weeklyfoo #147 is here: your weekly digest of all webdev news you need to know! This time you’ll find 36 valuable links in 5 categories! Enjoy!
🚀 Read it!
- A Runaway Model Hacked Hugging Face to Cheat Its Evals: OpenAI’s writeup of a model that broke out of the eval harness rather than solving the task by OpenAI / ai,security / 1 min read
📰 Good to know
- In-House LLM Serving at Netflix: How Netflix built an in-house LLM serving platform on vLLM and Triton instead of hosted APIs by Netflix / ai,infra / 14 min read
- Fetch Needs Error Codes: Fetch collapses every network failure into a bare TypeError, and a TC39 proposal would thread typed error codes through by James Snell / javascript,web / 17 min read
- No, We Can’t Harden Node.js Against Prototype Pollution: Matteo Collina argues you cannot harden Node core against prototype pollution and the fix belongs at the boundary by Matteo Collina / security,nodejs / 7 min read
- A Practical Guide to Reducing Token Spend: A swamp workflow structures agent tasks into deterministic code and cut token usage 8x by Adam Jacob / ai,agents / 10 min read
- Agent Swarms and the New Model Economics: Planner agents decompose work and cheaper workers execute it, beating a single frontier model on speed, quality and cost by Cursor / ai,agents / 36 min read
- You Only Need the Frontier Model for One Single Edit: Have the frontier model explore and dump its understanding to a todo list before handing execution to a cheaper model by Stencil / ai,agents / 11 min read
- Google Is Building a Chip With Gemini Baked Into the Silicon: Fusing one model’s architecture into the circuitry instead of running any model on general hardware by The Next Web / ai,hardware / 7 min read
- Human Mathematicians Are Being Outcounterexampled: AI tools now solve theorems and write Lean proofs to prove their own results, speeding up formalization research by Kevin Buzzard / ai,math / 20 min read
- One Document, Two Hands: Designing for the case where an agent is a guest in your app, editing the same document you are by Sunil Pai / ai,agents / 10 min read
- Engineering High-Performance Parsers with Data-Oriented Design: How data-oriented design built Yuku, a JavaScript parser faster than Oxc, SWC or Babel by Arshad Yaseen / javascript,performance / 18 min read
- How a Fake Interview’s Coding Challenge Steals Credentials: A North Korean campaign hides malware in SVG images inside a fake job interview JavaScript coding challenge by Elastic / security / 16 min read
- Models Are Worse at Reviewing Their Own Code: Claude Code and Codex each catch more bugs in the other model’s code than their own, so route reviews across models by Greptile / ai,agents / 14 min read
- Making 768 Servers Look Like 1: An animated explainer of how a petabyte of Postgres across 256 shards presents as one database by Ben Dicken / database / 13 min read
- Oxc Type-Aware Linting Is Stable: Rules relying on the TypeScript type system, like unhandled promises and unsafe assignments, now land in the Rust linter by Oxc / javascript,tools / 5 min read
- 53 Slopsquatting Targets Across 5 Popular LLMs: A study of the fake npm package names LLMs keep inventing, and how attackers register them by Sarah Gooding / security,ai / 10 min read
🧰 Tools
- LoopGain: Open-source tool that stops AI agent loops once iterations no longer improve by LoopGain / ai,agents,tools
- Nativ: Free open-source app for Apple Silicon that runs AI models locally with no accounts, subscriptions or cloud by Prince Canuma / ai,tools
- Flint: Visualization language that lets AI agents create charts by Microsoft / ai,tools
- NameThatUI: The Visual Dictionary of UI: See the element, learn its real name, and prompt your coding agent with precision by argofowl / design,tools
- Color.js: Library for working with colors following the latest specs, with new color spaces and smarter gamut mapping in 0.7 by Lea Verou / css,tools
- unwasm: Build-time tooling that lets you import .wasm files like any other ES module by unjs / javascript,tools
- Yuku: A JavaScript parser and analyzer engineered for speed, faster than Oxc, SWC or Babel by Arshad Yaseen / javascript,tools
- A11Y.md: Context system giving AI agents enforceable accessibility rules aligned with WCAG by Felipe Arriaga Carrico / ai,tools
- Puffgres: Rust tool using logical replication to mirror Postgres entities into Turbopuffer, and A24’s first open source release by A24 Films / database,tools
- pg_timetable 7.0: Job Scheduling for Postgres: Stand-alone scheduler storing state in your database, now with OpenTelemetry tracing and arm64 images by CYBERTEC / database,tools
- better-sqlite3 13.0: The popular SQLite library moves to N-API so prebuilt binaries work across Node and Electron versions, plus a new db.explain() by Joshua Wise / database,nodejs,tools
- Execa 10.0: Better process execution from Node, with a trimmed API and a killDescendants option to terminate a subprocess’s descendants by Sindre Sorhus / nodejs,tools
- Claude Thermos: Keeps the prompt cache warm during long agent sessions so waiting on subagents doesn’t re-bill re-encoding by Ivan Zeigerman / ai,agents,tools
- Shadscan: Static analysis for shadcn components with almost 60 deterministic accessibility, state and composition checks by OrcDev / react,tools
- GTKX: Build native Linux desktop apps with React and GTK4, now at 1.0 release candidate by GTKX / react,tools
🤪 Fun
- Decoy Font: A TTF Font That Hides What You Type: Uses separate spatial frequencies to show two different letters in the same space, misleading AI reading your screen by Mixfont / fun / 6 min read
- Which AI?: Daily guessing game where you spot whether a landing page was designed by Gemini, Claude or GPT by Which AI / fun,design / 0 min read
📺 Videos
- Designing 99 ‘No’s for a Single ‘Yes’: Screensharing dozens of unused concepts, novel art directions and brand identities behind one final pick by Michael Riddering / design,vids
- How to Think like a Performance Engineer: Which site-speed tools to trust and how to run tests that are realistic and actionable by Harry Roberts / performance,vids
- WhatsApp Private Processing: Meta engineers on running confidential VMs for privacy-preserving AI at WhatsApp scale by Meta / security,vids