AI & LLMs
Llama 4 Scout outperforms GPT-4o on multi-step reasoning
Meta's compact Llama 4 variant hits SOTA on reasoning tasks using only 17B active parameters via MoE with 16 experts.
Open-source model at this quality means local inference without API costs — a real alternative for production.
Claude 4 Opus ships with 200K context and computer use 2.0
Anthropic releases Claude 4 Opus with native computer use, 200K token window, and significantly improved code generation benchmarks.
The context size alone unlocks whole-codebase reasoning — this changes what you can hand off to AI agents.
Mixture-of-Experts is now the default LLM architecture
Analysis of the top 20 new models released in Q1 2025 shows 17 use MoE. Dense transformers are becoming legacy.
Understanding MoE routing is now table stakes for anyone deploying or fine-tuning foundation models.
OpenAI o3 mini API goes GA with structured output support
The reasoning model now supports JSON schema enforcement, making it practical for production pipelines that need deterministic output shapes.
Reasoning + structured output finally makes LLMs reliable enough to drive business logic without a validation layer.
Startups & Funding
Cognition AI raises $200M Series B at $2B valuation
The company behind Devin secured funding led by Founders Fund, expanding into multi-agent orchestration pipelines.
Agentic coding tools are moving fast — understanding Devin shapes where you invest your skills.
Vercel acquires Tremor to accelerate dashboard component ecosystem
Tremor's React component library for dashboards joins Vercel's portfolio alongside v0, signaling a push into internal tooling.
AI-generated UIs need high-quality component primitives — this gives Vercel a major advantage in the space.
Linear raises $35M to build AI-native project management
Linear is embedding AI across its entire workflow — from issue triage to roadmap generation — positioning against Jira.
If AI can eliminate Jira, that's a net positive for civilisation and your team's velocity.
Supabase hits 1M databases milestone, launches branching GA
Supabase announces database branching is generally available — preview environments with full Postgres isolation per branch.
Database branching removes the last painful friction point in the Postgres-as-a-service DX.
Open Source
uv 0.6 ships: Python package manager 100x faster than pip
Astral's Rust-based toolchain adds cross-platform lock files, workspace support, and `uv run` replacing virtualenv.
If you touch any Python, uv is now the undisputed standard — learn it now.
Zed editor hits 1.0 with remote collaboration and AI core
The Rust-built code editor by ex-Atom team ships stable with real-time collaboration, inline AI, and sub-millisecond keystrokes.
A credible VS Code alternative built for performance — worth a weekend eval if you've ever felt VS Code lag.
Bun 1.2 brings Node.js compatibility to 99% and ships S3 client
Bun 1.2 passes 99% of the Node.js compatibility tests and ships a first-party S3 client with zero config.
The migration path from Node to Bun is now trivially low-risk — the performance gains are real.
Rspack 1.0: Rust-based webpack drop-in replacement ships stable
ByteDance open-sources Rspack 1.0, a Rust rewrite of webpack that is 10x faster with near-full API compatibility.
If your team is stuck on webpack, Rspack is the lowest-friction way out.
New Products & Tools
GitHub Copilot Workspace goes GA with multi-file agent mode
Copilot Workspace exits beta: plan → edit → test across entire repositories, not just single files.
Whole-repo context is the threshold where AI coding tools become actually useful for real work.
Figma ships Dev Mode variables sync with CSS custom properties
Design tokens defined in Figma now export directly as CSS custom properties, eliminating the design-to-code translation layer.
This closes the last major gap between design and implementation — less bikeshedding, more shipping.
Warp terminal ships AI command search across shell history
Warp's new AI feature indexes your entire shell history and lets you search by intent rather than exact command syntax.
Every developer has a mental overhead tax from forgotten commands. This eliminates it.
Raycast AI gets web search and multi-step workflows
Raycast's AI layer now searches the web in real-time and chains multiple AI commands into repeatable workflow automations.
Your launcher becoming your agent orchestrator is a meaningful productivity unlock.
Frontend Dev
CSS anchor positioning is now baseline available
Safari 18.4 completes cross-browser support. Tooltips, popovers, and floating UI without JavaScript are now possible natively.
Replace entire JS libraries (Floating UI, Popper.js) with pure CSS — major DX and performance win.
React 19 Compiler ships stable — no more manual memoization
The React compiler automatically memoizes components and hooks, eliminating `useMemo`, `useCallback`, and most `memo()` wrappers.
Every React codebase has performance bugs from missing memoization. The compiler fixes the entire class of problems.
View Transitions API Level 2 enables cross-document animations
Level 2 brings View Transitions to MPA navigation — full-page transitions without a client-side router or JavaScript framework.
Native page transitions are the last missing piece that made SPAs feel necessary. Now MPAs can be just as smooth.
Tailwind CSS v4 drops PostCSS, ships Oxide Vite engine
Tailwind v4 is rewritten in Rust (Oxide), moves configuration to CSS, and ships a first-party Vite plugin that is 10x faster.
Configuration-in-CSS is a fundamental DX improvement — you stop context-switching between config and template files.
Architecture & Patterns
Temporal.io raises $75M — durable workflows are going mainstream
Temporal's funding signals enterprise appetite for durable execution primitives as async microservices complexity peaks.
If you're building anything with async workflows, queues, or retries, Temporal is the abstraction you want.
HTMX 2.0 and the return of hypermedia-driven applications
HTMX 2.0 ships with SSE support, better streaming, and a cleaner API — fuelling a broader conversation about SPA complexity.
Most apps don't need a full SPA. HTMX lets you ship faster and hand off easier without framework lock-in.
Event sourcing at scale: lessons from 5 years in production
Engineering team at Miro publishes a retrospective on their event-sourced architecture — what worked, what failed, and what they'd change.
Real production retrospectives on complex patterns are rare. This is worth a deep read before you adopt ES.
How Cloudflare built a globally consistent database on the edge
Deep dive into D1's consistency model — how Cloudflare achieves linearizable reads across 300 edge locations using SQLite.
Understanding the tradeoffs they made reframes how you think about CAP theorem in real-world edge deployments.
Career & Soft Skills
The senior-to-staff gap: what actually changes
Survey of 200 staff engineers: scope shifts from given to created, influence replaces authority, ambiguity tolerance is the core skill.
If leveling up is a goal, understanding the role is the first step to getting there.
AI is not replacing senior engineers — it's replacing junior ones
Analysis of hiring data from 500 tech companies shows junior engineer headcount down 34% YoY while senior roles grew 12%.
The skill ceiling is rising, not the floor. Staying senior means being the person who directs the AI, not competes with it.
How to run a technical design review that people actually find useful
A framework for design reviews from a Principal Engineer at Stripe: async RFC first, sync for tradeoffs only, explicit decisions logged.
Bad design reviews waste everyone's time. This framework makes them a force multiplier.
The art of the engineering estimation
Breaking down why estimates are always wrong and a practical 3-step method for giving ranges that are actually useful to stakeholders.
Every engineer underestimates. A structured approach protects your credibility and the team's planning.
Podcasts & YouTube
Changelog Podcast: The era of AI-native developer tools
In-depth discussion with founders of Cursor, Zed, and Warp on how they think about AI as a first-class interface primitive.
Understanding how tool creators think about AI helps you evaluate which bets are worth making in your own workflow.
Theo T3: Why I stopped using tRPC (and what I use instead)
Theo explores the tradeoffs of tRPC vs React Query + OpenAPI codegen in large teams, with honest takes on where tRPC struggles.
When your stack evangelist walks something back, that's signal worth paying attention to.
Software Unscripted: Functional architecture in the real world
Richard Feldman and Scott Wlaschin discuss how functional design principles apply in languages that aren't Haskell or Elm.
The ideas transfer directly to TypeScript. This episode will change how you structure modules.
Fireship: I rebuilt my entire stack with Astro and I'm never going back
15-minute breakdown of migrating a Next.js app to Astro SSR — performance numbers, DX wins, and the one thing Astro still can't do.
Real migration stories with honest tradeoffs are more valuable than marketing pages.