Shopify first released FlashList in 2022 and has now introduced FlashList v2, a complete rewrite of React Native's list component that eliminates the need for devs to provide item size estimates and achieves better load and scroll perf by using React Native's New Architecture for synchronous layout measurements. FlashList v2 is now production-ready and powers thousands of lists in Shopify's mobile app, with 2M+ monthly downloads.
This article outlines best practices for building agentic AI systems based on real production experience. The optimal architecture features a two-tier system with Primary Agents that maintain context and orchestrate tasks, plus stateless Subagents that work like pure functions without memory or state. Critical principles include keeping subagents stateless for parallel execution, using structured communication protocols, and employing orchestration patterns like MapReduce and Sequential Pipeline, depending on task dependencies.
Key practices to scaling code reviews include keeping changes small, responding to reviews within 24 hours, and using automated checks like unit tests and static analysis to let reviewers focus on high-level feedback. This article advocates for a structured 5-step workflow from PR creation through deployment, complemented by AI tools to reduce review time and improve developer velocity.
2025.08.27·
newsletter.systemdesign.one·4 min read·TLDR Web Dev
Cronicle is a multi-server task scheduler and runner with a web-based front-end UI that handles both scheduled, repeating, and on-demand jobs across multiple slave servers. It provides real-time statistics and a live log viewer for monitoring job execution.
JavaScript continues to evolve rapidly. The TC39 committee has advanced nine proposals, including the newly standardized explicit resource management feature using the using keyword for automatic cleanup, and Array.fromAsync for handling async iterables. Other notable developments include candidate features like immutable ArrayBuffers, draft proposals for deterministic random number generation and number clamping methods, and early-stage proposals for improved internationalization formatting and comparison utilities.
Bugs sneak out when less than 80% of user flows are tested before shipping. However, getting that kind of coverage (and staying there) is hard and pricey for any team.
Nvidia's Jetson AGX Thor robotics chip module is now on sale for $3,499 as a developer kit. The first kits will ship next month. The chips will allow customers to create robots that run generative AI models. They can also be used for self-driving cars. The Jetson Thor chips are based on Nvidia's Blackwell graphics processor, which is also used in the company's current generation of AI chips and chips for computer games.
Google has announced plans to begin verifying the identities of all Android app developers, no matter where they offer their content. Apps without verification will stop working on most Android devices in the coming years. The company plans to create a streamlined Android Developer Console that devs can use if they plan to distribute apps outside of the Play Store. Google won't check the content or functionality of apps outside of the Play Store.
UltraRAM promises DRAM-like speeds, 4,000x the durability of NAND, and data retention for up to a thousand years. Quinas Technology and IQE have successfully developed a scalable process for producing UltraRAM. The companies are now weighing pilot production with various foundries and other collaborators.
Standard Thermal is a company that aims to make energy from solar PV available at all times at a price that is competitive with US natural gas. The company's technology works by storing energy as heat in large piles of dirt. Electric heating elements embedded in the dirt convert electricity to heat, and pipes carrying fluid remove the heat to supply customers. The capital cost is comparable to natural gas storage at less than $0.10 per kilowatt-hour thermal and a thousand times cheaper than batteries. Standard Thermal's system can provide hundreds of megawatts of thermal demand as long as land is available.
FastMCP Cloud is here with production-ready AI infrastructure that just works. From the team behind FastMCP, instantly deploy your MCP servers as remote, authenticated endpoints for use with Claude, Cursor, or your favorite LLM. Get started for free [12] or check out
It's possible to dynamically change a function's source code at run time in Python using the compile and exec functions. This technique enables developers to build more flexible AI bots that can generate and execute code with access to the current environment. However, it also raises some serious security concerns. This post walks readers through how the technique works and the real security concerns with using the approach.
Prediction-Encoded Pixels is a format specifically designed for low-color pixel art. It uses 'Prediction by Partial Matching, Order-2' compression, which is able to compress packed-palette-indices smaller than GIF, PNG, and QOI. It's two to 10 times slower than GIF/PNG/QOI, depending on the image, but it often compresses images 20% to 50% smaller than GIF/PNG, and multiple times smaller than QOI. The format is still in an experimental phase.
Prediction markets don't provide better predictions about the future. The best, most efficient financial markets have standardized products, many participants, low transaction costs, and heterogeneous participants. Prediction markets lack participants with heterogeneous risk preferences. It's also possible for participants to manipulate some outcomes. Prediction markets are bad market design, and they induce bad incentives, so we should put the genie back in the box while there's still time.
President Donald Trump told Lip-Bu Tan, Intel's chief executive, to step down five months into his tenure because of his past ties to the Chinese military. Tan eventually smoothed things over with Trump, but at a cost: in return for his support, the administration converted nearly $9 billion in grants promised to Intel as part of the 2022 Chips Act into a 10% equity stake in the company. The US government is now Intel's biggest shareholder. Tan still needs to resolve core business challenges to get Intel back on track. Without new commitments from customers, converting government grants to stock will only leave the company in a worse financial shape by diluting shareholders.
Apple is ending its 'Friday Night Baseball' partnership with MLB after the 2025 season due to low viewership, high costs, and fan frustration with fragmented streaming access.
The sequence of 'revolutionary technology, abundant capital, speculative frenzy, and sudden reality checks' has played out with remarkable consistency for over 180 years.
David Lu, who worked for Eaton Corporation, wrote malicious code that crashed the company's network after a corporate realignment reduced his job responsibilities and restricted his access to certain servers.
Generative AI has turned financial fraud into an industrial operation. Voice cloning now takes just an hour of YouTube footage and an $11 subscription. Synthetic identity fraud already costs banks $6 billion annually. And fraudsters are scaling attacks using AI tools that traditional defenses can't handle.
Elon Musk's xAI filed a lawsuit accusing Apple and OpenAI of colluding to suppress competition in AI, citing Apple's partnership to embed ChatGPT in its ecosystem as anticompetitive behavior.
Perplexity has launched a $5 standalone subscription that gives users access to premium publisher content by sharing revenue with participating publishers based on human visits and AI interactions.
Modern language models use a building block called sliding window attention to handle long texts efficiently. The effective memory of these models has fundamental limitations. They struggle to use information from more than about 1,500 words ago, far less than the theoretical 100,000. This is because information gets diluted as it spreads through the network, and residual connections create an exponential barrier that blocks distant information.
In this interview, Jeff Huber from Chroma talks about what matters in vector databases, why modern search for AI is different, and how to ship systems that don't rot as context grows. Context engineering is the job of figuring out what should be in the context window at any given LLM generation step. It involves an inner loop, setting up the context window 'this time', and an outer loop that involves getting better at filling the context window with only relevant information.
Training frontier models? Turing [9] remains the independent choice for the world's top AI teams — from multimodal training to RL gyms. 1000s of domain-trained experts across 60+ languages, 700+ benchmark tasks built from real-world business and STEM use cases, 50+ frontier
ThinkMesh is a Python library for running diverse reasoning paths in parallel, scoring them with internal confidence signals, reallocating compute to promising branches, and fusing outcomes with verifiers and reducers. It works with offline Hugging Face Transformers, vLLM/TGI, and hosted APIs.ThinkMesh is confidence‑gated, strategy‑driven, and offline‑friendly. It is still in an early development stage.
This framework enhances Claude Code with 21 new slash commands, 14 specialized AI agents, and 6 MCP server integrations to create structured workflows for development tasks. The project has gained 14k GitHub stars in a month and claims to reduce context usage by 30-50%.
The IVA framework trains Vision-Language-Action models to detect and respond to instructions based on false premises, using clarification and alternative grounding strategies.
Memento is a memory-based framework that enables LLM agents to learn continually without altering model weights. It uses a planner–executor loop with case-based reasoning and supports a wide range of tools.
Perplexity will start paying publishers for news articles from a $42.5 million revenue pool. The money will be distributed when Perplexity's AI assistant or search engine uses a news article to fulfill a task or answer a search request. The payments come out of the subscription revenue generated by Perplexity's new news service, Comet Plus, which will roll out widely in the fall. Publishers will get 80% of Comet Plus revenue.
Sebastien Bubeck, an employee at OpenAI, claims that gpt-5-pro can prove new interesting mathematics. Bubeck took the first version of a paper on convex optimization with a clean open problem in it and asked gpt-5-pro to work on it. The model proved a better bound than what was in the paper. The proof is correct - the second version of the paper (already published) had already reached this conclusion.
Google's Gemini Live AI assistant will soon highlight items on your screen using your camera, initially available on the new Pixel 10 and other Android devices.
Andreessen Horowitz and OpenAI President Greg Brockman are among those launching the "Leading the Future" super-PAC network, which has already raised over $100 million to oppose strict AI regulations and support candidates favorable to the industry in the 2026 congressional races.
MCPMark is a challenging benchmark built to stress-test models with comprehensive tasks that balance different CRUD (create, read, update, and delete) operations while enforcing strict environment management.
Global financial regulators, including ESMA, IOSCO, and the World Federation of Exchanges, are urging the SEC to strengthen oversight of tokenized stocks, arguing these products misleadingly mimic traditional equities without providing equivalent investor protections. The organizations have expressed alarm at the growing number of brokers and crypto platforms offering tokenized stocks, claiming they are marketed as equivalent to actual stocks when they fundamentally are not, potentially exposing investors to significant risks. This regulatory pushback comes as tokenized assets reach $26 billion in value and major platforms like Coinbase, Kraken, and Robinhood expand into the space, mirroring previous successful lobbying efforts by traditional financial institutions to limit blockchain-based competition. Despite these concerns, SEC leadership has indicated openness to tokenization innovation while emphasizing that tokenized securities must still comply with existing securities laws.