Mitchell Hashimoto says Ghostty will leave GitHub after 18 years on the platform. He cites frequent GitHub outages affecting his work and says Ghostty will migrate to another hosting provider, with a read-only mirror kept on GitHub and his personal projects remaining there for now. Hashimoto notes a plan to remove dependencies on GitHub infrastructure gradually and promises to share details on the new hosting arrangement in coming months.
Top Stories
An Alberta-based startup is described in the title as selling no-tech tractors at half price. The text provided is a bot-verification page and contains no further details about the company or its products.
Apple announced that Tim Cook will become executive chairman of the board and John Ternus will become chief executive officer, effective September 1, 2026, with Cook remaining CEO through the summer to oversee a smooth transition. The transition was approved unanimously by the board, and Cook will assist with certain duties, including engaging with policymakers. Arthur Levinson will become lead independent director, and Ternus will join the board on September 1, 2026.
Zed is the 1.0 release of a new code editor announced by Nathan Sobo. It is built as a standalone desktop app with a custom UI framework called GPUI in Rust to avoid web-based limitations, and runs on macOS, Windows, and Linux. The editor is described as AI-native, supports multiple languages and tools, and uses DeltaDB CRDT-based synchronization for collaboration, with ongoing weekly releases and Zed for Business offering centralized billing and access controls.
DeepSeek v4 provides API access using formats compatible with OpenAI and Anthropic, with base URLs and an API key required. Models include deepseek-v4-flash and deepseek-v4-pro; the chat and reasoner variants (deepseek-chat and deepseek-reasoner) are slated for deprecation on 2026-07-24 and correspond to non-thinking and thinking modes, respectively. The documentation includes a curl example for the chat completions endpoint at api.deepseek.com/chat/completions, showing a payload with model, messages, thinking, reasoning_effort, and stream fields.
The article argues that software should rely on on-device AI rather than cloud-hosted models to improve privacy, reliability, and cost. It cites a native iOS client for The Brutalist Report where summaries are produced on-device using Apple’s local model APIs, avoiding server involvement and data storage. It notes tooling in Apple’s ecosystem that supports structured outputs and suggests using cloud models only when genuinely necessary.
A report alleges that Google Chrome silently installs a 4 GB on-device AI model onto users’ devices, writing a weights.bin file into a directory named OptGuideOnDeviceModel without user consent. The model is described as Gemini Nano and the piece says Chrome re-downloads the file if it is deleted, and that the behavior has potential privacy, legal, and environmental implications at scale. The article notes that disabling AI features or using enterprise policy can prevent the installation, but describes the default behavior as persistent.
Valve has released CAD files for the Steam Controller and its Puck under a Creative Commons license, allowing non-commercial use and attribution with designs shared back to the community. The files include the external-shell topology and engineering diagrams in STP and STL formats, with notes on areas that must remain uncovered for signal strength and proper function. Valve has previously released CAD assets for the Steam Deck and Valve Index, and commercial terms can be discussed directly with Valve.
The article claims that starting September 2026, Google plans to require all Android app developers to register centrally with Google, including a fee, terms, government ID, signing key disclosure, and listing all app identifiers. It asserts that unregistered apps could be silently blocked on devices worldwide, potentially limiting users to pre-approved software, including non-Play Store apps and hobbyist projects. The piece also cites advocacy groups and urges actions such as using alternative app stores and signing petitions.
The article argues that AI tools can generate work that appears competent without real expertise, enabling cross-domain production by novices and non-specialists. It links organizational incentives to momentum and visibility over accuracy, contributing to a decoupling between the produced output and the producer’s actual competence. It notes studies showing mixed productivity gains and overconfidence, and recommends using AI only for tasks that can be verified, with the human maintaining final judgment.
OpenAI released GPT-5.5 on April 23, 2026, describing it as a more capable model that can plan, use tools, and carry out multi-part tasks across coding, research, and office work. It notes stronger safeguards, improved efficiency with fewer tokens, and latency similar to GPT-5.4 while delivering higher performance, supported by external testing and early-access feedback. The rollout targets Plus, Pro, Business, and Enterprise users in ChatGPT and Codex, with GPT-5.5 Pro for additional plans and API deployment to follow under safety and security requirements.
The article's title asserts that Google broke reCAPTCHA for de-Googled Android users. It also discusses Brave's stripped-down option and who should care, noting that its usefulness depends on the reader’s privacy preferences.
A person conducted a one-month gym-based experiment to make friends by approaching one other gym-goer per day, starting with a fixed opener and tailoring it to each person. The conversations varied in length and receptiveness, with some positive responses and others brief or uninterested, and the person later shifted focus to nurturing existing gym connections. By the end, they had established ongoing contacts and reported reduced loneliness and greater social comfort.
VS Code's Git extension pull request 'Enabling ai co author by default' changes the default for git.addAICoAuthor from 'off' to 'all', enabling AI co-author trailers by default in commits. The update notes a mismatch between the schema default and the runtime fallback (which remains 'off') and records user feedback criticizing the default behavior.
Security researchers disclosed Copy Fail, a local privilege escalation affecting Linux kernels dating back to 2017, tracked as CVE-2026-31431. A 732-byte Python script can exploit a straight-line flaw in the AF_ALG path to gain root via the page cache, with a public PoC and demonstrations across multiple distributions. Mitigations include updating to kernels that revert the 2017 AF_ALG optimization (commit a664bf3d603d) or disabling algif_aead, with patches already rolling out on major distros.
The Framework Laptop 13 Pro is a modular, repairable laptop with a CNC aluminum chassis and compatibility with prior mainboards. It features Intel Core Ultra Series 3 processors, supports up to 64GB LPDDR5X memory and up to 8TB PCIe Gen5 NVMe storage, and includes a 13.5-inch 2880x1920 touchscreen with a haptic touchpad and Dolby Atmos audio. It can be shipped with Ubuntu (Ubuntu Certified) or Windows 11, with Linux-first support and a DIY Edition for other operating systems, plus a Marketplace for compatible parts.
The article outlines a dispute between Bambu Lab and an open-source fork of its software, centered on cloud-connected features and licensing. It describes Bambu Lab's public statements alleging impersonation by the fork's developer and the fork's counterclaims that AGPL-licensed code was used openly, with concerns about security and governance. The piece notes community responses, including financial support pledges to assist the open-source developer.
Cloudflare will cut more than 1,100 jobs globally as it reorganizes for an 'agentic AI era.' Founders say the move is a strategic restructuring, not a reflection of individuals' performance, with severance equal to full base pay through the end of 2026, continued US healthcare through year-end, and accelerated equity vesting for departing employees. Staff will be notified by email, and the company plans an earnings call at 2 PM PT to discuss the changes.
A Claude Code bug caused requests to route to extra usage billing instead of the included Max plan quota when a git commit message contained the exact string HERMES.md. The issue reportedly consumed about $200.98 in extra usage charges while the Max plan showed remaining capacity, and was later attributed to an overactive anti-abuse system that has been fixed.
Anthropic Labs unveiled Claude Design, a product that lets users collaborate with Claude to create visual work such as prototypes, slides, and marketing collateral. Powered by Claude Opus 4.7 and in research preview for Claude Pro, Max, Team, and Enterprise subscribers, it automatically builds a team design system, supports multiple import methods and inline editing, exports to Canva, PPTX, PDF, or HTML, includes a handoff to Claude Code, and is rolling out gradually.
Cursor Camp is presented with a brief welcome message. The text invites readers to enjoy their stay and includes the word 'Enter'.
The piece argues that defense and software capabilities hinge on deep, decades-spanning tacit knowledge, and that once experienced workers retire, rebuilding capacity can take years—evidenced by Stinger production restart, Europe’s artillery shell delays, and the Fogbank episode. It also notes that money and AI cannot compress this timeline, citing a MET randomized trial showing AI tools slowed experienced developers, and highlighting ongoing hiring bottlenecks and the need for senior engineers who understand end-to-end systems.
The article compiles a broad catalog of software engineering laws, principles, and patterns. It groups items by category—architecture, planning, quality, design, teams, and decisions—and lists well-known rules such as Conway's Law, YAGNI, CAP Theorem, and Brooks's Law.
On 2026-04-22, the author announces fundraising for exe.dev and explains his motivation to build a cloud he would actually want to use. He argues current cloud models are constrained by VM-centric abstractions and Kubernetes, and describes exe.dev’s approach: provisioning CPUs and memory directly, running VMs, with local NVMe storage, TLS/auth proxies, regional presence, and an anycast network. The post also notes ongoing work to add features like static IPs and automatic disk snapshots, as part of rethinking cloud infrastructure.