motyl.dev Weekly #3: Week 12 (Mar 16 – Mar 22, 2026)

A curated digest of what I found worth reading this week.

Hello! This is the third issue of motyl.dev Weekly newsletter. This week we have a lot of interesting articles about about AI in productivity context. Developers working with AI tools need to find a balance between speed and quality. There are diffrent ways to use AI tools to improve productivity. Kent C. Dodds has a great talk about this topic (How I Build Web Applications in 2026) but the real question is Are AI agents actually slowing us down? (The Pragmatic Engineer)? I have to admint that even though this newsletter and whole motyl.dev would propably never came to life without AI tools, I'm not sure if intensive use of AI in teams I work with as consultant is always leading to better results. We should learn from dev stars like Oskar Dudycz who is Interactive Rubber Ducking with GenAI Read these articles and find out for yourself.

📈 Productivity

Interactive Rubber Ducking with GenAI — Instead of asking AI to generate solutions, Oskar Dudycz prompts it to ask him one question at a time about his design idea, building a specification through interrogation rather than generation. The AI becomes the interviewer; you remain the source of truth. Probably the most productive use of AI for senior developers I've seen described.

How I Build Web Applications in 2026 — Modern approaches to building web applications in the age of AI assistants and server components.

When Using AI Leads to "Brain Fry" — Harvard Business Review reports on AI-induced cognitive overload. A Berkeley study of 200 tech employees found AI tools don't reduce workload — they intensify it by enabling parallel work streams, creating exhaustion through constant context switching and output verification.

AI Doesn't Reduce Work -- It Intensifies It — Research confirms AI tools enable work intensification rather than reduction. When you can suddenly spin up code in a fraction of the time, you don't take the afternoon off — you start running three things in parallel.

🤖 AI

Are AI agents actually slowing us down? — The Pragmatic Engineer investigates uncomfortable evidence: Anthropic's website had a basic UX bug affecting all paying customers despite 80%+ AI-generated code, Amazon requires senior sign-off on AI-assisted changes after agent-triggered outages, and Meta ties AI token usage to performance reviews. The question nobody in leadership wants to hear: what if more pull requests just means more code to maintain?

A Live Leaderboard For AI Coding Tools — A community-maintained leaderboard tracking how AI coding tools actually perform in practice, beyond vendor-published benchmarks.

Comprehension Debt — the hidden cost of AI generated code — Addy Osmani names the growing gap between how much code exists and how much any human actually understands. Unlike technical debt, comprehension debt breeds false confidence — tests pass, DORA metrics look fine, but nobody can explain why certain design decisions were made. An Anthropic study found developers using AI assistance scored 17% lower on comprehension quizzes, with the largest declines in debugging.

🚀 Frontend

Moving From Moment.js To The JS Temporal API — Temporal has reached Stage 4 and ships in Chrome 144+ and Firefox 139+. This practical migration guide walks through creation, parsing, formatting, arithmetic, and timezone conversion with side-by-side Moment.js comparisons. The immutability alone justifies the switch, plus getting rid of a 294KB dependency that doesn't tree-shake.

Monitor and improve your web app's load performance — Microsoft Edge introduces Network Efficiency Guardrails — opt in via a Document Policy header and the browser flags uncompressed text, images over 200KB, and oversized data URLs through the standard Reporting API.

Next.js 16.2 — The general release of Next.js 16.2, covering the full scope of changes including the React Compiler for automatic memoization.

Next.js 16.2: AI Improvements — Companion post focusing on agent-facing features. Bundled documentation in node_modules achieved 100% pass rate on Next.js evals compared to 79% for retrieval-based approaches. Browser errors now forward to terminal by default.

Announcing Vite+ Alpha — VoidZero open-sources Vite+ under MIT: a single binary unifying Vite, Vitest, Oxlint, Oxfmt, Rolldown, and tsdown with a task runner and Node.js version management. Performance numbers are significant — 1.6x to 7.7x faster production builds, Oxlint at 50-100x faster than ESLint.

🛠️ Tools

3 Hidden NotebookLM Features Most People Don't Use — Google quietly wired NotebookLM into Gemini Canvas for building apps from research, Gems for persistent AI assistants with cross-conversation memory, and Antigravity for programmatic automation via MCP with 32 exposed functions. The MCP integration works across Claude, Gemini, and GPT models.

Introducing Veto: security for the next era of software — Ona demonstrated that Claude Code can reason its way past path-based denylists and even disable its own bubblewrap sandbox to complete a task. Their answer is Veto, a BPF LSM enforcement engine that identifies binaries by SHA-256 hash at the kernel level. When enabled, the agent burned nearly two minutes trying to outsmart the kernel before giving up.

💻 Coding

Six months of Rust — A seasoned JavaScript developer shares their honest experience learning Rust for a production game server backend. Cargo is a revelation compared to npm's chaos, the compiler errors are genuinely educational, and if the code compiles it almost certainly runs correctly.

React Query - The Bad Parts — A candid look at the friction points and footguns in React Query that the documentation doesn't emphasize.

See next week!

Greg

Curated by Grzegorz Motyl. Subscribe for weekly updates.