Temporal Lands, Agentic DevTools Rise, and Compliance Gets Real
NEWSLETTER | Amplifi Labs
Ditch Moment.js: Practical Recipes for the JS Temporal API
Smashing Magazine • March 13, 2026
A hands-on guide shows how to migrate Moment.js code to the ECMAScript Temporal API, covering strict ISO/RFC 9557 parsing, immutable operations, time zones (ZonedDateTime), durations, and Intl-based locale-aware formatting. It explains how to choose the right Temporal types (Instant, PlainDate/Time, ZonedDateTime), the trade-offs versus Moment’s mutability and token formatting, and bundle-size wins. Temporal is shipping in Chrome/Edge/Firefox; Safari still needs the @js-temporal/polyfill (alpha, ~44 kB gzipped) or a TP flag, so plan rollout accordingly.
Engineering Productivity & DevTools
Shipping Production Code with LLMs: A Multi-Model Agent Workflow
Around the web •March 15, 2026
A developer shares a practical, low-defect workflow that treats LLMs as an architect–developer–reviewer team: Opus 4.6 plans, Sonnet 4.6 implements, and independent reviewers (Codex 5.4, Gemini 3 Flash, etc.) critique diffs for quality. Using the OpenCode harness, he demonstrates adding an email channel to a production agent via Cloudflare Worker webhooks, robust MIME parsing (mailparser), SMTP sending (nodemailer), config-gated features, and security-minded allowlists with domain wildcards that avoid spoofing edge cases. The takeaway: keep humans on architecture and tradeoffs, combine multiple models for checks and balances, and you can scale to tens of thousands of lines with fewer regressions.
Chrome DevTools MCP auto-connects coding agents to live sessions
Around the web •March 15, 2026
Chrome DevTools MCP now lets coding agents attach to an active Chrome session, reusing logged-in context and inspecting currently selected Elements or Network requests for AI-assisted debugging. The feature relies on Chrome M144 (beta) remote debugging—developers must enable it at chrome://inspect/#remote-debugging, after which --autoConnect requests a per-session user-approved connection with visible automation indicators. This closes the gap between manual and agent-driven workflows and sets the stage for broader panel data access in future updates.
Go 1.26 ships source-level inliner powering self-service refactors
Around the web •March 11, 2026
Go 1.26 introduces an all-new go fix with a source-level inliner and a //go:fix inline directive so package authors can encode safe, automated API migrations in code. The inliner—also used by gopls—can rewrite calls, types, and constants (e.g., migrating ioutil.ReadFile to os.ReadFile) while handling side effects, constant-folding hazards, shadowing, unused variables, and imports. This enables scalable, low-risk deprecation cleanup and modernization across Go codebases, though some results may still warrant manual tidying.
Frontend UX & Web Platform
Inside a 49MB News Page: Programmatic Ads vs. Web Vitals
Around the web •March 15, 2026
A teardown of a major news article shows 422 requests and 49MB transferred, driven by in-browser programmatic ad auctions, megabytes of JS, constant tracking beacons, and hostile UX patterns (double modals, sticky autoplay video, high CLS). For developers, it’s a case study in how ad-tech bloat throttles CPUs, drains batteries, and degrades Core Web Vitals/SEO—even as ad platforms enable it. The piece offers pragmatic fixes: defer/serialize popups until 60s or 50% scroll, reserve space for async content (aspect-ratio/min-height skeletons), lazy-load and refresh ads by scroll depth, pause videos with IntersectionObserver, and collapse unfilled slots via ResizeObserver.
CSS corner-shape Brings Squircles, Bevels, Scoops to Production UIs
Smashing Magazine •March 12, 2026
Chrome 139+ ships the CSS corner-shape property, which modifies border-radius to create squircles, bevels, scoops, notches, and more, with per-corner control and a superellipse() function for fine tuning. Unlike clip-path/SVG hacks, it also shapes outlines, borders, box shadows, and backgrounds and animates smoothly, cutting brittle UI code. Adopt via progressive enhancement using @supports and layered styles, with border-radius fallbacks until Firefox and Safari add support.
AI Systems, Risk & Governance
C-22 curbs warrantless lookups, mandates intercept capabilities and metadata retention
Around the web •March 15, 2026
Canada’s Bill C‑22 narrows warrantless access by replacing broad subscriber data demands with a limited “confirmation of service” query and judge‑approved production orders, though the low “reasonable grounds to suspect” threshold remains. Its SAAIA component compels telecoms and broader “electronic service providers” (including major platforms operating in Canada) to support surveillance capability testing, with designated core providers required to retain specified categories of metadata for up to one year (excluding content, web browsing history, and social media activity), keep requests secret, and operate under ministerial orders now subject to Intelligence Commissioner approval. For ISPs, cloud, and platform teams, this implies new engineering and compliance work on lawful‑intercept/testing interfaces, retention pipelines, and security reviews to avoid systemic vulnerabilities, with potential cross‑border exposure as the regime appears aligned with 2AP/CLOUD Act cooperation.
Raschka's LLM Architecture Gallery Maps Modern Open-Weight LLM Designs
Around the web •March 15, 2026
Sebastian Raschka released an updated, high-resolution gallery of LLM architecture diagrams and fact sheets from his comparison series, with clickable figures linked to each model section. Covering 2024–2026 open-weight releases, it surfaces current design patterns—sparse MoE with ~17–40B active paths (DeepSeek/GLM/Qwen), hybrid attention (MLA, Gated DeltaNet), widespread GQA+QK-Norm, sliding-window/global layers, Mamba-2 hybrids, and MTP. It’s a practical reference for model architects and infra teams comparing training/serving trade-offs (e.g., KV cache and latency) and invites community corrections; a poster version is available.
AI Is Scaling Bad UX Research Methods by Default
Nielsen Norman Group •March 13, 2026
UX research platforms increasingly embed methodological flaws—missing basics for quantitative testing (e.g., task randomization, multiple success URLs), analysis that ignores nonverbal behavior, and UIs that conflate interviews with usability tests. As vendors bolt on AI agents that plan, moderate, and analyze studies, these errors scale: auto-generated tasks are often leading or unrealistic, producing unreliable insights unless expert researchers review and control them. Teams should vet AI features on real studies, demand researcher oversight and guardrails, and avoid treating tool defaults as best practice.
