Don’t avoid office politics 💼, Cursor 1.7 🚀, memory engineering for agents 💾

Engineers often avoid internal politics, but this is a mistake. Politics is simply how humans coordinate and exert influence ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌  ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ ‌ 

Sign Up |Advertise|View Online TLDR Together With  TLDR Dev 2025-10-02 One API. Every LLM. No outages. (Sponsor) OpenRouter gives you a single API & SDK to access 500+ models - GPT, Claude, Gemini, Llama, Mistral, and more. 🔑 Bring your own API keys - or let users bring their own API keys via BYOK setup. Never worry about getting a huge bill for going viral! ♻️ Swap models at the request level, route across providers by price or latency, and have fallback options to survive outages. 💪 Full API functionality: stream responses, use structured JSON output, tools/function calling, and images (I/O).  📊 Track spend with API level limits and transparent activity view. No lock-in.  Get Started For Free in 30 seconds >>  🧑‍💻 Articles & Tutorials Blockdiff: How we built our own file format for VM disk snapshots (12 minute read) Cognition AI built Blockdiff to get fast block-level diffs and snapshots of VM disks for its AI software engineer, Devin. Blockdiff addresses the slow snapshot times experienced with EC2 by creating compact, instantaneous snapshots with zero overhead by only storing the differing blocks between a base image and VM disk. The tool uses Linux filesystem concepts like sparse files and copy-on-write to efficiently serialize the diffs. Why Multi-Agent Systems Need Memory Engineering (8 minute read) Multi-agent AI systems frequently fail because they don't have a proper shared memory infrastructure. While context engineering has improved individual agent performance by managing "the right information at the right time," this approach breaks down when multiple agents must coordinate without shared memory systems. The solution is memory engineering: creating a persistent, shared memory infrastructure that allows AI agents to evolve from single-agent tools to coordinated teams capable of handling enterprise-scale problems. What Every Programmer Absolutely, Positively Needs to Know About Encodings and Character Sets to Work With Text (34 minute read) Computers store text as bits, and encodings define how these bits translate into human-readable characters through lookup tables. Problems happen when text is interpreted with the wrong encoding, leading to incorrect character display. It's best to consistently use Unicode (specifically UTF-8) and proper encoding conversion techniques to prevent data loss and guarantee accurate text representation. 🧠 Opinions & Advice Stop Avoiding Politics (5 minute read) Engineers often avoid internal politics, viewing it as a negative game, but this is a mistake. Politics is simply how humans coordinate and exert influence within organizations. Avoiding politics doesn't eliminate it, it just allows "bad politics" and misinformed decisions to prevail. By building relationships, understanding incentives, and communicating effectively, engineers can engage in "good politics" to advocate for better outcomes. Increasing your practice surface area (8 minute read) The difference between good and great performers is in "practice surface area," not just talent or formal training. This is the total time and space in one's life where practice can occur. Elite performers dissolve the boundaries of formal training by integrating their craft into everyday life. No Figma, I won't fit in your little box (6 minute read) The separation of designers and developers reduces creativity and creates friction in the design process. This divide forces designers into a shrinking role, focusing on visuals while developers handle functionality, leading to annoying hand-offs and slow iteration. A better approach is to use tools built on HTML and CSS with visual interfaces that can unite design and development. 🚀 Launches & Tools ✂️ Cut your QA cycles down to minutes with QA Wolf (Sponsor) If QA is a bottleneck on your software engineering team and you're releasing slowly because of it, you need to check out QA Wolf. Their fully-managed service gets engineering teams to 80% automated end-to-end test coverage, and their AI-native platform helps dev teams ship 5x faster by reducing QA cycles from hours to minutes. With QA Wolf, Drata's team of 80+ engineers achieved 4x more test cases and 86% faster QA cycles. Schedule a demo to learn more React 19.2 (10 minute read) React 19.2 is now available, with new features like Activity components for controlling and prioritizing app sections, useEffectEvent for managing event logic in effects, and cacheSignal for cache lifetime awareness in React Server Components. Performance Tracks are added to Chrome DevTools to give more insights into React app performance, and Partial Pre-rendering is introduced for pre-rendering static parts of apps. Cursor 1.7 (2 minute read) Cursor 1.7 introduces several new features and improvements, including Agent Autocomplete, Hooks for customizing Agent behavior, and team rules for consistent project management. Shareable deeplinks for prompts are now available in beta, allowing for easier sharing of workflows and instructions. Commands now execute in a sandboxed environment for better security, and users can monitor Agents from the menu bar. 🎁 Miscellaneous Bits & Side Tables: How Reference Counting Works in Swift (23 minute read) This is a deep dive into Swift's reference counting mechanism, explaining how strong, weak, and unowned references are implemented at the bit level within the Swift Runtime. It explains how strong references use inline bit counting and may use a side table when counts overflow or weak references exist. Weak references are slower due to indirection and potential cache misses from accessing a side table, while unowned references offer better performance but can lead to crashes if accessed after deallocation, keeping zombie memory alive. We tried Go's experimental Green Tea garbage collector and it didn't help performance (8 minute read) DoltHub tested Go's experimental Green Tea garbage collector, which is supposed to improve performance by scanning objects based on their memory location to improve cache locality. Despite running comprehensive benchmarks with their Dolt database using both single and multi-threaded workloads, they found basically identical performance between the standard garbage collector and Green Tea, with no meaningful differences in throughput, median latency, or latency distribution. The results align with the Go team's own findings that most real-world applications don't see significant performance improvements from Green Tea, and if anything, it appeared to be a slight regression in garbage collection performance for their use case. ⚡ Quick Links Why Over-Engineering Happens (19 minute read) Over-engineering often comes from misguided intentions, misaligned incentives, and a desire to build for hypothetical future needs rather than solving present problems. The Temporal Dead Zone, or why the TypeScript codebase is littered with var statements (4 minute read) The TypeScript codebase still uses `var` statements despite their drawbacks due to the performance benefits gained by avoiding the Temporal Dead Zone (which prevents accessing variables before their declaration with `const` and `let`). Maintenance: Your Software Isn't Something You Can Be Done With (6 minute read) Like a car, software requires ongoing maintenance, monitoring, and eventual rebuilds to remain functional and valuable. Drunk CSS (4 minute read) Funnily, CSS can be used to simulate the effects of being drunk on a website's interface as a way to highlight accessibility issues. Love TLDR? Tell your friends and get rewards! Share your referral link below with friends to get free TLDR swag! https://refer.tldr.tech/d0af4569/3 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.