How good devs write bad code 💻, writing good CLAUDE.md 🤖, accidental performance 😅

Surprisingly, sloppy code isn't due to incompetent engineers, but rather the common practice of assigning engineers to unfamiliar codebases. ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌  ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ 

Sign Up |Advertise|View Online TLDR Together With  TLDR Dev 2025-12-01 Faster Releases With AI-Driven, Human-Verified Automated Testing (Sponsor) 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. QA Wolf solves this with a hybrid model that blends AI test generation with human expertise. Its multi-agent AI system—trained on 50M+ test runs with 700+ evaluation criteria—builds smarter, more resilient tests up to 5× faster than traditional approaches. And with expert QA engineers reviewing every test, you get end-to-end coverage that's accurate, stable, and always maintained. Companies like Cohere, Drata, and AutoTrader rely on QA Wolf to keep their releases on track. Schedule a demo to learn more → 🧑‍💻 Articles & Tutorials How we accidentally made route matching more performant by aiming for correctness (10 minute read) 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). Writing a good CLAUDE.md (10 minute read) 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. The Easiest Way to Build a Type Checker (8 minute read) 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. 🧠 Opinions & Advice How good engineers write bad code at big companies (10 minute read) 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. It's Always the Process, Stupid! (3 minute read) 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. Why Your Best Engineers Are Interviewing Elsewhere (23 minute read) 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. 🚀 Launches & Tools How to evaluate AI agents, a guide by Weights & Biases (Sponsor) AI agent development breaks the rules of traditional software development. Learn from teams at Canva, Cisco, Snowflake and more, on how to establish an evaluation framework so you can assess your agent across accuracy, latency, cost, safety, and user experience and more. Get a practical five-step recipe for effective evaluations in this guide from Weights & Biases. Materialize UI (Website) 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. Exposed by Default (Website) 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. 🎁 Miscellaneous GitHub → Codeberg: my experience (11 minute read) 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. You don't need AI for everything: A reality check for developers (12 minute read) 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. ⚡ Quick Links Zigbook is Plagiarizing the Zigtools Playground (5 minute read) Zigtools accuses Zigbook of plagiarizing its Zigtools Playground and violating the MIT license by not providing proper attribution. Advent of Code 2025 (7 minute read) 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. The space of minds (3 minute read) Both animal and LLM intelligences are forms of optimization. Imgur Geo-Blocked the UK, So I Geo-Unblocked My Entire Network (6 minute read) 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. Love TLDR? Tell your friends and get rewards! Share your referral link below with friends to get free TLDR swag! https://refer.tldr.tech/ef788db7/3 Track your referrals here. Want to advertise in TLDR? 📰 If your company is interested in reaching an audience of web developers and engineering decision makers, you may want to advertise with us. Want to work at TLDR? 💼 Apply here or send a friend's resume to jobs@tldr.tech and get $1k if we hire them! If you have any comments or feedback, just respond to this email! Thanks for reading, Priyam Mohanty, Jenny Xu & Ceora Ford Manage your subscriptions to our other newsletters on tech, startups, and programming. Or if TLDR Dev isn't for you, please unsubscribe.