Published on April 29, 2026 at 18:00 CEST (UTC+2)
Zed is 1.0 (395 points by salkahfi)
Zed is 1.0 – Zed, the Rust-based code editor built with a custom GPU-accelerated UI framework (GPUI), has reached version 1.0. The founders previously created Atom and Electron but argue that web-based foundations impose a performance ceiling. By writing the entire stack from scratch to feed data directly to GPU shaders, Zed achieves high performance and low latency. The editor now supports hundreds of thousands of daily users and promises a level of speed and responsiveness unavailable to editors built on Electron or similar platforms.
Tangled – We need a federation of forges (228 points by icy)
Tangled – We need a federation of forges – Tangled proposes a decentralized code collaboration platform using git for code transfer and the AT Protocol (from Bluesky) for communication of issues, pull requests, and social features. The project aims to free open-source software from dependence on a single provider like GitHub, which is increasingly unreliable. By federating events across self-hosted “knots,” Tangled allows cross-server forking and pull requests, similar to the old patch-by-email workflow but with modern synchronization.
Why AI companies want you to be afraid of them (57 points by rolph)
Why AI companies want you to be afraid of them – This BBC article critiques the marketing strategy of AI companies like Anthropic that publicly claim their models are too dangerous to release, while simultaneously selling or hinting at future releases. The author argues that such fear-mongering creates hype, positions the company as responsible, and attracts attention and funding. It warns that this narrative may be used to shape regulation in favor of incumbents rather than addressing real risks.
Soft launch of open-source code platform for government (390 points by e12e)
Soft launch of open-source code platform for government – The Dutch government has launched code.overheid.nl, a self-hosted Git platform based on Forgejo (a European, open-source alternative to GitHub/GitLab). The pilot aims to support digital sovereignty and enable government bodies to publish and develop open-source software collaboratively. Developers are invited to contribute, with the goal of scaling into a shared platform for all Dutch government organisations.
Ghostty is leaving GitHub (3100 points by WadeGrimridge)
Ghostty is leaving GitHub – Mitchell Hashimoto, creator of Vagrant and the Ghostty terminal emulator, announces that Ghostty will migrate away from GitHub, despite his personal 18-year attachment to the platform. He explains that GitHub’s recent direction (e.g., AI-driven changes, centralization, and potential lock-in) no longer aligns with the project’s values. The move underscores a growing trend among maintainers to reduce reliance on a single commercial code host.
Improving ICU handovers by learning from Scuderia Ferrari F1 team (29 points by embedding-shape)
Improving ICU handovers by learning from Scuderia Ferrari F1 team – This article (from 2012) examines how complex systems like intensive care units can reduce errors by adopting communication protocols used by Formula 1 pit crews. It highlights that seemingly minor handover failures accumulate into significant patient harm. The Ferrari example illustrates structured checklists, role clarity, and debriefing techniques that can improve safety in high-stakes medical environments.
GitHub – DOS 1.0: Transcription of Tim Paterson's DOS Printouts (56 points by s2l)
GitHub – DOS 1.0: Transcription of Tim Paterson's DOS Printouts – A GitHub repository hosts a transcription of Tim Paterson’s original printed source code for 86-DOS 1.0 and early PC-DOS versions, including the Microsoft BASIC-86 compiler runtime. The listings have been digitised, transcribed, and made compilable. This archival effort preserves a key piece of computing history and enables future study or reconstruction of early MS-DOS internals.
Mistral Medium 3.5 (66 points by meetpateltech)
Mistral Medium 3.5 – Mistral AI releases Medium 3.5, a 128B-parameter dense model under a modified MIT license. It powers remote coding agents in their “Vibe” environment, allowing developers to spawn async coding tasks from CLI or Le Chat. The model is designed for long-running instruction-following, reasoning, and code generation, and can be self-hosted on as few as four GPUs. A new “Work mode” in Le Chat also provides a powerful agent for multi-step tasks.
Bugs Rust won't catch (493 points by lwhsiao)
Bugs Rust won't catch – An analysis of 44 CVEs disclosed in uutils (Rust reimplementation of GNU coreutils) reveals bugs that Rust’s borrow checker and lints didn’t prevent. The most common pattern is TOCTOU (time-of-check, time-of-use) vulnerabilities where two syscalls on the same path can be interleaved by an attacker. The article argues that Rust eliminates memory safety issues but does not automatically solve logic, concurrency, or filesystem race conditions, citing the need for additional audits and careful API design.
Letting AI play my game – building an agentic test harness to help play-testing (63 points by jschomay)
Letting AI play my game – building an agentic test harness to help play-testing – The author describes an experiment where an AI agent (likely using a vision or keyboard/mouse interface) interacts with a game to automatically test gameplay, find bugs, and explore edge cases. The harness can simulate human-like play patterns, run for extended periods, and report issues. This approach reduces manual play-testing effort and can uncover issues that scripted tests miss.
AI Safety Rhetoric as a Marketing Tactic – The BBC article highlights a growing trend where AI companies publicly warn about the dangers of their own models while simultaneously selling them. This “scary but responsible” narrative helps shape regulation in their favor and attracts media attention. For AI/ML practitioners, this underscores the importance of separating genuine safety research from marketing spin, and the need for independent auditing and transparency.
Open-Weight Models Enable Self-Hosted and Sovereign AI – Mistral’s release of Medium 3.5 under a modified MIT license, capable of running on just four GPUs, continues the push toward smaller, efficient, open-weight models that can be deployed on private infrastructure. This trend matters because it reduces dependence on giant cloud providers, enables government and enterprise digital sovereignty, and lowers the barrier for fine-tuning and customisation (e.g., Dutch government’s open-source code platform and Ghostty leaving GitHub echo a similar desire for independence).
AI Agents Move from Laptop to Cloud for Asynchronous Coding – Mistral’s “remote agents” represent a shift from local coding assistants (e.g., Copilot on your IDE) to cloud-based agents that run long tasks in parallel and notify you when done. This trend could fundamentally change how developers offload work: instead of asking for a snippet, you assign a full feature, and the agent returns a draft. Implications include new orchestration challenges (managing parallel agents) and a redefinition of the developer’s role toward high-level review and architecture.
AI Play-Testing and Agentic Testing Gain Traction – The article about using an AI agent to play-test a game signals a broader move toward autonomous testing of interactive systems. This is relevant beyond games: agentic test harnesses can explore UIs, find regressions, and simulate user behavior at scale. For AI/ML, it requires robust vision or environment APIs, and highlights a practical application of reinforcement learning and multimodal models in QA.
Rust’s Safety Is Not a Panacea – AI Can Help but Also Needs Careful Design – The uutils audit shows that even in a memory-safe language like Rust, logical and time-of-check/time-of-use bugs still thrive. For AI/ML, this is a cautionary tale about over-relying on any single safety guarantee. AI-based code generation tools (like Mistral Medium 3.5) must be trained to avoid these patterns, and developers need to use static analysis, formal verification, and threat modeling in addition to borrowing liveness checks.
Decentralisation and Digital Sovereignty Are Major Motivators for AI/ML Infrastructure – Multiple articles (Tangled, Dutch government platform, Ghostty leaving GitHub) reflect a growing desire to move away from centralised code forges and cloud AI services. For AI/ML, this means the ecosystem will need to support federated model hosting, privacy-preserving fine-tuning, and distributed data annotation. Tools that enable self-hosted, sovereign ML pipelines (like Mistral’s four-GPU model) will become increasingly important.
Historical Software Preservation Benefits from AI-Assisted Transcription and Compilation – The DOS 1.0 printout transcription project, while non-AI itself, points to a broader use case: AI is now capable of digitising and reconstructing historical source code from scanned documents. For ML, this creates valuable datasets of legacy software patterns, and for the community, it ensures that foundational systems remain accessible for study and reclamation – a trend that will likely accelerate with multimodal models.
Analysis generated by deepseek-reasoner