July 2026 Summaries

Tailscale, a versatile networking tool utilized for various applications from homelab setups to enterprise infrastructures, faces challenges when connecting to multiple tailnets simultaneously due to its independent allocation of IPv4 addresses in each tailnet, which can lead to address conflicts. The author expresses frustration with the need to manually switch tailnets, disrupting existing connections and requiring administrative permissions to share nodes across tailnets. To address this, the author developed an open-source client called tailmix, which allows a single machine to connect to multiple tailnets by using tsnet.Server instances and a single TUN interface to assign unique IPv4 addresses from a local pool, leveraging MagicDNS for routing. This solution enables seamless connectivity across tailnets without the need to implement new policies, maintaining separate identities and policies for each tailnet while eliminating the need to switch manually.

Exe.dev offers two types of Slack integrations to enhance interaction with virtual machines (VMs) via Slack. The first integration, simply labeled "Slack," is a straightforward, send-only version that allows users to send messages from a VM to a Slack channel after setting up the integration through a simple process that involves selecting a workspace and channel. The second, more advanced integration, known as "Slack Bot," provides greater flexibility and functionality by permitting the creation of a full-fledged Slack bot capable of reading and writing messages, as well as executing tasks based on permissions granted during setup. This involves a more detailed setup process where users create an app and input necessary tokens to enable the bot's capabilities, allowing for simple message posting or more complex interactions like responsive chatbots. Detailed instructions and demo code are available in the documentation to facilitate the integration setup.

Carl Sverre, a software engineer at Antithesis, explores the balance of empowering AI agents effectively without compromising safety and efficiency. His experience with various platforms led him to exe, a tool offering a flexible and cost-effective solution for managing virtual machines (VMs) without the constraints of traditional sandboxes or the financial burden of per-VM charges. Exe allows Sverre and his team to access up to 50 VMs per user for a flat monthly fee, simplifying workflows and supporting persistent development environments crucial for projects like Snouty, Antithesis' command line tool. Key features such as root access to VMs and the elimination of delays associated with hibernating VMs enhance productivity and responsiveness. Sverre appreciates exe's flexibility and responsiveness to feature requests, such as restricting public VM sharing, and sees potential for expanding this approach into a more automated software factory, balancing the need for AI empowerment with safety considerations.

Shelley is a customizable tool that allows users to adjust its features by simply asking the agent to make changes, such as changing its appearance to pink. This innovative approach to software customization is distinct from traditional configuration and plug-in architectures by enabling direct code edits. When a new version of Shelley is released, it automatically integrates existing customizations onto the updated version, ensuring continuity of user preferences. Users are encouraged to experiment with customizing Shelley and to share successful, enduring features with the community on Discord, fostering a collaborative environment for enhancing the tool's capabilities.

In the text, the author discusses the evolution of software development and the role of Large Language Models (LLMs) like Claude, arguing that they are more than just compilers because they can work across different layers of software development, from strategy to machine code. Traditional software layers each involve specialized roles, but LLMs can integrate these roles by making decisions across the stack, accelerating and enhancing decision-making processes. The text illustrates this with a project involving the creation of a distributed DNS server, where LLMs assisted in researching, designing, and implementing solutions to complex problems, highlighting the advantages of LLMs in understanding and managing systems. Although LLMs are not yet as adept as humans in specific tasks, they provide a comprehensive perspective that can guide engineers in making informed decisions. The author emphasizes the importance of understanding the systems being worked on and suggests that while some software layers may become obsolete, those that allow for the expression of important decisions will remain valuable. The concept of "vibe-engineering," which involves using LLMs to support engineering tasks, is presented as an emerging facet of modern engineering, blending traditional practices with advanced AI capabilities.

Raffael Hickisch, a conservationist and co-founder of the Chinko Nature Reserve in the Central African Republic, has been leveraging technology to address the challenges of managing vast protected areas in Sub-Saharan Africa. Traditionally, Hickisch struggled with manually assembling data from various sources to monitor human activities, deforestation, and fires, crucial for enforcing conservation regulations. The introduction of exe.dev and its AI agent Shelley has revolutionized his work by enabling him to create the Five Megapixel Conservation app, which aggregates global conservation efforts using GPS data from park rangers and vehicles. Additionally, Hickisch has developed a system utilizing 100 virtual machines to process Austria's massive LIDAR data, identifying old forests in need of protection, and even created a settlers-style game to visualize LIDAR data interactively. While Hickisch's open-source apps aren't profit-driven, he aims for widespread adoption to empower conservation efforts, demonstrating that powerful technological tools can now be accessed by those with limited resources, like conservationists in developing countries.

San Anselmo's live music concert series faced challenges in terms of mobile accessibility, prompting the author to use the exe.dev iOS app to create a mobile-friendly, personalized concert calendar. By leveraging the app's features, such as virtual machines (VMs) and high-quality transcription, the author streamlined the process of managing tasks and ideas while on the go, moving away from cumbersome typing and intermediate steps. This approach allowed them to efficiently handle trivial tasks via voice mode while performing other activities, and to utilize the app's share sheet for seamless integration and transcription of ideas and media. The exe.dev platform thus provided a cohesive solution that enhanced productivity and eliminated the need for third-party services that often required additional manual effort or lacked integration capabilities.

You can now connect your ChatGPT subscription to exe.dev, allowing you to utilize the coding agent Shelley with OpenAI models through a simple integration process. By following the setup steps at exe.dev/integrations, your virtual machines (VMs) will be configured to use your OpenAI subscription. This integration leverages several exe.dev features, such as the tag-based configuration system for adding functionality to VMs, the reflection integration for identifying attached integrations, and the LLM integration for making requests and discovering models. On startup, Shelley offers available models to the VM, and a terminal-based agent, pi, is similarly pre-configured. Additionally, Codex can be configured using the exeuntu command, and both pre-configured tools and user-built tools can access the LLM integration to call models.

Exe.dev facilitates the operation of programs on virtual machines (VMs) by managing network connections through a dedicated process called exepipe, which handles network traffic without implementing additional features. The platform supports various functionalities such as encryption, authentication, SSH access, and terminal access to VMs, necessitating proxying of incoming connections and efficient routing of outgoing connections. Exepipe simplifies the process by maintaining stable connections and utilizing Unix socket ancillary data to duplicate network sockets, ensuring uninterrupted service even during software updates. In cases where updates to exepipe are necessary, a new instance is initiated to take over connections, allowing the old instance to continue managing existing ones until they close, thereby maintaining seamless user experiences.