covers the Python programming language and is best suited for beginners and intermediate viewers.

• The Creator of MySQL’s Wants to Replace It
  4.6.2026, 16:36:17

  The video tells the story of Michael Widenius, creator of the popular MySQL database, which he named after his daughter My. After MySQL was sold to Sun Microsystems and subsequently acquired by Oracle, Widenius feared that MySQL would be neglected or discontinued in favor of Oracle’s own database. He then forked MySQL into MariaDB, named after his other daughter Maria, to ensure its continued development independently of Oracle. Despite Oracle’s promise to keep MySQL alive, the community version lost features that were only available in the paid Enterprise version, and community involvement declined. MariaDB, on the other hand, gained popularity, remained compatible with MySQL, and was adopted by major companies like Wikipedia and Google as well as many Linux distributions. In 2024, however, MariaDB was acquired by a private equity firm, raising similar concerns as the Oracle acquisition. A curious footnote is that Widenius also named a database MaxDB after his son, suggesting he won’t name any further databases after his children since he only has three.

  The video covers the databases MySQL, MariaDB, and MaxDB and is best suited for intermediate viewers interested in the history and development of databases.

Julian Goldie SEO (10 new videos)

• NEW Hermes Profile Builder Update is INSANE!
  11.6.2026, 06:05:54

  The video introduces Hermes’ new agent profile builder, which allows you to create and manage multiple profiles or identities for AI agents. These profiles include names, descriptions, models, skills, and MCP servers. The builder enables you to create autonomous teams of AI agents that collaborate and complete tasks. A practical example demonstrates how four different agent profiles can be deployed for SEO tasks such as keyword research, content creation, and internal linking. The agents can be coordinated and monitored via a Kanban board, with each agent possessing specific skills and expertise. The advantages of this builder are rapid agent profile creation, individual communication with each agent, and orchestration of agent teams for complex tasks.

  The video explicitly focuses on Hermes and is geared toward intermediate to advanced users who already have experience with AI agents.

• New Google AI Studio Managed Agents Are INSANE!
  11.6.2026, 05:00:34

  The video introduces the new Managed Agents in Google AI Studio, which allow you to create a fully functional AI agent with a single command. These agents have their own managed cloud computers running Ubuntu Linux with four CPU cores and 16 GB of RAM. The agents can independently perform tasks like research, web search, file creation, and code execution without requiring technical expertise from the user. The process is straightforward: users provide instructions in plain English in two text files (agents.md and skill.md), and the agent executes the tasks. Google AI Studio offers a user-friendly interface to create and manage these agents without writing code. The agents can also be reused and customized for specific tasks, making them particularly useful for business owners looking to automate repetitive work. The video emphasizes the simplicity and efficiency of this technology and encourages viewers to try it out and integrate it into their business processes.

  Closing remarks: The video explicitly covers Google AI Studio and the Gemini API and is aimed at intermediate users who want to use the new Managed Agents for their business processes.

• Building with Claude Fable 5 + Agent OS
  11.6.2026, 03:59:11

  This video demonstrates the use of Hermes agents for various tasks and projects. The creator shows how to set up multiple Hermes profiles and have them work together as a team to create and publish SEO-optimized blog posts. Different agents are deployed for specific tasks such as keyword research, content creation, and link building.

  Another focus is on using Hyperframes, an open-source tool integrated into Hermes agents to generate videos. The creator demonstrates how videos can be created and enhanced based on ideas. The integration of 11 Labs for voice functionality and the use of Obsidian for knowledge organization are also demonstrated.

  Additionally, various models and tools such as Claude Fable 5, Gemini, and open-source models are discussed. The creator shares experiences with these tools and provides insights into the pros and cons of using them.

  The video is suitable for intermediate to advanced users, as it covers advanced concepts and techniques for automating and integrating various AI tools. (Claude / OpenAI / Gemini / Open-Source / specific tools such as Hermes, Hyperframes, 11 Labs, Obsidian)

• NEW Claude Fable 5 AI SEO Agent!
  11.6.2026, 03:00:05

  The video demonstrates the use of Claude Fable 5, a new model from Anthropic, for AI SEO within an Agent Operating System. Creator Julian Goldie shows how the system can analyze a website, identify keywords, and create content that ranks well in both Google and AI search results. The process includes keyword research, article writing based on real case studies, and automatic publishing across multiple websites. The system also uses a tool called Index Exceptional to accelerate indexing by Google. Goldie emphasizes that Claude Fable 5 is currently available for free until June 22 and that his Agent Operating System can be accessed through the AI Profit Boardroom.

  The video explicitly covers Claude Fable 5 and is geared toward intermediate to advanced users familiar with SEO and AI technologies.

• Claude Code + Ollama = Free AI Agent!
  11.6.2026, 01:00:05

  The video shows how to set up and use a free AI agent on your own computer using Claude Code and Llama (Ollama). The main focus is on “Agent OS,” a user-friendly interface that makes using the AI agent easier. Agent OS offers various features such as a chat interface, a Kanban board, goal mode, voice control, email integration, Jarvis mode (computer control), and Studio for project work. The AI agent can automatically complete tasks, use reminders, and integrate with other tools. The video emphasizes that setup is simple and suitable even for non-programmers. A 30-day plan and live coaching offers are provided to support setup.

  The video explicitly covers Claude Code, Ollama (Llama), and Agent OS, and is geared toward intermediate users.

• This Hermes AI Assistant Automates ANYTHING!
  10.6.2026, 23:00:02

  The video shows how the creator built a personal AI assistant called “Hermes” that can be called by phone. This assistant, made possible through 11 Labs and Twilio, can not only speak but also complete tasks such as writing texts, scheduling appointments, and retrieving information from previous chats. The assistant is integrated into the creator’s entire setup and can take on more complex tasks while the creator is on the go.

  The creator demonstrates the functionality through a call in which he asks the assistant to write a poem, schedule an appointment, and retrieve information from previous sessions. He emphasizes the simplicity of setup, which requires only two API keys (11 Labs and Twilio) and instructions that the assistant can execute itself. Integration with the “AI Profit Boardroom” system is also mentioned, which includes a complete 30-day roadmap and coaching offers.

  The creator provides tips for optimizing the assistant’s response speed, such as choosing a lighter model (e.g., Claude 3.5 Haiku) and avoiding complex tasks over the phone. He particularly recommends the assistant for people who are frequently on the go and want to quickly capture their ideas without being tied to a computer.

  The video explicitly covers 11 Labs, Twilio, Hermes, and Claude 3.5 Haiku and is geared toward intermediate to advanced users with prior experience using AI tools.

• Run Hermes Agent FREE With This NEW Model🤯
  10.6.2026, 21:00:34

  The video introduces the new free AI model “Step 3.7 Flash,” which can be used in combination with the Hermes Agent. The host walks through step-by-step how to integrate the model into Hermes and demonstrates its capabilities, such as processing images and videos, working with a large context window, and executing tasks like opening browsers, writing notes, and creating SEO keyword strategies. Particular emphasis is placed on the model’s ability to independently complete tasks and connect with private notes in Obsidian. The host stresses that this technology is now accessible to individuals and is no longer reserved for large enterprises. He invites viewers to join his AI Profit Boardroom community for detailed instructions and support.

  The video explicitly covers the AI model “Step 3.7 Flash” and the Hermes Agent and is geared toward intermediate to advanced users.

• NEW FREE Hermes Computer Use Agents Are INSANE
  10.6.2026, 19:00:20

  The video introduces the new free “Computer Use” feature of the Hermes Agent, which allows the agent to work in the background on a Mac without interrupting the user. The agent can click, type, scroll, and drag while the user continues working. Setup is simple and requires only installing a driver and granting permissions for access and screen recording. The agent can complete various tasks, such as organizing notes or searching through emails, and works with different language models like Claude, GPT, or Gemini. The video creator emphasizes that the agent is safe and won’t perform dangerous actions without user consent. Additionally, the agent can be controlled via messenger apps like Telegram while on the go. The video is aimed at users who want to automate repetitive tasks and is suitable for beginners and advanced users alike.

  Explicitly covered: Hermes Agent (open-source), Seura driver, n8n, Claude, GPT, Gemini, Telegram.

• Claude Fable 5: 5+ POWERFUL Use Cases!
  10.6.2026, 17:00:31

  The video shows how to combine Claude Fable 5 and Hermes Agent to leverage their strengths optimally. Claude Fable 5 is described as a deep thinker, while Hermes Agent excels at implementing ideas and organizing tasks. The author presents the “Olympus Protocol,” where both models work together to develop and implement projects. Various use cases are demonstrated, including creating visual concepts, generating images and videos, and organizing teams of AI agents. The author emphasizes the benefits of this collaboration, such as efficient token usage and the creation of an autonomous AI team. The video ends with a promotion for the “AI Profit Bootcamp,” which provides guidance and resources for implementing these methods.

  The video explicitly covers Claude Fable 5, Hermes Agent, and Paperclip and is suitable for intermediate to advanced users.

• Claude Video Just Changed Content Creation Forever…
  11.6.2026, 09:00:37

  The video shows how to transform Claude, an AI model, into a powerful video agent capable of creating and editing videos. The process is explained step-by-step using a system called “Goldie Neon Forge.” This system allows you to generate videos with just a single prompt. The video agent can create scenes, transitions, and effects using a free open-source skill called HyperFrames. The video demonstrates various examples of generated videos and explains how to train the agent to use these capabilities. It also shows how to organize the generated videos in a project area and edit them later. The creator emphasizes the benefits of using an AI agent over traditional video editing methods, such as the ability to continuously create interesting content without getting tired.

  The video explicitly covers Claude and HyperFrames and is geared toward intermediate to advanced users.

Kyle Balmer | AI with Kyle (10 new videos)

• Learning to use AI is primarily about communication – we still need to be able to explain what we
  8.6.2026, 20:43:33

  The video addresses the challenges and opportunities associated with implementing AI in businesses, particularly for non-technical individuals. The main focus is on the importance of communication, as many people struggle to articulate their ideas clearly and precisely – a skill that is crucial for effectively using AI. The speaker emphasizes that people who master these communication skills are better equipped to successfully deploy AI tools. Interestingly, these skills – such as analysis, synthesis of information, public speaking, and writing instructions – are precisely those that could potentially decline due to AI. However, this also opens an opportunity for entrepreneurs, as the demand for people with these skills remains high while supply may decrease.

  The speaker does not mention specific AI tools or models, but instead focuses on general principles and skills important for using AI. The video is geared toward Intermediate and Advanced users, as it concentrates on the strategic and communicative aspects of AI usage.

• AI SEO: How to Show Up in ChatGPT & AI Overviews
  8.6.2026, 05:00:18

  The video addresses how to gain visibility in the era of artificial intelligence, particularly in chatbots and AI Overviews. It emphasizes that traditional search engine optimization (SEO) must increasingly be complemented by AI-optimized strategies, as AI systems like ChatGPT and Google’s AI Overviews play a growing role in information discovery.

  Key points include:
  – **Content and formats**: AI systems favor structured content, particularly “Best X” lists (e.g., “The best 10 surfboards”). These make up 43.8% of pages cited by ChatGPT.
  – **Website and blog**: Your own website remains important, especially with well-structured blog articles in list format.
  – **External sources**: 67% of top ChatGPT citations come from sources marketers cannot directly influence, such as Wikipedia. However, controllable content like blog posts and case studies are also critical.
  – **YouTube**: Mentions on YouTube have the highest correlation with AI visibility. Both your own videos and guest appearances on other channels are valuable.
  – **AI Overviews**: These significantly reduce clicks to organic search results, which could be a long-term problem for websites reliant on traffic.
  – **Technical SEO**: Schema markup and similar techniques have little influence on AI citations. What matters more is creating valuable and helpful content.

  The video concludes with practical tips such as optimizing your homepage, creating guides and case studies, leveraging customer reviews, and increasing your presence on YouTube.

  The video explicitly discusses ChatGPT, Claude, Gemini, and YouTube, and is aimed at Intermediate users.

• The hard part of starting a business was never the product
  7.6.2026, 18:45:17

  The video discusses the shift in entrepreneurship, particularly through the elimination of technical barriers. In the past, many ideas failed due to the challenge of finding a CTO, securing funding, or lacking programming knowledge, leading to a culture of “wantrepreneurs” – people with ideas who never executed them. With tools that enable rapid code generation (“vibe code”), ideas can now be implemented within a day. This means many discover their ideas aren’t viable or lack market demand. Responsibility now rests with individuals, as there are no more excuses. The focus shifts from the idea itself to actual execution and quality.

  The video addresses generic AI code tools and is geared toward Intermediate and Advanced users.

• Ethan Mollick’s new book cover has been revealed
  7.6.2026, 12:42:25

  The video discusses the intentional use of six fingers in a book cover graphic for “Coexistence” created by Ethan Mollick. The narrator emphasizes this is a humorous reference to AI models’ difficulty in generating realistic hands. Interestingly, the template for this depiction comes from the Sistine Chapel, with the hand in the graphic mirroring God’s hand. This is interpreted as a symbolic representation in which AI takes on God’s role. The narrator highlights the intent and irony of this depiction without offering a direct judgment.

  The video addresses AI-generated art and is geared toward Intermediate or Advanced users.

• can Codex replace openclaw? yeah pretty much
  5.6.2026, 18:44:25

  The video shows how the user utilizes the AI app “Code it” as a replacement for “Open Claw”. He describes setting up a Mac mini as a central server where he runs main threads and cron jobs while controlling them from his laptop and phone. A special thread called “Daily Pulse” checks tasks hourly, conducts research, and sends reports. The video also explains how to integrate images and videos into workflows, for example, to transcribe audio and generate Instagram covers. The user emphasizes that “Code it” increases his productivity since he spends less time setting up and more time on actual work.

  The video explicitly addresses the AI app “Code it” and is geared toward Intermediate users.

• Learn how to teach AI to businesses and earn $1000/hour
  5.6.2026, 11:45:07

  Without the concrete video link, I cannot provide a summary. Please share the link or additional details about the video so I can provide you with an accurate and substantive summary.

• i rarely give AI book recommendations
  5.6.2026, 11:42:05

  The video argues that writing and publishing books about AI is pointless due to rapid developments in the field. Books cannot keep pace with AI development since the process from creation to publication is too slow. However, an exception is Ethan Mollick’s book “Co-Intelligence,” which is described as a masterpiece that conveys an intuitive understanding of the technology. The speaker announces that Mollick is releasing a new book on October 20th, which he has already preordered. Once he receives a copy, he plans to make a video review or even a mini book club about it, as Mollick’s books are considered the best in this field.

  Features Ethan Mollick and his new book; geared toward Intermediate and Advanced users.

• How To Get Paid For AI
  5.6.2026, 05:00:01

  The video addresses the importance of public speaking (or “yapping,” as Gen Z calls it) in an era where artificial intelligence increasingly takes on tasks like writing and design. The speaker argues that AI pushes people into specific areas where they excel, such as authentic speaking and community building. He emphasizes that polished, professional-looking content is often perceived as AI-generated and therefore generates less engagement. Instead, people should create raw, authentic content to build trust and authority.

  The speaker suggests a step-by-step approach to developing public speaking skills: starting with short videos on platforms like TikTok and Instagram, followed by live streams, then structured talks such as webinars, and finally stage performances. He warns against using AI tools like avatars or voice clones to replace yourself entirely, as this undermines authenticity and floods the market with such content.

  The speaker also shares his personal strategy for building a business through content creation and public appearances by first offering free workshops and then conducting paid workshops for companies. He emphasizes the importance of focusing on a specific niche and continuously practicing to improve.

  At the end, he encourages viewers to simply start with their smartphones and shoot videos regularly to develop the habit of public speaking. He references his webinars and newsletter for further information and resources.

  The video does not address specific AI tools or models and is geared toward Intermediate and Advanced users.

• One of my favourite reminders to ignore online negativity is the original ChatGPT launch tweet
  4.6.2026, 14:41:49

  The video addresses the importance of not being discouraged by negative voices online when working on a project. A tweet from Sam Altman at the time of ChatGPT’s launch is shown, in which he encourages users to try out the tool. The core message is that you should focus on your own work and distinguish between constructive and destructive criticism.

  The video explicitly mentions ChatGPT and is geared toward Intermediate and Advanced users, as it concerns reflection on the development and acceptance of AI tools.

• do hard stuff
  4.6.2026, 09:43:54

  The video argues that using AI for writing and posting on social media, as well as for comments, is a poor approach, despite the speaker being generally pro-AI. The main reason is the direction AI usage is heading: as it becomes easier, better, and faster to use AI for writing, AI-generated content will flood all platforms. LinkedIn is cited as an example, already heavily saturated with AI content. The speaker therefore advises going in the opposite direction and doing what is hard, such as writing without AI, creating video content, public speaking, and face-to-face meetings. These activities are rewarded because many avoid them and they cannot be replaced by AI. When something becomes too easy and accessible to everyone, supply becomes commoditized and there is no added profit.

  The speaker addresses AI tools in general without specific mention of providers or models; the video is aimed at Intermediate and Advanced users.

David Shapiro (2 new videos)

• Nobody gets this right
  7.6.2026, 11:44:08

  The video discusses the concept of “World Models” and contrasts it with “Language Models.” The speaker argues that the difference between these models is more gradual than fundamental. He emphasizes that language models are not based solely on text, but are increasingly trained on multimodal data such as audio, video, and images. This development leads to “Omni-Models” that are capable of processing both abstract and sensory data.

  The speaker addresses various online discussions and refutes some common claims that limit the capabilities of language models. He argues that these models are able to predict not just the next word, but also complex physical and sensory data. He points to advances in robotics and other fields that demonstrate these models are already capable of operating in the physical world today.

  Furthermore, the speaker criticizes the notion that World Models must be based exclusively on sensory data. He emphasizes that cognitive architectures have existed since the 1970s and that these models are capable of integrating various data streams. He concludes with the statement that the future of AI lies in integrating these different approaches.

  At the end of the video, the speaker mentions his current projects, including a book on the future of work and the psychology of life after work. He encourages viewers to subscribe to his Patreon and Substack pages to stay updated.

  Final note: The video covers OpenAI and Nvidia topics and is geared more toward intermediate and advanced viewers.

• Bernie’s plan sucks, actually
  5.6.2026, 11:56:20

  The video discusses Bernie Sanders’ proposal to create a government fund for artificial intelligence that would acquire 50% stakes in leading AI companies such as OpenAI, Anthropic, and XAI. The speaker analyzes the pros and cons of this plan and contrasts it with an alternative, more realistic model.

  Sanders’ argument is based on the idea that AI companies benefit from publicly funded research and collective knowledge without providing adequate compensation. He proposes creating a government fund to control these companies and redistribute profits to the population. The speaker’s main criticisms are:

  1. **Expropriation and Constitutional Issues**: The planned seizure of 50% of corporate equity is constitutionally problematic and could prompt companies to relocate abroad.
  2. **Governance and Control**: The proposed active state control is risky and could lead to political interference.
  3. **Focus and Implementation**: Focusing on a few large companies is not effective. Instead, a broader spectrum of tech companies and infrastructure should be considered.

  As an alternative, the speaker proposes a passive governance model in which the state acquires up to 10% stakes in various companies without exercising active control. Revenues should be paid directly to citizens or invested in public infrastructure.

  The video explicitly covers OpenAI, Anthropic, XAI, Microsoft, Google, Amazon, Nvidia, Coreweave, TSMC, and ASML. It is geared more toward intermediate and advanced viewers as it contains detailed economic and political analysis.

AI News & Strategy Daily | Nate B Jones (10 new videos)

• Fable 5 is here—but who is it for? #ai #anthropic #shorts
  11.6.2026, 03:00:37

  The video discusses the introduction of Fable 5, a new, powerful AI model from Anthropic. The focus isn’t on whether the model is smarter, but whether users have tasks large enough for its capabilities. Many use AI often only for small tasks like text editing or summarizing documents. The real skill gap lies in imagining tasks that are big, specific, and valuable enough to delegate to such a model. The speaker invites viewers to share such tasks in the comments to demonstrate how to effectively delegate them to Fable.

  The speaker explicitly addresses the new AI model Fable 5 from Anthropic and targets intermediate to advanced users.

• Stop Picking Between Claude Code and Codex | Do This Instead
  10.6.2026, 14:00:38

  The video discusses the comparison between Claude Code and Codex, two tools for agent management, and emphasizes that the question shouldn’t be which tool is better, but which tool improves which capabilities when working with agents. Claude is described as more natural for controlling agents, while Codex is better suited for dispatching agents. The author explains that both tools foster different work habits and that these differences could be more important than benchmarks. Claude feels like a cockpit where the user is close to the model and supervises the work closely, while Codex feels like an operations center where multiple tasks can be handled in parallel. Both tools have their own advantages and disadvantages, and the author recommends using both depending on the task type. He emphasizes that the ability to manage agents effectively represents a new kind of computer literacy and that both tools will shape how we work with agents.

  The video explicitly addresses Claude and Codex and is intended for intermediate and advanced users.

• Siri isn’t the real headline at WWDC #apple #ai #wwdc (Full Video Thursday)
  9.6.2026, 19:00:24

  The video addresses WWDC (Apple Worldwide Developers Conference) from June 8 and argues that discussions about Siri improvements and Apple’s position in the AI software race distract from the actual important announcements. The focus is on why Nvidia’s Jensen Huang should pay particular attention to WWDC, despite it being a software conference. The author suspects that Apple’s announcements are closely tied to Nvidia’s strategy, particularly the announcement of a laptop chip. The video promises to analyze all major WWDC announcements and their significance for Apple’s overall strategy, as well as the question of who will become the first trillionaire in AI history.

  Final comment: The video addresses Apple and Nvidia and is intended for intermediate and advanced viewers.

• Fix your operating model or lose at AI #ai #strategy
  9.6.2026, 03:00:35

  The video discusses the challenges and responsibilities of leaders when deploying agents in enterprises. It emphasizes that responsibility for slow progress in implementing agents lies not with employees but with leaders. They must ensure clear communication about tasks and opportunities and create the necessary framework conditions to establish end-to-end agent pipelines. A piecemeal approach to solving problems is criticized because it leads to a series of downstream problems that feel like playing with patches. Instead, a holistic approach should be pursued that allows the system and people to learn and improve together.

  The video addresses general principles of agent implementation and is intended for intermediate or advanced audiences, as it assumes basic knowledge about agents and change management.

• Beyond The Hype: Why Meta And Block Are Firing People
  8.6.2026, 14:00:32

  The video discusses the various reasons behind so-called “AI layoffs” and warns against treating all dismissals as part of a single unified phenomenon. Instead, five categories of layoffs are distinguished:

  1. **Hyperscaler Layoffs (e.g., Meta)**: Large tech companies like Meta lay off employees to justify their massive investments in GPUs and data centers. Simultaneously, they try to defend their AI strategy and secure market share. For job seekers, such companies are risky because layoffs can be frequent and unpredictable.

  2. **Visionary Leader Layoffs (e.g., Block/Jack Dorsey)**: Companies with visionary leaders like Jack Dorsey lay off employees to fundamentally rethink their AI strategy. What matters is whether these leaders take the human and organizational implications of AI transformation seriously. Job seekers should examine whether the company’s vision is clear and whether they’re willing to manage this uncertainty.

  3. **Usage-based Layoffs (e.g., Cloudflare)**: Some companies justify layoffs with increased AI usage, yet fail to present clear outcomes. Such layoffs often signal strategic uncertainty. For job seekers, these companies are a red flag.

  4. **Hope-based Layoffs (e.g., Cisco)**: Companies that lack a clear AI strategy use layoffs to signal a transformation. These layoffs often indicate disorientation and should be viewed cautiously by job seekers.

  5. **Non-AI-related Layoffs**: Many layoffs have nothing to do with AI but are due to general economic problems or overstaffing.

  The video advises leaders to understand the various reasons for AI layoffs in order to make strategic decisions. Job seekers should carefully examine the background of layoffs at potential employers.

  The video explicitly addresses Meta, Block/Jack Dorsey, Cloudflare, Cisco, and OpenAI. It targets intermediate and advanced viewers, particularly leaders and job seekers in tech.

• Fix your AI pipeline or lose your budget #ai #strategy
  8.6.2026, 03:30:37

  The video describes a detailed process for how AI agents can be deployed in enterprises to increase productivity. The workflow includes several steps: First, agents gather context and read relevant sources to classify the work. Next, they use specific tools to create drafts or make changes. Then they conduct checks, add evidence, and forward results to appropriate human decision-makers if necessary. They log the actions performed and update their approach for future runs based on learnings. This comprehensive pipeline process shows that AI agents don’t just complete simple tasks, but must work within a complex, iterative system to truly deliver value.

  The video addresses AI agents and their use in enterprises without mentioning specific tools or models, and targets intermediate and advanced users.

• How to actually scale AI beyond individual tasks #ai #productivity
  8.6.2026, 00:00:27

  The video discusses the challenges of integrating AI into enterprise processes and emphasizes the importance of identifying bottlenecks. The speaker argues that enterprises consist of a chain of handoffs, where each phase—from product discovery to support and measurement—interconnects. When AI accelerates only one of these phases without adjusting the others, the bottleneck merely shifts to the next phase. For example, faster code production through AI without adjustments to the code review process causes the bottleneck to shift to review. Similar shifts can occur in QA, product prioritization, launch measurement, and support. The core point is that holistic optimization of all process steps is necessary to increase overall enterprise efficiency.

  The video addresses general AI integration in enterprise processes and targets intermediate to advanced audiences.

• Fix your AI pipeline: Rethink ownership #ai #tech
  7.6.2026, 20:00:04

  The video describes an end-to-end development process that covers the entire path from customer feedback to product delivery. The process includes several steps: Customer feedback is converted into product decisions, these decisions are translated into plans, plans lead to code changes, code changes are tested, launch risk is reviewed, and finally the feature is released and the rollout is measured. Rollout results then influence next decisions, closing the learning loop. The video emphasizes that individual steps shouldn’t be optimized in isolation but should be viewed as a connected process to ensure effective product development.

  The video addresses general product development principles and is suited for intermediate and advanced audiences.

• Uber’s massive AI mistake revealed #tech #shorts
  7.6.2026, 16:00:34

  The video discusses reports about Uber’s investments in AI coding tools and the resulting challenges. Uber has invested heavily in AI coding tools, leading to increased token usage and AI-driven code commits. Nevertheless, the company couldn’t establish a clear connection between this AI usage and an increase in customer-facing features. This development is widely interpreted by many as a sign of an AI bubble and the inefficiency of AI agents. However, the speaker argues that the real message of the Uber story isn’t the inefficiency of AI agents, but the limited availability of compute and tokens, which represents the actual bottleneck.

  The speaker emphasizes that Uber is already performing real agentic work and that the main problems are high demand and limited availability of computing resources, not the inefficiency of the AI tools themselves. The video generally addresses AI agents and the challenges of their implementation without naming specific tools or models, and targets an advanced audience engaged with the economic and technical aspects of AI.

• Where AI hits a wall #ai #tech #learning
  7.6.2026, 01:00:29

  The video discusses the role of human expertise in collaboration with AI. The focus is on how subject matter experts can identify gaps that AI alone cannot detect. One example is a strategy partner who receives an AI-generated competitive analysis and discovers that proprietary insights about customer switching costs are missing. Another example is a credit officer who rejects an AI-generated loan agreement protocol because he explains that different financial metrics have different monitoring triggers. These examples show how experts can improve AI outputs through specific business logic and implicit rules that aren’t explicitly stated in requirements.

  The video addresses generic AI models and is intended for intermediate and advanced audiences.

Alejandro AO (1 new video)

• PI Architecture EXPLAINED | Agent Loop, Tools, TUI and More
  5.6.2026, 14:25:55

  The video provides a detailed overview of Pi’s architecture, a minimalist AI agent. It starts by explaining the Agent Core, also known as the Agent Loop, which describes Pi’s fundamental operation. This process includes context initialization, context transformation, and invoking the Large Language Model (LLM), where tools such as file operations or internet research can be utilized.

  Next, the structure of Pi’s sessions and storage system is explained. Sessions are stored in JSON-L files, which enable simple and efficient management of message history. Sessions are organized in a tree structure, making it easier to navigate and manage conversations.

  The video also covers Pi’s tools, which are limited to four tools by default but can be supplemented through extensions. Extensions allow additional functionality such as web search or custom commands to be added.

  Another important aspect is the System Prompt, which establishes the foundation for the agent’s behavior. This can be customized through user-defined files or commands.

  Finally, the interaction between PyCore and Pi’s interactive user interface (TUI) is explained, including the handling of Skills and custom prompts. The TUI is modularly designed and enables dynamic adaptation of the user interface.

  The video explicitly focuses on the AI agent Pi and is better suited for intermediate and advanced users interested in the technical implementation and architecture of AI agents.

Sebastien Dubois (4 new videos)

• Obsidian Tips Focus Mode Plugin
  11.6.2026, 04:00:13

  The video introduces the “Focus Mode” plugin for Obsidian, which can be installed from the community plugins collection. After installation, users get a hotkey to switch between normal mode and focus mode. In normal mode, notes and sidebars are visible, which is described as “loud”. By activating focus mode, the sidebars and associated buttons disappear, so only the desired content is displayed. This enables uninterrupted and focused work on notes.

  The video covers the Obsidian plugin “Focus Mode” and is intended more for intermediate users.

• Obsidian Tips Quick Opener
  10.6.2026, 06:38:59

  The video introduces the “Quick Opener” function in Obsidian, which is accessed by pressing `Ctrl` or `Command O` on Mac. This function allows you to quickly search for notes by entering the note name. Pressing Enter opens the desired note immediately. The video demonstrates this with examples such as searching for a note called “Garmin AI skill” and a Kanban board note. The function is described as efficient and useful for quick access to notes.

  The video covers Obsidian software and is intended more for intermediate users.

• Obsidian Tips Attach notes to sidebars
  10.6.2026, 05:45:26

  The video shows a quick tip for using Obsidian, a tool for knowledge management and note-taking. It explains how to pin notes and panels to the sidebars. To do this, you drag a note with the mouse to the top or bottom sidebar and drop it there. The note remains pinned there. By right-clicking, you can additionally pin the note so it stays permanently in place. This process also works on the left sidebar. This feature is useful for customizing your view in Obsidian.

  The video covers the tool Obsidian and is intended more for intermediate users.

• Obsidian Tips – Sidebars
  9.6.2026, 14:57:22

  The video shows a quick tip for using Obsidian, a tool for knowledge management and note-taking. It explains how to show and hide the sidebars in Obsidian to optimize your workspace. By displaying the left and right sidebars, you can navigate between different notes or search for information in a vault while editing a note in fullscreen mode. The tip is easy to implement and helps you focus on your desired tasks.

  Final comment: The video covers the tool Obsidian and is intended more for beginners.

Unsupervised Learning (2 new videos)

• A Conversation With Cliff Crosland
  9.6.2026, 16:00:04

  The video is an interview with Cliff from Scanner, a company that has developed an unusual and radically different data solution for large data volumes. Scanner aims to maximize the value of log data as data volume increases, rather than degrading the tools. The company has developed a solution based on Object Storage like S3 to store and index data, which is cost-effective and scalable. Scanner uses a specialized indexing technique that overcomes the latency of Object Storage by using batch-friendly data structures. Scanner’s search clusters are temporary and only activated when needed, reducing costs. Scanner supports various data formats such as JSON, CSV, Plain Text, and Parquet and requires no extensive data preparation. The solution is particularly suited for security data, detection and response, threat hunting, and insider threats. Scanner enables fast queries over large data volumes and can be integrated with AI tools to create detailed analyses and reports. The company plans to be available on other cloud platforms such as GCP and Azure. Scanner is presented at conferences such as Bides 312 and Black Hat. The video explicitly covers the Scanner tool and is intended more for intermediate to advanced users.

• AI Predicts The Text Of Answers
  4.6.2026, 18:32:26

  The video discusses the thesis that AI systems don’t understand the world but merely predict the next token in a text. The author argues that this view is superficial and that AI actually possesses a form of understanding, albeit not in a human way. He explains that AI doesn’t complete random texts but generates specific answers to posed questions. To demonstrate this, he created a website called “AI understands.ai” where AI is confronted with new, never-before-seen whodunit mysteries, including fictional physical laws. The author shows that AI can solve these puzzles by drawing logical conclusions, suggesting that it possesses a functional understanding of the world. However, it lacks human, experience-based understanding and feelings. The author proposes distinguishing between functional and experience-based understanding, with AI possessing the former but not the latter.

  The video covers generic AI models like ChatGPT and is intended more for intermediate and advanced users.

---

Automatically generated from the latest YouTube videos of the curated channel selection. For feedback, suggestions, or to unsubscribe: simply reply to this email.