Forem: Adam

10x TypeScript, Intent Debt, and Patch That Kernel

Adam — Fri, 08 May 2026 04:43:24 +0000

TypeScript 7.0 Beta drops with a Go-powered rewrite and roughly 10x compiler performance — your morning coffee break just got a lot shorter.

The AI discourse this week is having an identity crisis. Martin Fowler argues that treating prompts as first-class, version-controlled artifacts is the next maturity leap for LLM-assisted development. Tim Kellogg maps out three mutable agent memory types — Files, Memory Blocks, and Skills — and why editable memory is the real unlock. Simme introduces "intent debt" — the hidden cost of deploying agents without documenting what they're actually supposed to do — while The Last Software Engineer argues that when agents handle all implementation, judgment is all that's left. And Mario Zechner and Armin Ronacher, who actually shipped production agentic systems, are walking back their own enthusiasm: junior engineers still win on value.

Security corner: Andrew Nesbitt shows how insecure Actions defaults make GitHub CI the weakest link in your supply chain, and CVE-2026-31431 is a Linux kernel privilege escalation hiding since 2017 — any unprivileged user, Python script, root access. Patch now.

Stripe runs 5% of tests per CI build in a 50M-line monorepo using C++ file-access tracking. If your test suite runs everything on every PR, this one should sting a little. Josh Comeau returns with the definitive guide to CSS scroll-driven animations — no JavaScript, no excuses.

On tools: Dirac cuts AI coding agent API costs by 64.8% using AST manipulation, and Perry compiles TypeScript straight to native executables without a runtime dependency.

Enjoy!

Signup here for the newsletter to get the weekly digest right into your inbox.

Find the 12 highlighted links of weeklyfoo #135:

Structured Prompt-Driven Development

by Martin Fowler

Treating AI prompts as first-class version-controlled artifacts to make LLM-assisted changes governable, reviewable, and reusable

🚀 Read it!, ai, engineering, prompting

Selective Test Execution at Stripe

by Stripe Engineering

How Stripe runs only 5% of tests per CI build in a 50M-line Ruby monorepo — C++ file access tracking selects exactly which tests to run based on changed code paths

📰 Good to know, engineering, ci, testing

Agent Memory Patterns

by Tim Kellogg

The three types of mutable memory for agents — Files, Memory Blocks, and Skills — and how editable memory makes agents dramatically more capable

📰 Good to know, ai, agents

TypeScript 7.0 Beta: 10x Faster Compilation

by Microsoft

The Go-powered TypeScript port in beta — about 10x faster compiler performance, already close to production-ready

📰 Good to know, typescript, javascript

GitHub Actions Is the Weakest Link

by Andrew Nesbitt

How insecure defaults make Actions a major supply chain attack vector — and what maintainers can do today without waiting for GitHub to fix the defaults

📰 Good to know, security, ci, supply-chain

Scroll-Driven Animations

by Josh W. Comeau

Deep dive into the new CSS animation-timeline API for native scroll-driven animations without JavaScript

📰 Good to know, css, frontend, animation

Building Pi and What Makes Self-Modifying Software Fascinating

by Gergely Orosz

Mario Zechner and Armin Ronacher on automation bias, declining AI code quality, and why junior engineers are more valuable than AI agents

📰 Good to know, ai, agents, engineering, opinion

The Last Software Engineer

by Epic Product Engineer

As AI agents automate implementation, engineering value shifts to judgment — what to build and why, and owning the consequences of those decisions

📰 Good to know, ai, engineering, opinion

The End of Just Ask Sarah

by Simme

AI agents only know their context window — the shift from human institutional memory to explicit machine-readable docs reveals a new kind of technical debt called intent debt

📰 Good to know, ai, agents, documentation

Copy Fail (CVE-2026-31431)

by security researchers

Critical Linux kernel flaw since 2017 — unprivileged local users gain root access via a Python script, posing severe risk to container clusters without kernel update or disabling algif_aead

📰 Good to know, security, linux

Dirac

by dirac-run

Open-source AI coding agent for VS Code and CLI that cuts API costs by 64.8% through AST manipulation and optimized context curation

🧰 Tools, ai, coding, tools

Perry

by perryts

Cross-platform TypeScript compiler that compiles directly to native executables without a runtime

🧰 Tools, typescript, tools

Want to read more? Check out the full article here.

To sign up for the weekly newsletter, visit weeklyfoo.com.

Stay ahead in web development: latest news, tools, and insights #135

Adam — Mon, 04 May 2026 04:47:14 +0000

Signup here for the newsletter to get the weekly digest right into your inbox.

weeklyfoo #135 is here: your weekly digest of all webdev news you need to know! This time you'll find 35 valuable links in 4 categories! Enjoy!

🚀 Read it!

Structured Prompt-Driven Development: Treating AI prompts as first-class version-controlled artifacts to make LLM-assisted changes governable, reviewable, and reusable by Martin Fowler / ai, engineering, prompting / 37 min read

📰 Good to know

Selective Test Execution at Stripe: How Stripe runs only 5% of tests per CI build in a 50M-line Ruby monorepo — C++ file access tracking selects exactly which tests to run based on changed code paths by Stripe Engineering / engineering, ci, testing / 2 min read
Thoughts and Feelings Around Claude Design: Figma's locked-down format kept it out of AI training data while LLMs learned code — the source of truth is shifting back to code, making Figma look pre-AI by Sam Henri Gold / ai, design / 7 min read
The Woes of Sanitizing SVGs: How Scratch's repeated manual SVG sanitization failures led to XSS and HTTP leaks — and why iframe+CSP sandboxing is the only sustainable path by muffin / security, frontend / 16 min read
Agent Memory Patterns: The three types of mutable memory for agents — Files, Memory Blocks, and Skills — and how editable memory makes agents dramatically more capable by Tim Kellogg / ai, agents / 8 min read
AI Fluency Index: Anthropic studied 9,830 Claude conversations and identified 24 fluency behaviors — only 30% of users set collaboration terms upfront, showing most are stuck at the prompt-and-fix level by Anthropic / ai, research / 14 min read
How AI Is Changing Operating Systems: Ubuntu's approach to AI hardware enablement — native CUDA packaging, NPU support, x86_64 architecture variants, and early plans for agentic workflows at the OS level by Gergely Orosz / engineering, ai, linux / 14 min read
TypeScript 7.0 Beta: 10x Faster Compilation: The Go-powered TypeScript port in beta — about 10x faster compiler performance, already close to production-ready by Microsoft / typescript, javascript / 20 min read
GitHub Actions Is the Weakest Link: How insecure defaults make Actions a major supply chain attack vector — and what maintainers can do today without waiting for GitHub to fix the defaults by Andrew Nesbitt / security, ci, supply-chain / 12 min read
Scroll-Driven Animations: Deep dive into the new CSS animation-timeline API for native scroll-driven animations without JavaScript by Josh W. Comeau / css, frontend, animation / 20 min read
Why Ghostty Is Leaving GitHub: Mitchell Hashimoto moves Ghostty off GitHub after 18 years due to frequent outages affecting professional collaboration by Mitchell Hashimoto / tools, engineering, opinion / 5 min read
What's Actually New in JavaScript: All the useful features from ES2025 and ES2026 in one place: iterator helpers, Promise.try, Map.getOrInsert, Temporal, and more by Neciu Dan / javascript, frontend / 35 min read
Building Pi and What Makes Self-Modifying Software Fascinating: Mario Zechner and Armin Ronacher on automation bias, declining AI code quality, and why junior engineers are more valuable than AI agents by Gergely Orosz / ai, agents, engineering, opinion / 9 min read
The End of Responsive Images: Mat Marquis reflects on 14 years of srcset/sizes headaches — the new sizes=auto for lazy-loaded images is now cross-browser and eliminates manual calculations entirely by Mat Marquis / css, frontend, responsive / 20 min read
Constructable Stylesheets and adoptedStyleSheets: One Parse, Every Shadow Root: Create CSSStyleSheet objects in JavaScript and share them across Shadow DOM — parse CSS once per component class instead of per element by Rob Levin / css, web-components, frontend / 14 min read
React Compiler at 18 Months: Eighteen months in, the compiler's real wins are eliminating forgotten-dependency bugs and missing memos — not benchmark speedups. Adoption blockers in brownfield apps and what's next by Sascha Bäuerlein / react, javascript, engineering / 12 min read
The Last Software Engineer: As AI agents automate implementation, engineering value shifts to judgment — what to build and why, and owning the consequences of those decisions by Epic Product Engineer / ai, engineering, opinion / 10 min read
The End of Just Ask Sarah: AI agents only know their context window — the shift from human institutional memory to explicit machine-readable docs reveals a new kind of technical debt called intent debt by Simme / ai, agents, documentation / 8 min read
Copy Fail (CVE-2026-31431): Critical Linux kernel flaw since 2017 — unprivileged local users gain root access via a Python script, posing severe risk to container clusters without kernel update or disabling algif_aead by security researchers / security, linux / 5 min read
Fresh 2.3: Deno's full-stack framework gains first-class WebSocket support, ships zero JS for static pages, and adds one-attribute View Transitions API integration by Bartek Iwańczuk / javascript, deno, frontend / 10 min read
The Pulse: Token Spend Breaks Budgets: Companies are hitting AI token budget overruns and switching strategies — from real-time measurement to cheaper models and local alternatives by Gergely Orosz / ai, engineering / 17 min read
Firefox Maker Pushes Back Against Google's Prompt API: Mozilla argues Chrome's built-in LLM Prompt API harms web interoperability and neutrality — the API lets web pages directly prompt a browser-provided language model by Mozilla / browser, ai, standards / 11 min read

🧰 Tools

AIMAC: Tests 43 AI models by prompting them to build web pages and auditing the output for accessibility violations by AIMAC / ai, accessibility, testing
Awesome Harness Engineering: Awesome list for AI agent harness engineering: tools, patterns, evals, memory, MCP, permissions, observability, and orchestration. by ai-boost / ai, tools
Skills: Skills for Real Engineers. Straight from my .claude directory. by Matt Pocock / skills, ai
wuphf: Slack for AI employees that build and maintain their own wiki. Get Claudes, Codexes, OpenClaws and local LLMs to collaborate and do your work autonomously while never losing context. by wuphf.team / ai, agents
Dirac: Open-source AI coding agent for VS Code and CLI that cuts API costs by 64.8% through AST manipulation and optimized context curation by dirac-run / ai, coding, tools
Pi: Minimalist self-modifying AI coding agent built for stability — designed because Claude Code became unpredictable as features accumulated by Mario Zechner / ai, agents, tools
Honker: Postgres NOTIFY/LISTEN semantics for SQLite — low-latency cross-process events without a daemon or broker by Russell Romney / databases, sqlite, tools
OpenSpec: Spec-driven development framework that adds a lightweight specification layer between humans and AI coding assistants — align on requirements before any code is written by Fission AI / ai, tools, github
Perry: Cross-platform TypeScript compiler that compiles directly to native executables without a runtime by perryts / typescript, tools
Pu.sh: Minimalist coding-agent harness in 400 lines of shell script — a lightweight alternative to heavier AI coding frameworks by pu.sh / ai, agents, tools

📺 Videos

Prompting 101: Anthropic's applied AI team just showed how to actually prompt Claude properly. by Hannah Moran, Christian Ryan / ai, anthropic
Design at Scale at Ramp: Diego Zaks on how Ramp uses AI at 1,500 people — 99.5% daily usage, 1,500 internal apps in six weeks, non-engineers writing 12% of production code by UX Tools / ai, design, agents
OpenClaw for Designers: Tommy's walkthrough of his AI agent harness for designers — SOUL.md, USER.md, MCP connections to Figma and Slack, and what it actually costs to run a self-improving design stack by UX Tools / ai, design, agents

Want to read more? Check out the full article here.

To sign up for the weekly newsletter, visit weeklyfoo.com.

Agent Vaults, Slop Cops, and No More PRs

Adam — Fri, 01 May 2026 05:13:28 +0000

The theme this week writes itself: AI agents are running production systems, they're getting faster to deploy, they're occasionally breaking in, and someone finally built a lockbox for your API keys.

Addy Osmani charts four architectural bets for the next generation of agentic systems. Cloudflare shares actual numbers: 93% of their R&D now uses AI tools built on their own platform, pushing weekly merge requests from 5,600 to over 8,700 — the most concrete productivity benchmark any major infrastructure company has published. Slack Engineering rounds it out with the hardest problem: keeping long-running agents on-mission without context drift.

The security picture is getting complicated. The Vercel April 2026 incident is essential reading — a compromised third-party AI tool gave attackers access to sensitive customer data, and Guillermo explicitly called out the role of AI in accelerating the attack. Agent Vault from Infisical is the direct response: an open-source HTTP credential proxy that prevents AI agents from ever directly touching your API keys, blocking both accidental exfiltration and prompt injection at the network layer.

Meanwhile, the ecosystem is quietly reshuffling. One open-source maintainer is no longer accepting external PRs — LLM-generated code only, human contributions redirected to feedback and architecture. Evil Martians maps which LLM discoverability techniques actually work versus the eight that sound plausible but don't. Frontend Masters delivers the accessibility wake-up call: AI-generated UI looks right but fails screen readers, optimizing for visual fidelity while ignoring the accessibility tree entirely.

On tools: Rspack 2.0 ships meaningful Rust-powered build performance without touching your existing webpack config, Slop Cop catches LLM prose patterns before your writing sounds like everyone else's AI output, and aube enters the package manager competition from the creator of mise — performance-first, security-conscious, drop-in npm replacement.

Enjoy!

Signup here for the newsletter to get the weekly digest right into your inbox.

Find the 11 highlighted links of weeklyfoo #134:

The Agent Stack Bet

by Addy Osmani

Most production agents today feature fragile session logic, shared service accounts, and weak security models — four architectural bets for the next generation of agentic systems

🚀 Read it!, ai, engineering

Vercel April 2026 Security Incident

by Vercel

A compromised third-party AI tool gave attackers access to sensitive Vercel customer data — Guillermo noted the attack was significantly accelerated by AI

📰 Good to know, security

The AI Engineering Stack Cloudflare Built Internally

by Cloudflare

93% of Cloudflare's R&D uses AI tools built on their own platform — MCP servers, access layer, and agent tooling pushed weekly merge requests from 5,600 to over 8,700

📰 Good to know, ai, engineering

Managing Context in Long-Run Agentic Applications

by Slack Engineering

Slack Engineering on maintaining alignment across complex long-running agent systems — techniques for balancing continuity and creativity in multi-agent teams

📰 Good to know, ai, engineering, agents

I Don't Want Your PRs Anymore

by dpc.pw

An open source maintainer closes external PRs in favor of LLM-generated code — future contributions should focus on feedback, architecture, and bug reports

📰 Good to know, ai, engineering, open-source

Making Your Site Visible to LLMs: 6 Techniques That Work, 8 That Don't

by Evil Martians

Practical guide to LLM discoverability — structured data, llms.txt, clean semantic HTML, and eight techniques that actually don't help

📰 Good to know, ai, seo, frontend

AI-Generated UI Is Inaccessible by Default

by Frontend Masters

AI tools produce React components that look correct but fail screen readers — div soup with no roles, keyboard support, or ARIA state because models optimize for visual fidelity while ignoring the accessibility tree

📰 Good to know, ai, accessibility, frontend

Rspack 2.0

by Rspack

Rust-powered webpack-compatible bundler ships v2.0 with significantly faster builds, improved ESM output, and broad framework support across Next.js, Vue, Svelte, and Solid

📰 Good to know, javascript, tools, performance

Slop Cop

by awnist

Browser-based writing editor that flags rhetorical and structural patterns common in generic LLM prose

🧰 Tools, ai, writing, tools

aube

by Jeff Dickey

New Node.js package manager from the creator of mise — drop-in replacement for npm with raw performance focus and security-conscious defaults

🧰 Tools, node, tools

Agent Vault

by Infisical

Open-source HTTP credential proxy that prevents AI agents from handling sensitive API keys directly — transparently injects credentials at the network layer to block exfiltration and prompt injection

🧰 Tools, ai, security, tools, github

Want to read more? Check out the full article here.

To sign up for the weekly newsletter, visit weeklyfoo.com.

Stay ahead in web development: latest news, tools, and insights #134

Adam — Mon, 27 Apr 2026 04:57:32 +0000

Signup here for the newsletter to get the weekly digest right into your inbox.

weeklyfoo #134 is here: your weekly digest of all webdev news you need to know! This time you'll find 32 valuable links in 5 categories! Enjoy!

🚀 Read it!

The Agent Stack Bet: Most production agents today feature fragile session logic, shared service accounts, and weak security models — four architectural bets for the next generation of agentic systems by Addy Osmani / ai, engineering / 8 min read

📰 Good to know

Introducing Claude Design: Anthropic's new product for creating designs, prototypes, and presentations using Claude Opus 4.7 — available in research preview for Pro, Max, Team, and Enterprise by Anthropic / ai, design / 7 min read
Claude Opus 4.7 System Prompt Changes: Claude's system prompt changes from Opus 4.6 to 4.7 — less pushy, more concise, new browser and Office agent tools, expanded safety guidelines by Simon Willison / ai / 7 min read
Six Levels of Dark Mode: Eight progressive methods for dark mode implementation — from basic HTML meta tags and CSS properties to advanced media queries and JavaScript color scheme switching by Matthias Beitl / css, frontend / 10 min read
Vercel April 2026 Security Incident: A compromised third-party AI tool gave attackers access to sensitive Vercel customer data — Guillermo noted the attack was significantly accelerated by AI by Vercel / security / 7 min read
Building a UI Without Breakpoints: Using flex calc(infinity) and container queries to build responsive layouts that naturally adapt without explicit media query breakpoints by Frontend Masters / css, frontend / 14 min read
The AI Engineering Stack Cloudflare Built Internally: 93% of Cloudflare's R&D uses AI tools built on their own platform — MCP servers, access layer, and agent tooling pushed weekly merge requests from 5,600 to over 8,700 by Cloudflare / ai, engineering / 23 min read
Managing Context in Long-Run Agentic Applications: Slack Engineering on maintaining alignment across complex long-running agent systems — techniques for balancing continuity and creativity in multi-agent teams by Slack Engineering / ai, engineering, agents / 17 min read
OWASP NPM Security Best Practices: Continually updated checklist covering lifecycle script disabling, typosquatting, trusted publishing, and dependency confusion by OWASP / security, npm / 22 min read
I Don't Want Your PRs Anymore: An open source maintainer closes external PRs in favor of LLM-generated code — future contributions should focus on feedback, architecture, and bug reports by dpc.pw / ai, engineering, open-source / 5 min read
Making Your Site Visible to LLMs: 6 Techniques That Work, 8 That Don't: Practical guide to LLM discoverability — structured data, llms.txt, clean semantic HTML, and eight techniques that actually don't help by Evil Martians / ai, seo, frontend / 20 min read
AI-Generated UI Is Inaccessible by Default: AI tools produce React components that look correct but fail screen readers — div soup with no roles, keyboard support, or ARIA state because models optimize for visual fidelity while ignoring the accessibility tree by Frontend Masters / ai, accessibility, frontend / 21 min read
Rspack 2.0: Rust-powered webpack-compatible bundler ships v2.0 with significantly faster builds, improved ESM output, and broad framework support across Next.js, Vue, Svelte, and Solid by Rspack / javascript, tools, performance / 18 min read
Details That Make Interfaces Feel Better: Small UI polish details that compound into better-feeling interfaces — balanced text wrapping, concentric border radius, tabular numbers, interruptible CSS transitions, and more by Jakub / css, frontend / 10 min read

🧰 Tools

MDV: Markdown superset with embedded charts, KPI cards, and data visualizations using declarative syntax — renders to self-contained HTML or PDF with a VS Code live preview extension by drasimwagan / markdown, tools, github
Slop Cop: Browser-based writing editor that flags rhetorical and structural patterns common in generic LLM prose by awnist / ai, writing, tools
Yuku: High-performance JavaScript/TypeScript compiler and toolchain written in Zig — 100% JS spec compliant by Yuku Toolchain / javascript, typescript, tools, github
Claude Context: MCP plugin that gives AI coding agents deep codebase understanding — indexes millions of code lines with vector search and AST-based chunking to cut token usage by Zilliz / ai, tools, mcp, github
Cloudflare cf CLI: Preview of Cloudflare's new CLI tool for working with its various services — local dev and resource explorer features by Cloudflare / tools, cloudflare
Laws of Software Engineering: Collection of 56 principles and patterns that shape software systems, teams, and decisions by Community / engineering, tools
Is It Agent Ready?: Scan any website to check how well it's structured for AI agent access — checks llms.txt, structured data, and agent-readable content by Is It Agent Ready / ai, tools, agents
Argent: Agentic toolkit for AI assistants to autonomously control, profile, and debug iOS Simulators with native React Native support — operates locally without data collection by Software Mansion / ai, tools, ios, github
Broccoli: Converts Linear tickets into GitHub Pull Requests using AI models like Claude and Codex — runs entirely in your own Google Cloud project for privacy with automated planning, implementation, and code reviews by BeSImple / ai, tools, github
aube: New Node.js package manager from the creator of mise — drop-in replacement for npm with raw performance focus and security-conscious defaults by Jeff Dickey / node, tools
Agent Vault: Open-source HTTP credential proxy that prevents AI agents from handling sensitive API keys directly — transparently injects credentials at the network layer to block exfiltration and prompt injection by Infisical / ai, security, tools, github
Portless: Replace port numbers with named local URLs for local development — cleaner URLs for multi-service dev environments by Vercel Labs / tools, github
TSRX: TypeScript language extension described as a spiritual successor to JSX — compiles to React, Preact, Solid, and Ripple with control flow, scoped styles, and a live playground by Dominic Gannaway / typescript, javascript, tools
React Arborist 3.5: Tree view component for React with folders, drag and drop sorting, filtering, and keyboard navigation — the VS Code sidebar experience in one component by James Kerr / react, tools, github

🤪 Fun

Liquid Glass WebGL Effects for HTML: WebGL-powered glass refraction and blur effects for any HTML element — reproduce iOS 26's liquid glass aesthetic with an interactive playground by Yassine Bouanane / webgl, css, fun / 13 min read

📺 Videos

Designing The Next Flow State: Amelia Wattenberger on the missing abstraction in AI coding — a living spec, a markdown document between you and the agents as the shared source of truth by UX Tools / ai, design, engineering
The Future of MCP: MCP co-creator David Soria Parra on what agents look like in 2026 — connectivity as the key abstraction, skills and MCP as the integration stack, and why 2026 is the year agents go to production by David Soria Parra / ai, mcp, agents
Agentic Design: Video essay for designers building agent harnesses — how stochastic, scholastic, and deterministic design represent fundamentally different bets on design labor and who owns the final artifact by UX Tools / ai, design

Want to read more? Check out the full article here.

To sign up for the weekly newsletter, visit weeklyfoo.com.

Builders, Coasters, and 30 Plugins Gone Rogue

Adam — Fri, 24 Apr 2026 04:49:33 +0000

Gergely Orosz surveyed 900+ software engineers and found three archetypes emerging in the AI era: builders, shippers, and coasters. The breakdown of who's thriving, who's just shipping faster, and who's quietly coasting is the most grounded portrait of the profession right now.

The bigger question is what comes next. James Stanier maps three scenarios for whether the traditional mentorship pipeline can survive an AI-compressed learning curve — and the answer matters for everyone at every level of seniority. Addy Osmani formalizes the new skill that will separate high-functioning AI teams from the rest: Agentic Engine Optimization — structuring your docs for agent discoverability, parsability, and token efficiency. Tim Davis names the fundamental shift: we're no longer building deterministic systems — we're assembling probabilistic ones, from code no single human designed end-to-end, reviewed under time pressure. The accountability questions that creates haven't been answered yet.

Orhun Parmaksiz makes the principled case against vibe-coding: letting AI handle the tedious parts while you keep only the interesting ones isn't a productivity hack — it's how craft and maintainability erode together. The practical complement: TkDodo's vertical codebase — stop organizing your codebase by technical type and start organizing by domain, so everything that belongs together actually lives together.

Christian Ekrem demonstrates how branded types and discriminated unions let TypeScript carry proof that validation already happened — no defensive if-checks scattered across files, zero runtime overhead.

Security brought two head-turners this week: someone purchased 30 WordPress plugins specifically to plant backdoors across the portfolio (the acquisition vector most threat models never account for), and Cal.com closed its main repository after five years of open source because AI can now find and exploit public vulnerabilities faster than maintainers can patch them. The security calculus for open-source projects has quietly shifted.

Jonathan Chart's index deep dive covers composite index order, partial indexes, functional indexes, and EXPLAIN — the fundamentals that turn mysterious slow queries into five-minute fixes.

Finally: GitHub Stacked PRs just went native with one-click merging and AI agent integration — the pattern that was too painful to adopt at scale has officially graduated to mainstream. And Google's Magika is now open source: AI-powered file type detection for 200+ content types, the same model powering Google's own user safety infrastructure.

Enjoy!

Signup here for the newsletter to get the weekly digest right into your inbox.

Find the 12 highlighted links of weeklyfoo #133:

The Impact of AI on Software Engineers in 2026

by Gergely Orosz

Survey of 900+ engineers reveals three archetypes - builders, shippers, coasters - and how AI tools affect each differently in cost, productivity, and professional identity

🚀 Read it!, ai, engineering

Agentic Engine Optimization (AEO)

by Addy Osmani

How to structure docs so AI coding agents can actually use them - discoverability, parsability, token efficiency, and access control

📰 Good to know, ai, engineering

Who Will Be the Senior Engineers of 2035?

by James Stanier

Three scenarios for how AI will reshape the way senior engineers emerge - and whether the traditional mentorship pipeline can survive

📰 Good to know, engineering, ai

Write Less Code, Be More Responsible

by Orhun Parmaksiz

Against the vibe-coding trend - why letting AI handle the tedious parts while you keep the fun ones is a recipe for quality and maintainability decline

📰 Good to know, ai, engineering

Parse, Don't Validate in TypeScript

by Christian Ekrem

Branded types and discriminated unions let TypeScript carry proof that validation already happened - no more defensive if checks scattered across files

📰 Good to know, typescript, javascript

30 WordPress Plugins Backdoored After Acquisition

by anchor.host

A buyer purchased 30 plugins of varying popularity and planted a backdoor in all of them - a supply chain attack via legitimate ownership

📰 Good to know, security

Things You Didn't Know About Indexes

by Jonathan Chart

Composite index order, partial indexes, functional indexes, and EXPLAIN - a practical deep dive on where indexes help and where they hurt

📰 Good to know, database, engineering

Cal.com Goes Closed Source After 5 Years

by Cal.com

AI can now rapidly find and exploit open-source vulnerabilities - Cal.com closes its main repo while releasing Cal.diy under MIT for hobbyists

📰 Good to know, open-source, engineering

The Vertical Codebase

by TkDodo

Stop splitting code by technical type (components, hooks, utils) - group by functionality instead so everything in a domain lives together regardless of type

📰 Good to know, engineering

Probabilistic Engineering and the 24-7 Employee

by Tim Davis

The shift from deterministic to probabilistic engineering - code generated by stochastic systems, reviewed under time pressure, assembled into wholes no single human ever designed end-to-end

📰 Good to know, engineering, ai

GitHub Stacked PRs

by GitHub

Native stacked PR workflow - break large changes into small reviewable pull requests and merge the whole stack in one click, with AI agent integration

🧰 Tools, git, github, tools

Magika

by Google

Google's AI-powered file type detection that identifies 200+ content types using a lightweight deep learning model - used internally for user safety

🧰 Tools, ai, tools, github

Want to read more? Check out the full article here.

To sign up for the weekly newsletter, visit weeklyfoo.com.

Stay ahead in web development: latest news, tools, and insights #133

Adam — Mon, 20 Apr 2026 04:57:10 +0000

Signup here for the newsletter to get the weekly digest right into your inbox.

weeklyfoo #133 is here: your weekly digest of all webdev news you need to know! This time you'll find 36 valuable links in 4 categories! Enjoy!

🚀 Read it!

The Impact of AI on Software Engineers in 2026: Survey of 900+ engineers reveals three archetypes - builders, shippers, coasters - and how AI tools affect each differently in cost, productivity, and professional identity by Gergely Orosz / ai, engineering / 16 min read

📰 Good to know

Agentic Engine Optimization (AEO): How to structure docs so AI coding agents can actually use them - discoverability, parsability, token efficiency, and access control by Addy Osmani / ai, engineering / 16 min read
How Complex Is My Code?: Code complexity beyond algorithmic efficiency - human cognitive effort, psycholinguistic insights, and why metrics are better for prioritizing refactoring than measuring quality by Phil / engineering / 13 min read
Why MDN Was Rebuilt Without React: React was just a wrapper around static content - MDN now uses a custom render pipeline and Lit web components by MDN / frontend, engineering / 33 min read
Moving Railway's Frontend Off Next.js: Railway migrated 200+ routes from Next.js to Vite and TanStack Router with zero downtime - faster builds, better type safety, and a simpler client-side architecture by Railway / frontend, engineering / 7 min read
Uber's Agentic System for Instant Design Specs: How Uber built an agentic pipeline to automate design spec generation in minutes, cutting manual handoff work between designers and engineers by Ian Guisard / ai, design / 13 min read
Who Will Be the Senior Engineers of 2035?: Three scenarios for how AI will reshape the way senior engineers emerge - and whether the traditional mentorship pipeline can survive by James Stanier / engineering, ai / 16 min read
Write Less Code, Be More Responsible: Against the vibe-coding trend - why letting AI handle the tedious parts while you keep the fun ones is a recipe for quality and maintainability decline by Orhun Parmaksiz / ai, engineering / 6 min read
The Hidden Cost of Building Workflow Editors: Build vs buy analysis for React Flow-based editors - custom node state, undo/redo, and real-time sync pile up faster than teams expect by workflowbuilder.io / engineering, frontend / 10 min read
TanStack Start Gets RSC But Not Like Next.js: Client-owned tree, server functions via createServerFn only, and RSC as a plain React Flight stream you fetch with TanStack Query - a genuinely different approach than Next.js by TanStack / frontend, engineering / 19 min read
Parse, Don't Validate in TypeScript: Branded types and discriminated unions let TypeScript carry proof that validation already happened - no more defensive if checks scattered across files by Christian Ekrem / typescript, javascript / 13 min read
SVG Filters Guide: Getting Started with the Basics: Filter regions, color interpolation, and cross-browser consistency - a solid primer on a power feature most devs skip by Ana Tudor / css, frontend / 18 min read
Squash and Stretch on the Web: Disney animation principles applied to web motion - how squash and stretch adds polish and physicality to UI interactions by Josh W. Comeau / css, frontend / 15 min read
Why AI Sucks at Front End: AI design is trained on ancient solutions and unoriginal UI patterns - Adam Argyle's rant on why AI still can't do frontend well by Adam Argyle / ai, frontend / 14 min read
30 WordPress Plugins Backdoored After Acquisition: A buyer purchased 30 plugins of varying popularity and planted a backdoor in all of them - a supply chain attack via legitimate ownership by anchor.host / security / 10 min read
Things You Didn't Know About Indexes: Composite index order, partial indexes, functional indexes, and EXPLAIN - a practical deep dive on where indexes help and where they hurt by Jonathan Chart / database, engineering / 12 min read
Cal.com Goes Closed Source After 5 Years: AI can now rapidly find and exploit open-source vulnerabilities - Cal.com closes its main repo while releasing Cal.diy under MIT for hobbyists by Cal.com / open-source, engineering / 19 min read
What We Learned Building a Rust Runtime for TypeScript: Encore's experience replacing the Node.js runtime with a custom Rust-based one for TypeScript - tradeoffs, surprises, and lessons by Ivan Cernja / engineering, typescript / 18 min read
The Vertical Codebase: Stop splitting code by technical type (components, hooks, utils) - group by functionality instead so everything in a domain lives together regardless of type by TkDodo / engineering / 11 min read
Probabilistic Engineering and the 24-7 Employee: The shift from deterministic to probabilistic engineering - code generated by stochastic systems, reviewed under time pressure, assembled into wholes no single human ever designed end-to-end by Tim Davis / engineering, ai / 22 min read
Design and Engineering, as One: Most teams still use an 1898 industrial model that separates design from engineering - a case for deeper integration beyond better handoff processes by Matthias Ott / design, engineering / 30 min read
Cloudflare Email Service: Now in public beta. Ready for your agents by Thomas Gauvin, Eric Falcão / emails, cloudflare / 13 min read

🧰 Tools

Multica: Open-source platform integrating AI coding agents into dev teams as autonomous teammates - they claim tasks, write code, and report progress via a Go-based local daemon by multica.ai / ai, tools, github
Letta Code: Memory-first coding harness for long-lived AI agents that persist across sessions and accumulate skills like a permanent teammate by docs.letta.com/letta-code / ai, tools, github
GitHub Stacked PRs: Native stacked PR workflow - break large changes into small reviewable pull requests and merge the whole stack in one click, with AI agent integration by GitHub / git, github, tools
Delphitools: 40+ simple online tools with no logins or tracking - SVG optimizer, Tailwind shade generator, background remover, palette generator, and more by Ruby Morgan Voigt / tools
Magika: Google's AI-powered file type detection that identifies 200+ content types using a lightweight deep learning model - used internally for user safety by Google / ai, tools, github
Pompelmi 1.0: Antivirus File Scanning from Node: ClamAV wrapper for scanning malicious uploads in Node apps - standalone, no daemon required, spawns ClamAV on-demand by pompelmi contributors / javascript, tools, security
Micro-ML: Forecasting and Clustering for Node: 56KB WASM-powered library with regression, smoothing, clustering, and classification - no TensorFlow.js needed by Adam Perliński / javascript, tools, github
Hyperframes: Open-source HTML-based video rendering framework with first-class AI agent support - non-interactive CLI for agent-driven workflows, renders to MP4 locally or in Docker by HeyGen / tools, ai, github
Typescale AI: Typography scale and design token generator - creates consistent text hierarchies from H1 to H6 and body text with design token output by typescale.ai / design, tools
Animata: ~200 animation-focused React components built with Tailwind CSS - animated beams, spreading cards, glitching text, Slack-style hero screen by Codse / react, tools
Marky: A lightweight easy to use markdown viewer by Garrett Graves / markdown

🤪 Fun

100 Jumps: Jump without missing - a small physics-based browser game by boredzebra.com / fun, game / 1 min read
Chess in Pure SQL: A fully playable chess game rendered and run entirely with SELECT and UPDATE statements - no app server, no JavaScript, just SQL by dbpro.app / fun, sql / 5 min read
Slitscanner: Upload an image, set scan settings, move it around - creates slit-scan photography distortion art, neat for background images by Matt Soria / fun, art / 3 min read

Want to read more? Check out the full article here.

To sign up for the weekly newsletter, visit weeklyfoo.com.

Vibe Wars, Leaked Prompts, npm Under Siege

Adam — Fri, 17 Apr 2026 04:44:28 +0000

Martin Fowler and Kent Beck have joined forces to examine the AI disruption — and they're in agreement: this one is categorically different from Agile and OOP in magnitude and speed. When the two architects who helped define modern software engineering land on the same page, you slow down and read it.

The vibe coding debate found its sharpest edges this week. DHH has abandoned keyboard-first coding for a dual-AI tmux setup and barely writes code by hand anymore — with no quality regression reported. Bram Cohen has some thoughts: fully delegating code to AI without reading the output isn't a philosophy, it's a technical debt subscription with deferred payment terms. Both positions are defensible — which is the most unsettling thing about it.

Drew Breunig dissected the accidental Claude Code source leak to map exactly which context components are always vs conditionally included in its system prompt — the most revealing look at professional context engineering from inside a major AI tool that anyone outside Anthropic has published. Martin Fowler wrote the practical companion piece: how to encode your team's conventions into CLAUDE.md, linters, and CI so agents generate code that already passes review without constant correction.

Also in this issue: Konrad Piechowski's five git commands that map a codebase's health, bug clusters, and shipping confidence before you open a single file — an immediately adoptable pre-reading ritual.

Security this week deserves two reads: Dani Akash makes the case for a single package manager config change that blocks fast-moving supply chain attacks before they propagate, and socket.dev documents active social engineering campaigns specifically targeting high-value npm maintainers. Every package publisher should know these techniques before becoming the next target.

Tools this week: dryrun feeds AI agents a Postgres schema snapshot instead of a live database connection, Caveman strips filler words to cut Claude Code token usage by 65%, Boneyard auto-generates pixel-perfect skeleton screens straight from your DOM, and Little Snitch finally comes to Linux.

Enjoy!

Signup here for the newsletter to get the weekly digest right into your inbox.

Find the 12 highlighted links of weeklyfoo #132:

DHH's New Way of Writing Code

by Gergely Orosz

DHH switched from typing all his code to running two AI models in tmux — now he barely writes any code by hand, while his quality standards haven't budged

🚀 Read it!, ai, engineering

Cycles of Disruption in the Tech Industry

by Gergely Orosz

Martin Fowler and Kent Beck compare the AI shift to Agile and OOP — and explain why this time it's different in magnitude and speed

📰 Good to know, ai, engineering

Minimum Release Age is an Underrated Supply Chain Defense

by Dani Akash

A single package manager config change that can block fast-moving supply chain attacks before they reach your project

📰 Good to know, security, npm

The Cult of Vibe Coding Is Insane

by Bram Cohen

Fully delegating code to AI without reading the output is not a development philosophy, it's a debt factory — AI is only effective when humans actively review and guide it

📰 Good to know, ai, engineering

Encoding Team Standards

by Martin Fowler

Practical patterns for putting your team's conventions into CLAUDE.md, linters, and CI so AI agents generate code that passes review without constant correction

📰 Good to know, ai, engineering

How Claude Code Builds a System Prompt

by Drew Breunig

The accidental source code leak reveals how Claude Code assembles its context — some components always included, others conditional — showing just how complex context engineering has become

📰 Good to know, ai, engineering

The Git Commands I Run Before Reading Any Code

by Konrad Piechowski

Five git commands that reveal a codebase's story before you open a single file — who built it, where bugs cluster, whether a team ships with confidence or tiptoes around landmines

📰 Good to know, git, engineering

Attackers Hunting High-Impact Node.js Maintainers

by socket.dev

Ongoing social engineering campaigns targeting npm package maintainers — know these techniques before you become a target

📰 Good to know, security, nodejs

dryrun

by Radim Marek

Offline-first Postgres MCP server — lets AI agents access what they need from your database using a JSON snapshot, never a live connection

🧰 Tools, postgres, mcp, ai

Caveman

by Julius Brussee

Claude Code skill and Codex plugin that compresses LLM communication by stripping filler words while maintaining technical accuracy — cuts token usage by an average of 65%

🧰 Tools, ai, claude, tools

Little Snitch for Linux

by Objective Development

The beloved macOS network monitor finally comes to Linux — see every hidden app network connection, block unwanted traffic, manage blocklists, write custom rules, and view detailed traffic history

🧰 Tools, linux, security

Boneyard

by 0xGF

Snapshots your DOM and auto-generates pixel-perfect skeleton screens — no manual placeholders, supports React, Vue, Svelte, and Angular

🧰 Tools, react, ui

Want to read more? Check out the full article here.

To sign up for the weekly newsletter, visit weeklyfoo.com.

Stay ahead in web development: latest news, tools, and insights #132

Adam — Mon, 13 Apr 2026 04:40:30 +0000

Signup here for the newsletter to get the weekly digest right into your inbox.

weeklyfoo #132 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!

DHH's New Way of Writing Code: DHH switched from typing all his code to running two AI models in tmux — now he barely writes any code by hand, while his quality standards haven't budged by Gergely Orosz / ai, engineering / 11 min read

📰 Good to know

Cycles of Disruption in the Tech Industry: Martin Fowler and Kent Beck compare the AI shift to Agile and OOP — and explain why this time it's different in magnitude and speed by Gergely Orosz / ai, engineering / 13 min read
How to Use HTML Video and Audio Lazy-Loading: HTML video and audio lazy loading is now a web standard in Chrome 148, with Firefox and WebKit implementations close behind by Scott Jehl / html, performance / 12 min read
What to Know in JavaScript (2026 Edition): A comprehensive map of the JavaScript landscape today — ECMAScript 2025/2026 additions, frameworks, runtimes, and build tools by Chris Coyier / javascript / 27 min read
Claude Is Not Your Architect. Stop Letting It Pretend: AI agents are agreeable pattern-matchers that produce generic designs without understanding your real constraints — and nobody owns the result when things break by hollandtech.net / ai, engineering / 10 min read
Minimum Release Age is an Underrated Supply Chain Defense: A single package manager config change that can block fast-moving supply chain attacks before they reach your project by Dani Akash / security, npm / 11 min read
Good APIs Age Slowly: Robust APIs prioritize long-term stability by establishing explicit boundaries and avoiding exposure of internal details — the disciplined approach that turns your API into reliable infrastructure by Yusuf Aytas / engineering / 8 min read
What Is CSS Containment and How Can I Use It?: Harry Roberts explains CSS containment from first principles — a long-available but underused feature that lets browsers skip layout, paint, and style recalculations for isolated subtrees by Harry Roberts / css, performance / 20 min read
The Cult of Vibe Coding Is Insane: Fully delegating code to AI without reading the output is not a development philosophy, it's a debt factory — AI is only effective when humans actively review and guide it by Bram Cohen / ai, engineering / 7 min read
Encoding Team Standards: Practical patterns for putting your team's conventions into CLAUDE.md, linters, and CI so AI agents generate code that passes review without constant correction by Martin Fowler / ai, engineering / 15 min read
What Is Inference Engineering?: A new engineering discipline is emerging around deploying and optimizing LLMs at scale — Gergely maps what inference engineering actually involves and why teams need it by Gergely Orosz / ai, engineering / 40 min read
How Claude Code Builds a System Prompt: The accidental source code leak reveals how Claude Code assembles its context — some components always included, others conditional — showing just how complex context engineering has become by Drew Breunig / ai, engineering / 11 min read
The Git Commands I Run Before Reading Any Code: Five git commands that reveal a codebase's story before you open a single file — who built it, where bugs cluster, whether a team ships with confidence or tiptoes around landmines by Konrad Piechowski / git, engineering / 5 min read
State of Prototyping: Spring 2026 Report: 1,478 designers surveyed — Claude is now the #2 weekly design tool after Figma, 43.8% spend more than half their time vibe coding, and heavy vibe coders rate job satisfaction 7.39 vs 5.93 for non-vibe coders by Tommy (UX Tools) / ai, design / 13 min read
CSS Subgrid Is Super Good: A practical guide to using CSS subgrid (available across browsers since 2023) for laying out content from CMS systems you don't fully control by David Bushell / css / 6 min read
CSS position:sticky Can Now Track the Nearest Scroll Container per Axis: A nine-year CSS spec issue is resolved — sticky elements can now track different scroll containers horizontally and vertically, enabling previously impossible layouts, landing in Chromium first by Bramus Van Damme / css / 10 min read
light-dark() Is About to Support Images: The CSS light-dark() function is expanding beyond color values to handle images too — swap light and dark mode backgrounds with a single declaration, landing in Chromium and Firefox soon by Bramus Van Damme / css / 10 min read
Attackers Hunting High-Impact Node.js Maintainers: Ongoing social engineering campaigns targeting npm package maintainers — know these techniques before you become a target by socket.dev / security, nodejs / 12 min read

🧰 Tools

view-transitions-toolkit: Utility functions for working with View Transitions — patterns, helpers, and demos collected from Bramus's years of experimentation by Bramus Van Damme (Google) / css, animation
dryrun: Offline-first Postgres MCP server — lets AI agents access what they need from your database using a JSON snapshot, never a live connection by Radim Marek / postgres, mcp, ai
rpg: A psql-compatible Postgres terminal written in Rust with built-in DBA diagnostics and an AI assistant by Nik Samokhvalov / postgres, rust, cli
Fuse.js 7.3: Lightweight fuzzy-search library — v7.3 adds per-term fuzzy matching and static single-string matching, zero backend required by Kiro Risk / javascript, search
Xilem: Experimental high-performance reactive UI framework for Rust inspired by React and SwiftUI by Linebender / rust, ui
Hippo Memory: Biologically-inspired memory system for AI agents that prevents context loss through memory decay and consolidation — hybrid search and multi-tool management for more adaptive agents by Keith So / ai, memory
Caveman: Claude Code skill and Codex plugin that compresses LLM communication by stripping filler words while maintaining technical accuracy — cuts token usage by an average of 65% by Julius Brussee / ai, claude, tools
Gemma Gem: Chrome extension that runs Google's Gemma 4 model entirely on-device via WebGPU — no API keys, no cloud, no data leaving your machine by Yaniv Kessler / ai, chrome
human.json: Machine-readable file for documenting the humans behind a website — a structured JSON alternative to the classic humans.txt by Beto Dealmeida / tools
Little Snitch for Linux: The beloved macOS network monitor finally comes to Linux — see every hidden app network connection, block unwanted traffic, manage blocklists, write custom rules, and view detailed traffic history by Objective Development / linux, security
Skrun: Open-source platform for deploying Agent Skills as callable API endpoints — multi-model support, stateful memory, and tool invocation via local scripts or MCP servers by skrun-dev / ai, agents, mcp
TinyTTS: Fast English text-to-speech for Node.js and Python with a 3.4MB model — CPU-only, no cloud dependency by tronghieuit / tts, ai, node
Boneyard: Snapshots your DOM and auto-generates pixel-perfect skeleton screens — no manual placeholders, supports React, Vue, Svelte, and Angular by 0xGF / react, ui
dev-browser: A Claude Skill to give your agent the ability to use a web browser by Sawyer Hood / ai, claude, browsers
Agent Skills: Production-grade engineering skills for AI coding agents. by Addy Osmani / ai, claude, skills
Awesome DESIGN.md: A collection of DESIGN.md files inspired by popular brand design systems. Drop one into your project and let coding agents generate a matching UI. by VoltAgent / ai, design

🤪 Fun

CSS or BS: Test your CSS knowledge by guessing whether each property name is real or completely made up — trickier than it looks by Keith Cirkel / css / 8 min read

📺 Videos

React Paris 2026 Talks: 23 talks from React Paris 2026 now on YouTube by React Paris / react, conference

Want to read more? Check out the full article here.

To sign up for the weekly newsletter, visit weeklyfoo.com.

Agents in a Box, npm on Fire

Adam — Fri, 10 Apr 2026 05:03:06 +0000

This week's theme is basically: AI agents are eating the world, and the world is starting to eat back.

Daniel Miessler drops his April AI synthesis — the quarterly anchor you need when the field is moving faster than your git pull. Joel Andrews offers the honest counterpoint with some uncomfortable truths about AI coding agents in real production environments. And Drew Breunig maps the shift from agentic experiments to spec-driven idiosyncratic tooling — the clearest mental model for where we go from here.

Meanwhile, Haseeb Qureshi dug through the full 1,900-file TypeScript leak of the Claude Code source to reverse-engineer how it actually works — the most revealing accidental transparency from any AI lab this year. The security side of the week is busy: axios versions on npm were quietly installing remote access trojans (audit your dependencies, now), and the ecosystem responded with two serious containment tools — Stanford's jai brings kernel-level jailing for AI agents, while nono adds cryptographic immutable audit chains and atomic rollback. If you're running agents in production, both are worth a look.

Also in this issue: 113 hard-won engineering lessons from Jack Kinsella on scaling a monolith to a million lines of code, pgmicro reimplements PostgreSQL semantics on top of SQLite storage in-process (technically audacious), and git-lrc gives you free AI code reviews on every commit with zero friction.

Enjoy!

Signup here for the newsletter to get the weekly digest right into your inbox.

Find the 10 highlighted links of weeklyfoo #131:

Inside the Claude Code source

by Haseeb Qureshi

Anthropic's Claude Code CLI source code leaked onto GitHub recently. All of it. About 1,900 files and a lot of TypeScript.

🚀 Read it!, claude

The Most Important Ideas in AI Right Now (April 2026)

by Daniel Miessler

Self-improvement and transparency change everything in unexpected ways

📰 Good to know, ai, engineering

Some uncomfortable truths about AI coding agents

by Joel Andrews

I’ve been following the development of generative AI closely for several years now.

📰 Good to know, ai

Scaling a Monolith to 1M LOC

by Jack Kinsella

113 Pragmatic Lessons from Tech Lead to CTO

📰 Good to know, engineering

axios Compromised on npm

by Ashish Kurmi

Malicious Versions Drop Remote Access Trojan

📰 Good to know, security, axios

The 2nd Phase of Agentic Development

by Drew Breunig

Yesterday we talked about how cheap code is fueling an era of idiosyncratic tooling, and previously we’ve talked about the rise of spec driven development.

📰 Good to know, ai, engineering

pgmicro

by Glauber Costa

An in-process reimplementation of PostgreSQL, backed by a SQLite-compatible storage engine

🧰 Tools, postgres, sqlite

jai

by stanford-scs

Jail your AI agent

🧰 Tools, ai

nono

by nono.sh

Kernel-enforced agent sandbox. Capability-based isolation with secure key management, atomic rollback, cryptographic immutable audit chain of provenance. Run your agents in a zero-trust environment.

🧰 Tools, ai, sandbox

git-lrc

by hexmos.com

Free, Unlimited AI Code Reviews That Run on Commit

🧰 Tools, ai, git, reviews

Want to read more? Check out the full article here.

To sign up for the weekly newsletter, visit weeklyfoo.com.

AI coding agents nail the code. They break on the scaffolding.

Adam — Mon, 06 Apr 2026 05:28:16 +0000

Signup here for the newsletter to get the weekly digest right into your inbox.

When AI coding agents fall short, the failure is almost never "it generated bad code." The failure is everything else — the environment state, the implicit decisions baked into a config three years ago, the reason that dependency is pinned to that version.

Joel Andrews published one of the more grounded take-downs of AI coding agents this week. Not a hype piece in either direction — just a practitioner's account of where these tools consistently fall apart in real production environments. The pattern he describes maps to something Daniel Miessler flagged in his April AI synthesis: most of what developers call "work" was actually maintaining an elaborate, fragile state required for work to happen. The AI exposed it by stumbling directly into it.

This is what "scaffolding" means in practice. Your agent can write a correct function. It cannot know why your test suite is configured with that specific mock, or why a particular environment variable gets overridden in staging but not local. That context lives in heads and in Slack threads and in the institutional memory of whoever set it up in 2022. The model doesn't have it.

Drew Breunig's framing of the second phase of agentic development makes more sense through this lens. Phase one was: can the AI generate code? (Yes, mostly.) Phase two is: can the agent operate within the scaffolding well enough to do something useful end-to-end? The answer right now is: inconsistently, and the variance is high. Spec-driven development isn't just a prompt engineering trick — it's an attempt to externalize the scaffolding into something the agent can read.

What's encouraging is that we're getting honest data now. The honeymoon period where everything was demos and benchmarks is ending. Practitioners are writing up where it works and where it doesn't, and the failure modes are getting specific enough to be actionable.

Some uncomfortable truths about AI coding agents — grounded production observations from Joel Andrews on where agents consistently fall short, with specifics that let you actually plan around them

The 2nd Phase of Agentic Development — Drew Breunig's model for the phase shift from "can it code?" to spec-driven idiosyncratic tooling

The Most Important Ideas in AI Right Now (April 2026) — Daniel Miessler's quarterly synthesis, worth reading for the self-improvement and transparency threads that contextualize where the capability ceiling actually sits

Inside the Claude Code source — Haseeb Qureshi's reverse-engineering of the Claude Code CLI from the full TypeScript leak; seeing the internals clarifies how the scaffolding problem is being handled at the tool level

The next useful thing the AI tooling community could build isn't a better code generator. It's better scaffolding capture — structured context files, agent instructions, decision logs that encode why the codebase is the way it is. The teams building that layer now will get substantially more out of agents than teams that keep treating them as fast autocomplete.

Want to read more? Check out the full issue here.

To sign up for the weekly newsletter, visit weeklyfoo.com.

TypeScript 6 Ships, Agents Remember, IDEs Panic Quietly

Adam — Sat, 04 Apr 2026 11:11:10 +0000

TypeScript 6.0 is officially out — native ES module output by default, sweeping type system upgrades, and a must-read before it starts appearing in your dependency trees.

The big AI debate this week: Daniel Miessler builds the most complete argument yet for why AI will replace knowledge work — and somehow makes it sound like good news. Meanwhile, Addy Osmani makes the case that agentic orchestration is already eating the IDE's lunch — if your workflow still centers on a code window, this might be the signal you've been waiting for.

On the agent tooling front: Simon Willison breaks down Claude Code's new auto mode, which removes permission friction and quietly redefines what "agentic" means in daily practice. Mozilla dropped an open standard for shared agent learning called cq — so agents can stop rediscovering the same failures in parallel. And Cog bakes persistent memory and self-reflection directly into Claude Code projects, because amnesia is a terrible trait in a coding assistant.

Also this week: James Garbutt and the e18e community pinpoint exactly three architectural patterns responsible for the bulk of JavaScript bloat — a surgical diagnosis that finally makes the fix feel actionable. In the counterintuitive result of the week, someone rewrote their Rust WASM parser in TypeScript and it ran 3x faster — the right algorithm, as always, beats the right language. Storybook gains an MCP server so AI agents can browse and generate components directly. Stripe Projects quietly lays the infrastructure for the agentic economy. And Sugar High is a syntax highlighter under 1KB gzipped — the kind of tool that does exactly one thing and nothing else.

For lighter reading: Andrew Nesbitt assembled the top 10 open source conspiracies with just enough factual backup to make you genuinely uncomfortable. And fifteen engineering leaders pool their real-world AI workflows into one essential guide.

Enjoy!

Signup here for the newsletter to get the weekly digest right into your inbox.

Find the 13 highlighted links of weeklyfoo #130:

How to Do AI-Assisted Engineering

by Gregor Ojstersek

15 experienced engineers and engineering leaders share their real-world experiences with AI-assisted engineering.

🚀 Read it!, ai, engineering

Exactly Why and How AI Will Replace Knowledge Work

by Daniel Miessler

And why this is actually a good thing

📰 Good to know, ai, engineering

The Three Pillars of JavaScript Bloat

by James Garbutt

Over the last couple of years, we’ve seen significant growth of the e18e community and a rise in performance focused contributions because of it.

📰 Good to know, javascript, bloat

Is the IDE dead?

by Addy Osmani

How Agent orchestration is replacing the editor as the center of developer work

📰 Good to know, ai, ide

Announcing TypeScript 6.0

by Daniel Rosenwasser

Today we are excited to announce the availability of TypeScript 6.0!

📰 Good to know, typescript

Rewriting our Rust WASM Parser in TypeScript

by Thesys Engineering Team

We rewrote our Rust WASM Parser in TypeScript - and it got 3x Faster

📰 Good to know, wasm, typescript

Auto mode for Claude Code.

by Simon Willison

Really interesting new development in Claude Code today as an alternative to --dangerously-skip-permissions

📰 Good to know, claude, ai

The Top 10 Biggest Conspiracies in Open Source

by Andrew Nesbitt

This is gold!

📰 Good to know, oss

Storybook MCP for React

by Kyle Gach

Storybook-powered agentic UI development

📰 Good to know, storybook, ai, mcp

cq

by Mozilla

An open standard for shared agent learning. Agents persist, share, and query collective knowledge so they stop rediscovering the same failures independently.

🧰 Tools, ai, memory

Sugar High

by Jiachi Liu

Super lightweight code syntax highlighter, around 1KB after minified and gzipped

🧰 Tools, syntax, highlight

Stripe Projects

by Stripe

Stripe Projects lets you or your agents provision multiple services, generate and store credentials, and manage usage and billing from the CLI. Set up hosting, databases, auth, AI, analytics, and more in a few commands

🧰 Tools, stripe, ai, infra

Cog

by Marcio Puga

Cognitive architecture for Claude Code — persistent memory, self-reflection, and foresight

🧰 Tools, ai, memory, claude

Want to read more? Check out the full article here.

To sign up for the weekly newsletter, visit weeklyfoo.com.

Stay ahead in web development: latest news, tools, and insights #130

Adam — Mon, 30 Mar 2026 04:55:17 +0000

Signup here for the newsletter to get the weekly digest right into your inbox.

weeklyfoo #130 is here: your weekly digest of all webdev news you need to know! This time you'll find 39 valuable links in 5 categories! Enjoy!

🚀 Read it!

How to Do AI-Assisted Engineering: 15 experienced engineers and engineering leaders share their real-world experiences with AI-assisted engineering. by Gregor Ojstersek / ai, engineering / 41 min read

📰 Good to know

Why your next mobile app is probably headless: For years, companies fought to get their icon on your phone. by Tuan-Anh Tran / engineering / 5 min read
Exactly Why and How AI Will Replace Knowledge Work: And why this is actually a good thing by Daniel Miessler / ai, engineering / 39 min read
The Three Pillars of JavaScript Bloat: Over the last couple of years, we’ve seen significant growth of the e18e community and a rise in performance focused contributions because of it. by James Garbutt / javascript, bloat / 14 min read
How I'm Productive with Claude Code: Automating the grunt work, Killing the wait, Let Claude see what happens by Neil Kakkar / claude / 7 min read
Is the IDE dead?: How Agent orchestration is replacing the editor as the center of developer work by Addy Osmani / ai, ide / 9 min read
These coders want AI to take their jobs: Vibe coding and what it means for the future of programming, explained. by Ariana Aspuru, Sean Rameswaram / ai, engineering / 13 min read
Announcing TypeScript 6.0: Today we are excited to announce the availability of TypeScript 6.0! by Daniel Rosenwasser / typescript / 36 min read
Rewriting our Rust WASM Parser in TypeScript: We rewrote our Rust WASM Parser in TypeScript - and it got 3x Faster by Thesys Engineering Team / wasm, typescript / 8 min read
Auto mode for Claude Code.: Really interesting new development in Claude Code today as an alternative to --dangerously-skip-permissions by Simon Willison / claude, ai / 5 min read
The Great CSS Expansion: CSS now does what Floating UI, GSAP ScrollTrigger, Framer Motion, and react-select used to require JavaScript for. by Pavel Laptev / css / 22 min read
Sneaky Header Blocker Trick: One of the coolest transitions on this blog. by Josh Comeau / css / 9 min read
SVG favicons that respect theme preference: Style your SVGs! by Paweł Grzybek / svg / 4 min read
The Top 10 Biggest Conspiracies in Open Source: This is gold! by Andrew Nesbitt / oss / 15 min read
Storybook MCP for React: Storybook-powered agentic UI development by Kyle Gach / storybook, ai, mcp / 4 min read

🧰 Tools

Arrow.js: The first UI framework for the agentic era — tiny, performant, with WASM sandboxes for safe code execution. by arrow-js.com / ui
Shadify: Stream AI generated Shadcn components, not just text. by shadify.copilotkit.ai / ai, ui
Claude Code Cheat Sheet: Summarized on one page by storyfox.cz / claude
cq: An open standard for shared agent learning. Agents persist, share, and query collective knowledge so they stop rediscovering the same failures independently. by Mozilla / ai, memory
OpenUI: The Open Standard for Generative UI by openui.com / ui, ai
Sugar High: Super lightweight code syntax highlighter, around 1KB after minified and gzipped by Jiachi Liu / syntax, highlight
Gridland: Terminal apps that run anywhere - even the browser! by Thoughtful / ui
Email.md: Render markdown into email-safe HTML by emailmd.dev / emails, markdown
pgtui: Postgres TUI client that handles basic CRUD operations on databases. by Kris Warner / postgres, cli, tui
docmd: Generate minimal, fast & beautiful docs from Markdown. No React, no bloat, just content. Built for developers. by docmd.io / docs
Tooscut: A powerful NLE editor with GPU compositing, keyframe animation, and real-time preview. No installs required. by Mohamad Mohebifar / video, tools
Elena: Elena is a simple, tiny library for building Progressive Web Components. by elenajs.com / pwa
Gea: A batteries-included, reactive JavaScript UI framework. No virtual DOM. Compile-time JSX transforms. Proxy-based stores. Surgical DOM patching. by Armagan Amcalar / javascript, framework
Optio: Workflow orchestration for AI coding agents, from task to merged PR. by Jon Wiggins / ai, orchestration
visual-json: The Visual JSON Editor. Schema-aware, embeddable, extensible. by Vercel / json, viewer
knex.js: A query builder for PostgreSQL, MySQL, CockroachDB, SQL Server, SQLite3 and Oracle, designed to be flexible, portable, and fun to use. by knexjs.org / sql
Stripe Projects: Stripe Projects lets you or your agents provision multiple services, generate and store credentials, and manage usage and billing from the CLI. Set up hosting, databases, auth, AI, analytics, and more in a few commands by Stripe / stripe, ai, infra
Cog: Cognitive architecture for Claude Code — persistent memory, self-reflection, and foresight by Marcio Puga / ai, memory, claude
React Joyride: Create guided tours in your apps by Gil Barbara / react, tours
Semiotic: A data visualization for AI and Streaming by semiotic.nteract.io / visualization
Stroke: Create Hand-drawn SVG animations for logos, signatures, and illustrations. by Abhishek / drawings, animation

🤪 Fun

Web Rewind: An interactive journey through 30 years of the web by Opera / web, rewind / 1 min read
Github Space Shooter: Visualizes GitHub contribution graphs as Space Shooter! by Zane Chen / shooter, gif / 1 min read

📺 Videos

How to burn $30m on a JavaScript framework...: The famo.us rendering engine was supposed to change web dev forever back in 2012. by Fireship / javascript, famo

Want to read more? Check out the full article here.

To sign up for the weekly newsletter, visit weeklyfoo.com.