Full-text search across article titles and summaries. Refine with sector, year, content type, and reading-time filters.
Bun, a high-performance JavaScript runtime, bundler, and package manager, has been acquired by Anthropic. This acquisition provides Bun with long-term stability and resources to focus on building the best JavaScript tooling, especially for AI-driven software like Claude Code. Bun will remain open-source, MIT-licensed, and developed in public on GitHub by the same team, with continued focus on Node.js compatibility and performance.
SQLite's embedded nature and single-writer limitation are usually seen as drawbacks, but are actually important to its scalability. By removing network latency and enabling batching, SQLite can achieve much higher transactions per second (TPS) compared to networked databases like Postgres, especially when dealing with power-law distributed workloads and interactive transactions. This is shown by benchmarks of SQLite reaching over 100,000 TPS even with concurrent reads, while Postgres is much worse with added network latency.
The tech industry is facing a hiring crisis in hiring junior-level engineers due to AI automation replacing entry-level roles and senior engineers not prioritizing mentorship. Universities are aware of students struggling to find jobs, but they can't do much about it. As a result, juniors should focus on developing their networks and their interpersonal skills.
Amazon RDS has a managed environment that imposes a unique set of constraints which require non-trivial engineering solutions. Working with the service is a constant exercise in navigating abstractions. It requires building tools that are aware of the managed environment's specific constraints and behaviors. The key is not to create RDS like a black box.
Sponsored
Push a branch, get a custom preview env URL, same services, routes, and fresh real data. See it happen in seconds, then try it on your repo. No kubectl, no staging drift, just honest previews.
zmx enables session persistence for terminal processes. It can attach and detach from a shell session without killing it. Re-attaching a session restores the previous terminal state and output. Multiple clients can connect to the same session.
The new errors.AsType function in Go is a generic version of errors that is type-safe, faster, and easier to use. errors.As has not been deprecated yet, but errors.AsType is recommended for new code. AsType can handle everything that As does, so it's a drop-in improvement for new code. Code examples are available in the article.
WeChat is ubiquitous in China. The app allows users to access a wide variety of services, including food delivery, ride hailing, and bike rentals, without leaving the app. WeChat Mini apps prove that it's sometimes a better customer experience to just scan a QR code rather than using a native app. This article looks at how to create a WeChat Mini app experience outside of WeChat using regular QR codes and open web standards.
Compilers use the x86's `lea` instruction for efficient three-operand integer addition, using its memory addressing capabilities to overcome the limitations of standard `add` instructions.
Vercel's platform handled a record 115.8 billion requests during the 2025 Black Friday-Cyber Monday weekend.
At AWS Re:Invent 2025, Amazon announced new AI agents (Kiro) for software development, DevOps, and security, claiming they address the issues of existing AI coding tools and improve efficiency.
Zig's new approach to asynchronous programming uses a generic `Io` interface to hide asynchronous operations, allowing code to function correctly with either synchronous or asynchronous I/O and providing fine-grained control over execution.
Sponsored
GitHub analyzed 2,500+ `agents.md` files and found the successful ones were specialists with clear jobs, not vague helpers. Good agents.md files give your agent specific commands to run, concrete code examples to follow, and explicit boundaries (like “never touch these files”). Start simple with one focused task like writing tests or docs, then iterate based on what breaks.
CSS doesn't let you set one property based on another's value (like `border-radius: height`), and there's no magic `compute()` function coming. Your best bets right now are to use CSS custom properties if you know the values, `aspect-ratio` to link width/height, `currentColor` to reuse colors, or container query units like `100cqh` to steal parent dimensions.
AWS Lambda Managed Instances is a new capability that can run AWS Lambda functions on Amazon Elastic Compute Cloud compute while maintaining serverless operational simplicity. It allows developers to access specialized compute options and optimize costs for steady-state workloads using a familiar serverless development experience. Lambda Managed Instances can be used to define how Lambda functions run on EC2 instances. Developers can just select compute profiles optimized for specific workload requirements without taking on the operational burden of managing Amazon EC2 infrastructure.
While React makes sense as a server-side authoring tool (developers love JSX and component architecture), shipping it to the browser hurts users with unnecessary JavaScript. Frameworks like Astro let you write in React but render plain HTML to users, while Next.js defaults to the “hydration” pattern that sends all that server-side JavaScript to the client anyway.
A Staff Engineer shares his Saturday morning ritual for staying current: he batches all newsletters into a 2-3 hour block using Superhuman's split inbox (to separate articles from urgent emails) and Mailbrew (to group 60+ weekly emails into just 2 digests). His process is simple: queue interesting articles by title only, read them one by one, then immediately take action by either saving to Notion for reference or adding to Todoist as a work idea.
Sponsored
Your organizations are just as important as your users. Now track
gives you retention, churn, and cohort analysis for every organization in your app. Daily, weekly, or monthly active orgs. Flexible filtering. Click any chart segment to drill into specific cohorts. Stop flying blind on B2B metrics.
AWS Interconnect - multicloud makes it easy to connect AWS networking services to other cloud service providers. It will start in preview with Google Cloud and then open up to Microsoft Azure later in 2026. The service is now available in preview in five AWS regions. Cloud service providers can easily adopt the service via a published open API package on GitHub.
Rad gives users Python-like scripting with CLI superpowers built in. It allows users to write maintainable scripts with declarative argument parsing, built-in JSON processing, and more. The project is still in early development, but it is already useful for real scripts. Rad is optimized for CLI scripting - a general-purpose language may be more appropriate for enterprise applications, high-performance computations, and projects that require specialized libraries.
Anthropic researchers evaluated AI agents' ability to exploit smart contracts on a newly created benchmark, SCONE-bench. The agents, including Claude Opus 4.5 and GPT-5, successfully exploited vulnerabilities in smart contracts, collectively amassing $4.6 million in simulated stolen funds from contracts exploited after March. In a separate simulation, the AI agents uncovered two novel zero-day vulnerabilities in recently deployed contracts, showing the technical feasibility of profitable, real-world autonomous exploitation.
Despite devices and networks improving, the performance gap between rich and poor users is still growing. For 2026, the budget is ~150KiB of HTML/CSS/fonts and ~300-350KiB of JavaScript (gzipped) to serve the global 75th percentile, but less than half of mobile sites pass Core Web Vitals because developers keep shipping way more JavaScript than reasonable.
Apple announced that John Giannandrea will retire as senior vice president for Machine Learning and AI Strategy, and Amar Subramanya (previously part of Microsoft and ex-VP of the Gemini App at Google) will join as the new vice president of AI.
After initially rejecting JPEG XL despite community support, Chromium has reversed its decision and will now support the format.
Google's advancements in AI with Gemini 3 and its TPU offerings are challenging the dominance of OpenAI and Nvidia, forcing them to re-evaluate their moats and strategies.
This dev has lost confidence in macOS due to silent failures, uninformative error messages, and the resulting reliance on workarounds.
Sponsored
Low test coverage leaves critical flows untested and slows your release velocity. Even with AI-assisted tools, teams can still struggle to reach high coverage and maintain tests as products evolve.
TanStack Router achieved a 20,000× performance improvement in route matching by switching from a flat list iteration approach to a segment trie data structure. This changed the algorithmic complexity from O(N), based on the number of routes, to O(M), based on pathname segments. It also added backwards stack processing, bitmasking for optional segments, and reusing typed arrays for parsing to further boost performance across both small apps (60× faster) and large apps (10,000× faster).
CLAUDE.md, a file for onboarding Claude to a codebase, provides important context about the project's purpose, technology stack, and operational instructions. The CLAUDE.md should be kept concise and universally applicable. Instead of stuffing everything into CLAUDE.md, progressive disclosure should be used by pointing to separate, task-specific markdown files when needed and relying on deterministic tools like linters for code styling.
Bidirectional Type Checking is a conceptually simple method for building type checkers. This article provides a functional type checker implementation for a small language, showing how it uses "infer" and "check" functions to determine and validate types.
Surprisingly, sloppy code at big tech companies isn't due to incompetent engineers, but rather the common practice of assigning engineers to unfamiliar codebases. Engineers often work outside their expertise, rotate frequently, and are incentivized to change jobs, resulting in a lack of long-term ownership and expertise in specific systems.
Implementing AI on top of broken business processes will only accelerate the generation of inefficiencies. Instead of viewing AI as a magical solution, enterprises should focus on optimizing their own processes first, especially when dealing with unstructured data.
Many companies struggle with retaining engineers because of poor information flow within their hierarchical structure, leading to executives being unaware of critical issues until it's too late. Companies need to prioritize open communication channels like skip-level meetings and act on the feedback they receive to show that engineers' input is valued.
Sponsored
Materialize is a platform that allows users to quickly build beautifully designed mobile apps, prototypes, flows, and screens using AI. You can describe your app in English, and Materialize will generate a complete React Native app that can be exported. It can also integrate with design systems.
This website shows the amount of data your browser automatically exposes, creating a unique "fingerprint." This fingerprint includes information about your audio, browser, device, display, fonts, hardware, location, network, and more. The data can be used to identify and track you across the web.
This post details how a developer moved their website from GitHub Pages to Codeberg. The migration took more than a click of a single button, but it's actually not that complicated. The post covers how to migrate repositories, repoint links, stub out GitHub repositories, port CI, and rehost. The whole process of migrating 45 repositories took the developer a weekend to complete.
Devs should resist the pressure to integrate AI into every feature, and instead use deterministic code for simple, well-defined tasks. Using LLMs for straightforward problems like email validation introduces unnecessary costs, latency, and reliability issues compared to traditional code solutions. AI should be reserved for genuinely complex problems involving unstructured data, natural language understanding, or creative generation.
Zigtools accuses Zigbook of plagiarizing its Zigtools Playground and violating the MIT license by not providing proper attribution.
Advent of Code is an annual Advent calendar of programming puzzles for all skill levels, solvable in any language, used for various purposes like interview prep or friendly competition.
Both animal and LLM intelligences are forms of optimization.
This author implemented a network-level solution, routing all i.imgur.com traffic through a VPN-connected container, to access the site to bypass Imgur's UK geo-blocking without impacting network speed or requiring individual device configuration.
Input handling on Linux can be divided into two parts, separated by a common layer. Kernel-level handling deals with what happens in the kernel and how events are exposed to user-space. User-space handling includes the user-space device manager and hardware database for device management and setup, the libinput library for general input and other libraries such as XKB for keyboards, and Widgets, X Server, X11 window managers, and Wayland compositors. The middle exposed layer includes the event abstraction subsystem, devtmpfs for device nodes, sysfs for kernel objects and device attributes, and procfs for an introspection interface of the input core.
Virtual threads simplify concurrency by reducing the cost of waiting threads. They allow developers to use blocking functions without needing to refactor existing code or rely heavily on reactive libraries like Reactor. By using virtual threads, developers can combine modern reactive workflows with traditional Java code, improving code maintainability and understanding.
This post discusses two ways to significantly reduce the size of a static library. One uses classic tools like ld and objcopy, while the other uses LLVM-specific tools. They both produce similar-sized outputs, but there are some trade-offs. The 'classic' bintools approach works with both GNU bintools and LLVM bintools, and it's significantly faster than the LLVM tools, which need to recompile everything from the intermediate representation to machine code. However, the LLVM approach works on macOS, Linux, Windows, and probably others, making it more versatile.