From Night-Time Work with AI Agents to Critical Linux Vulnerabilities: This Week’s Tech Highlights
Wednesday, May 13, 2026
🎧 Podcast Episode (3.7 min)
Hello, this weekly newsletter guides you through the most important new videos from a curated selection of AI and coding YouTube channels. Each video gets a compact summary, plus a daily overview of dominant topics. If interested, simply click the link under the summary.
This week was marked by two dominant themes: the use of AI agents for automated tasks and security vulnerabilities in operating systems. Brian Casel demonstrated in several videos how AI agents can handle routine tasks at night, such as SEO optimization or evaluating GitHub pull requests. Particularly fascinating was his focus on Claude Code, which leverages local capabilities and files not available in the cloud version. The convergence of various tools like Claude Code, Cursor, and Codex toward plan-based development (spec-driven development) was also a topic of discussion.
Another major focus was operating system security. Fireship provided detailed explanations of a critical Linux vulnerability (CVE-2023-31431) discovered by an AI tool. This vulnerability allows local users root access and affects all Linux distributions since 2017. In parallel, the growing strain on GitHub from AI agents, leading to outages and technical problems, was discussed.
An outlier this week was NeuralNine’s video on the ONNX format, which exports machine learning models into a universal format. This enables the use of models independent of the original frameworks like PyTorch or TensorFlow. The introduction of FastEmbed, a Python library for local embeddings, also stood out. It uses ONNX Runtime to avoid dependencies like PyTorch or TensorFlow, and is particularly suitable for less powerful systems.
Overall, there was a clear dominance of Claude and OpenAI in the topics covered, while open-source tools like ONNX and FastEmbed provided interoperability and efficiency. The discussions between creators remained fairly objective this week, with a focus on technical solutions and best practices.
Brian Casel (3 new videos)
- How I build agents that work the night shift
12.5.2026, 12:01:07The video introduces the concept of “Night Shift,” a method for deploying AI agents to handle recurring business tasks. The author describes how he uses agents to automatically complete tasks during his absence (e.g., at night), such as checking and optimizing SEO metadata on his website or reviewing pull requests in his open-source projects on GitHub. The process consists of three main components: a shared interface (e.g., a markdown file or custom app), human feedback in short, focused sessions, and an AI agent executing a skill on a set schedule. The author emphasizes that the real effort lies in setting up the system, but the long-term benefits come from automating and delegating routine tasks.
The author shows two concrete examples: an agent that checks and updates SEO metadata as needed, and another that reviews pull requests on GitHub and provides recommendations for merging or closing requests. In both cases, the author receives reports that he briefly reviews and provides feedback on if needed, before the agent executes the next steps.
The video explicitly covers Claude and OpenAI, and is geared toward intermediate or advanced users who already have experience with AI agents and are ready to develop their own systems and processes.
- Why You Need Claude Code Server Mode?
6.5.2026, 14:01:40The video demonstrates how Claude Code can use locally installed skills and files that are not available in the cloud version. The user shows how he uses various custom-built skills on his local machine that are stored directly on his file system. This enables extended functionality, as Claude Code can access these resources locally, which is not possible in the web version. The advantage is that tasks requiring access to local files or installed skills can be completed more efficiently.
**Final Note:** The video explicitly covers Claude Code and is geared toward intermediate users who are already familiar with local development environments.
- Why Every AI Coding Tool is Converging on Plan Mode?
1.5.2026, 14:01:11The video shows how various developer tools are adopting the trend toward plan-based development (spec-driven development) and integrating it as a standard feature. The author demonstrates this using a product requirements document (PRD) for a customer portal, which he uses in three tools—Claude Code, Cursor, and Codex—to create an implementation plan. In Claude Code, plan mode is automatically activated, guiding the user through a series of questions to create a detailed plan. In Cursor, plan mode is accessed via the Shift+Tab keyboard shortcut, underscoring the standardization of this feature across different tools. The author emphasizes that this convergence of tools makes adoption and switching between them easier.
The video explicitly covers Claude Code, Cursor, and Codex, and is aimed at intermediate to advanced users who are familiar with spec-driven development.
Niklas Steenfatt
No new videos in this period.
Fireship (3 new videos)
- Every operating system concept in one video…
7.5.2026, 17:32:34The video explains in detail how an operating system works from the moment the power button is pressed until shutdown. It begins with the bootloader that loads the operating system, then goes through various stages such as privileges, virtual memory, file systems, device drivers, and interrupts. The kernel builds the system, manages processes and memory, enables inter-process communication, and ensures system security and stability. The video also highlights the importance of system calls, schedulers, and threads. At the end, the shutdown process is described, where all processes are terminated and the system is safely shut down.
The video doesn’t address specific AI tools or models and is geared more toward intermediate viewers seeking a deeper understanding of operating systems.
- 732 bytes of Python just borked every Linux machine on earth…
4.5.2026, 18:40:40The video covers a critical security vulnerability in the Linux kernel, known as “copy fail” (CVE-2023-31431), which has existed since 2017 and was discovered by an AI tool. The vulnerability allows a local, unprivileged user to gain root access to a Linux system by writing four controlled bytes into the page cache of a read-only file. The exploit leverages a weakness in the AF_ALG interface and affects all Linux distributions updated since 2017. While the vulnerability is not remotely exploitable, it requires an update on all affected systems. The video explains the technical details of the exploit and emphasizes the importance of AI tools for improving code quality, with Code Rabbit cited as an example.
The video explicitly addresses AI tools such as the AI agent tool used by Theori and Code Rabbit, and is geared more toward intermediate and advanced users.
- GitHub is having some major issues right now…
30.4.2026, 16:37:31The video “It’s 10 p.m. Do You Know Where Your Children Are?” from The Code Report discusses the current issues and challenges facing GitHub, the leading platform for software development. GitHub, founded in 2008 and acquired by Microsoft in 2018 for $7.5 billion, is essential for many developers. However, outages and technical problems have been mounting lately, causing massive dissatisfaction in the developer community. The video creator mentions several incidents, such as accidental deletion of pull requests, search function outages, and critical security vulnerabilities. A particularly striking example is the decision by Mitchell Hashimoto, a prominent developer, to migrate his open-source projects away from GitHub, viewed as a sign of growing frustration among many developers.
The video creator analyzes the causes of these problems and points to the increasing strain from AI agents placing heavy demands on the platform. Despite the criticism, he emphasizes that Microsoft and GitHub are working to improve the situation. Alternatives to GitHub mentioned include platforms like GitLab, Codeberg, and Source Hut. The video closes with an appeal for choosing a meaningful and fulfilling career, supported by the organization 80,000 Hours.
The video addresses GitHub and Microsoft and is aimed at an intermediate to advanced audience.
NeuralNine (7 new videos)
- JavaScript Crash Course: Full Beginner Tutorial
13.5.2026, 16:01:18The video provides a comprehensive introduction to the fundamentals of JavaScript programming. It starts with simple concepts like outputting text to the console and using variables. The author explains the differences between the keywords `let`, `var`, and `const` and their scopes. Various data types are covered, including numbers, strings, booleans, and special values like `undefined`, `null`, `Infinity`, and `NaN`. The author also demonstrates how to work with operators, including arithmetic, assignment, comparison, and logical operators.
Control structures such as `if-else` statements and `switch-case` constructs are explained. The author demonstrates how to process user input using the `readline` library. Various loop constructs like `while`, `do-while`, and `for` loops are introduced, including the use of `break` and `continue` to control loop execution.
Functions and their usage are explained in detail, including function definition, parameter usage, return values, anonymous functions, lambda expressions, and the use of functions as first-class objects (callbacks). The author also demonstrates how to work with arrays, maps, and sets, as well as the definition and usage of objects and classes in JavaScript.
Finally, it briefly shows how to embed JavaScript in HTML to create interactive web pages.
The video is suitable for beginners who have basic programming knowledge and are looking for a quick but comprehensive introduction to JavaScript. No specific AI tools or models are discussed.
- ONNX: The PDF Format For Neural Networks
11.5.2026, 16:00:44The video demonstrates how to use the ONNX format (Open Neural Network Exchange) to export machine learning models into a universal format that works independently of the original frameworks like PyTorch or TensorFlow. The author demonstrates this with simple examples, where a model that increments a number by one is created in both frameworks and then exported to ONNX format. It then shows how to use the exported model with ONNX Runtime without needing PyTorch or TensorFlow. A more complex example with an MNIST classifier trained in PyTorch and then exported to ONNX format is also presented. Finally, it demonstrates how to download and use a model from Hugging Face in ONNX format. The author emphasizes the advantages of interoperability and reduced dependencies through using the ONNX format.
The video explicitly covers ONNX, PyTorch, TensorFlow, Hugging Face, and ONNX Runtime and is aimed at intermediate and advanced users.
- Why you should start creating tech content…
10.5.2026, 11:04:45The video argues for the benefits of creating tech content and sharing knowledge. First, teaching forces deep understanding, as you must explain concepts clearly and simply, similar to the Feynman Technique. Additionally, you can directly link new knowledge with teaching and expand your portfolio. Second, consistent content creation demonstrates proactivity and discipline, which stands out positively to employers and partners. Third, reach and recognition open doors, as large followings are often associated with competence, even if this shouldn’t be the primary motivation. The author emphasizes that tech knowledge should be shared, even without a camera, such as through blogs or LinkedIn.
The video doesn’t discuss specific AI tools but is aimed at intermediate and advanced tech experts considering sharing their knowledge.
- OpenClaw: Simple VPS Setup Guide
8.5.2026, 16:00:03The video shows step-by-step how to install Open Claw on a virtual private server (VPS) with Ubuntu and connect it to Telegram. First, a VPS is set up with Hetzner and Ubuntu is installed. Node Version Manager (NVM) is then installed to use the required Node.js version (24). Open Claw is installed globally and the onboarding process is started. OpenAI is selected as the model provider and Telegram as the communication channel. After entering the OpenAI API key and Telegram bot token, the bot is configured and started. The bot can then be used via Telegram to perform tasks like creating and editing files. The video emphasizes the security risks of Open Claw and recommends running it in an isolated environment.
The video explicitly covers OpenAI, Telegram, and Open Claw and is aimed at intermediate users.
- Coding Slack Bots in Python: Quick Start Guide
4.5.2026, 16:00:21The video shows how to create a simple Slack bot with Python. It focuses on creating a Minimal Viable Product (MVP) that can execute commands and communicate with a language model. The process starts with setting up a Slack app and creating a bot in a Slack workspace. A Python project is then set up with the necessary packages (python-dotenv and Slack-Bolt). The bot is programmed to respond to certain messages, execute commands, and respond to mentions. As an example, it shows how to connect the bot to the OpenAI API to get responses from GPT-4O. The video demonstrates the bot’s basic functionality and how to use it in a Slack channel.
The video explicitly covers the use of OpenAI (GPT-4O) and is aimed at intermediate users.
- How to stay up-to-date in AI/ML?
3.5.2026, 22:40:53The video offers tips on how to stay updated on the latest developments in AI and Machine Learning in just 15 minutes a day. The author recommends three main strategies: First, read Hacker News, as it regularly publishes current articles and discussions about new models, frameworks, and papers. Second, subscribe to the right YouTube channels, such as Fire Ship for quick updates and AI Explained for in-depth analysis. Third, connect with like-minded people and join communities to automatically receive relevant news. The author emphasizes that these methods provide natural filtering and deliver only the most relevant information.
The video covers Hacker News, YouTube channels like Fire Ship and AI Explained, and communities, and is suitable for intermediate and advanced users.
- FastEmbed: Local AI Embeddings in Python
1.5.2026, 16:00:07The video introduces the Python library FastEmbed, which is specifically optimized for creating embeddings locally, even on less powerful systems without a dedicated GPU. FastEmbed uses ONNX Runtime to avoid dependencies like PyTorch or TensorFlow, and provides a simple and fast way to work with text and image embedding models.
The tutorial shows the installation of FastEmbed and demonstrates the creation of text embeddings using an example with three sentences that compare different meanings of “Apple” and “Microsoft”. It shows how to calculate the similarity between sentences using vector distances. The use of image embedding models with two example images of dogs is then explained.
Additionally, it explains how FastEmbed is implicitly used by Qdrant, a vector store optimized for filtering and custom tagging. Finally, it briefly discusses the possibility of GPU usage with FastEmbed, emphasizing that this does not happen automatically and requires additional configuration.
The video explicitly covers FastEmbed and Qdrant and is aimed at intermediate users.
Automatically generated from the latest YouTube videos of the curated channel selection. For feedback, suggestions, or to unsubscribe: simply reply to this email.