Signup here for the newsletter to get the weekly digest right into your inbox.
weeklyfoo #158 is here: your weekly digest of all webdev news you need to know! This time you'll find 40 valuable links in 5 categories! Enjoy!
🚀 Read it!
- How ChatGPT Intelligent UI works: Reverse-engineering tour from a model-written DIL format through server-side compilation, sandboxed client execution and native rendering, including how streaming recompiles partial output and preserves keyed state by OpenUI / ai, ui, frontend / 22 min read
📰 Good to know
- Agents don't need memory. They need documentation.: Vector-search memory surfaces similar snippets without proving they are current or relevant, while a structured Markdown workspace of specs, decisions and instructions gives agents auditable context by liao.gg / ai, agents / 6 min read
- Ephemeral testing: Test a software component by having an agent build disposable applications on top of it, exposing hidden state, unstable invariants, surprising defaults and weak documentation by Daniel Lemire / ai, testing / 9 min read
- Introducing Beam, Reflection's 501B open-weight model: Sparse Mixture-of-Experts model with 501 billion total and 23 billion active parameters aimed at coding, reasoning and agentic workloads, with Apache 2.0 weights planned for later this month by Reflection AI / ai, llm, opensource / 19 min read
- Capturing and replaying real database workloads at Airbnb: Airbnb captures MySQL traffic at ProxySQL, rebuilds transaction order offline and replays production workloads for load testing, migration checks and the move from MySQL 5.7 to 8.0 by Airbnb / databases, mysql, testing / 13 min read
- The agent said it was done. The database disagreed: ThinkingBox evaluates agents by the backend state they leave behind across 507 workflows, where 67% of 79,853 failed runs ended without a tool error by Microsoft / ai, agents, evals / 19 min read
- Introducing Mistral Large 4: One-trillion-parameter multimodal model in public preview with 49 billion active parameters, 160+ languages and open weights planned for later this month by Mistral AI / ai, llm, opensource / 13 min read
- A terminal protocol for program status (OSC 7501): Spec for an escape sequence that lets any program tell the terminal whether it is idling, working, waiting on the user, finished or failed, and why by Mitchell Hashimoto / terminal, tools / 7 min read
- Upgrading a 56TB database by mostly not moving it: Facing a mandatory Aurora reboot deadline, Trigger moved just 2% of its data to PlanetScale, with a candid postmortem covering a pgcopydb patch and a failed first cutover by Daniel Sutton / postgres, databases, migration / 21 min read
- pgvector 0.8.7 released: Fixes a bug where creating an IVFFlat index could lead to arbitrary code execution by pgvector / postgres, security, vector / 1 min read
- A hitchhiker's guide to Postgres text search: Cortex replaced a Lucene-based search architecture with Postgres full-text search, a custom query parser, weighted ranking and CDC-fed indexes, cutting public API p95 latency from 13.3 to 1.54 seconds by Cortex / postgres, search, databases / 16 min read
- Claude Haiku 5.5: Fast, low-cost model for summarization, classification, browser use and subagent tasks, with average running costs about 75% lower than Haiku 4.5 and adjustable effort by Anthropic / ai, llm, anthropic / 11 min read
- GPT-6 and Intelligent UI for everyone: GPT-6 rolls out in ChatGPT with Intelligent UI, which combines text, graphics, forms, charts and interactive tools in a response and can start answering while still reasoning or searching by OpenAI / ai, llm, openai / 1 min read
- We built our own cloud agents runtime: The shared runtime behind PostHog's agents built on Temporal workflows, VM sandboxes, snapshots and resumable run logs, with every run designed to survive the loss of its sandbox by PostHog / ai, agents, infrastructure / 13 min read
- Don't merge what you didn't run: A test gate for agent-written code that reinstalls from scratch, runs the full suite three times, rejects removed or skipped tests and restores the base branch's tests against the new code by Runtime / ai, agents, testing / 11 min read
- A single POST freezes any Next.js server: React Server Components flaw where a roughly 900 KB request forces quadratic FormData work and blocks the Node.js event loop, fixed in React 19.0.6, 19.1.7 and 19.2.6 by Simon Koeck / security, react, nextjs / 6 min read
- Deno joins Cloudflare: Cloudflare is acquiring Deno, with a Rust rewrite of workerd, first-class self-hosting for Durable Objects and a secured future for the Deno runtime on the roadmap by Cloudflare / deno, cloudflare, javascript / 26 min read
- Introducing the Personal Agent Protocol: A new protocol for personal agents that is built on top of OAuth by Bret Taylor, Clay Bavor / ai, agents, auth / 8 min read
🧰 Tools
- Panda CSS 2.0: Same css() API with a much faster compiler rebuilt in Rust, plus publishable design systems, view-transition styling, an ESLint plugin and a new typography preset by Panda CSS / css, cssinjs, frontend
- Postgres MCP Server: Microsoft's Rust-powered MCP server, installed via npm, that lets AI coding agents work with Postgres on Azure or anywhere else by Microsoft / postgres, mcp, ai
- pgx-bm25 1.0: Pure C BM25 ranked full-text search extension whose indexes live in the index relation's pages, so WAL logging, replication and VACUUM work as usual by PGX Inc. / postgres, search, opensource
- Next.js 16.4: Cache Components on by default for new apps, static-render guarantees, finer prefetch controls, agent-assisted upgrades and React 19.3 by Vercel / nextjs, react, frontend
- Remix 3: Rebuilt full-stack framework with no React required, shipping a type-safe router, auth, an ORM, an asset server and its own component model by Remix / javascript, framework, fullstack
- Lightpanda 1.0: Headless browser built from scratch in Zig that uses a fraction of Chrome's CPU and memory, driven by Puppeteer, Playwright or plain JavaScript by Francis Bouvier / browser, automation, zig
- Pi 1.0: The minimal Node.js-based coding agent adds MCP and non-LLM model support through Codemode by Earendil / ai, agents, nodejs
- JavaScript Framework Benchmark for Chrome 154: Results for 196 ways to implement the same app, how each performs and how large they are by Stefan Krause / javascript, performance, benchmark
- e2e: Agent-driven end-to-end testing for web and mobile apps that records successful agent steps and replays them without model calls until the app changes by Tester Army / ai, testing, playwright
- TanStack Charts 1.0: Composable, type-safe charting system built from marks, scales, axes and interactions instead of rigid chart types, with custom renderers and SVG or Canvas output by TanStack / charts, typescript, frontend
- TanStack Intent: CLI to generate, validate and ship agent skills alongside your library, versioned with the package and discovered by agents from node_modules by TanStack / ai, agents, tooling
- Node.js 26.11: New --process-timeout flag that ends a process after a set time even mid-busy-loop, virtual table support in node:sqlite, Buffer.stringLength() and tier 2 support for Alpine on x64 by Antoine du Hamel / nodejs, javascript
- H3 v2: The minimal HTTP framework behind Nitro and Nuxt rebuilt around standard Request and Response, so one app runs on Node, Bun, Deno and Workers by Pooya Parsa / javascript, http, framework
- pnpm 12.10: Experimental loaded linker lets Node load dependencies straight from the pnpm store, skipping node_modules linking by pnpm / pnpm, nodejs, tooling
- Hono 5.0 RC: The web-standards server framework goes ESM-only, requires Node 22.12+ and moves runtime adapters into separate packages by Hono / javascript, framework
- Bootstrap 6 alpha: Ground-up modernization with the Sass module system, more native browser APIs and elements, ESM-only JavaScript plugins and a slew of new CSS standards by Bootstrap / css, frontend, bootstrap
- calc-input: Custom input element that accepts arithmetic expressions, shows the formula on focus and the result on blur, with a safe parser and native form validation instead of eval() by Bramus Van Damme / webcomponents, forms, frontend
🤪 Fun
- Destroy this website: Steer a tiny orange character that jumps across headings and images and shoots page elements to pieces by Sprite Fusion / fun, web / 3 min read
- I put Smash Bros Melee in my living room: Hours after the Melee decompilation hit 100%, a developer had Fox and Marth fighting in their apartment through a Meta Quest, with a write-up of how the match got into the headset by Kevin / fun, gaming, vr / 31 min read
📺 Videos
- Building resilient systems with Sam Newman: The Building Microservices author on microservices as a last resort, three rules of distributed systems, idempotency keys versus fingerprints and resisting cognitive surrender to AI by Gergely Orosz / architecture, distributedsystems, ai
- Harness engineering on Rails: An agent is really a loop, so move judgment out of AGENTS.md and into deterministic tools like custom RuboCop cops, generators, hooks and focused review skills by Joël Quenneville / ai, agents, rails
- Responsibly dimming the lights in the software factory: Steps and milestones for raising the quality of AI-generated code until code review intensity can safely go down, without cognitive debt and skill erosion by Victor Rentea / ai, agents, codereview
Want to read more? Check out the full article here.
To sign up for the weekly newsletter, visit weeklyfoo.com.
Top comments (0)