Zod 75× faster, CSS @function, Tokio overhaul

Zod‑compiler rewrites your Zod schemas into pure JavaScript validators during the build, eliminating runtime overhead. Benchmarks show 2‑75× faster validation without changing any source code, even in projects with tens of thousands of schemas. Faster checks mean lower latency for APIs and tighter type safety in production.

WebKit released an interactive Field Guide for its new Grid Lanes layout (the renamed CSS masonry). The site offers live playgrounds, cheat‑sheet reference, and six real‑world demos, letting developers experiment with waterfall and brick modes, flow‑tolerance, and responsive behavior directly in Safari’s inspector. It speeds up adoption of the emerging layout model.

Custom CSS @function lets developers define reusable functions that return values, encapsulate intermediate variables, and accept arguments, all within CSS. It reduces duplication, improves dev experience, but currently returns only a single value and silently fails on argument mismatches. This shifts complex styling logic from JavaScript toward native CSS, promising cleaner codebases.

Diplomat is a new unidirectional FFI system that generates bindings for Rust libraries across many languages, C++, JavaScript, Dart, JVM and more, from a single Rust API definition. It removes the need for hand‑written glue code, letting teams expose Rust code to heterogeneous ecosystems with far less effort.

Alice Ryhl outlines a completion‑based async I/O model that could replace the current future‑driven approach in Tokio. The design promises lower latency and tighter control over I/O operations, potentially reshaping Rust's async ecosystem. If adopted, developers may see simpler code paths and better performance for high‑throughput services.

The author released DashClaw, an open‑source observability and governance platform for LangChain‑style AI agents. By instrumenting agents with a Node.js SDK and a Next.js dashboard, it surfaces hidden token usage and decision logs that saved a client $2,400 in unnoticed LLM costs.

The Rio de Janeiro "homegrown" LLM, claimed to be a 397B model trained by IplanRIO, is actually a 60% Nex, 40% Qwen weight blend. Independent checks of prompts and tensor values show no unique training, exposing a misrepresentation of government AI capabilities.

The U.S. Commerce Department issued an export‑control directive forcing Anthropic to suspend all foreign‑national access to its Claude Fable 5 and Claude Mythos 5 models. Anthropic complied, shutting the models down worldwide while keeping other models available. The move highlights how national‑security concerns can instantly curtail cutting‑edge AI deployments.

cURL’s maintainer Daniel Stenberg announced a month‑long “summer of bliss”, the project will not accept any vulnerability reports in July 2026. The pause aims to curb maintainer burnout, but it leaves users without a formal way to report critical issues, shifting risk to paid support contracts.

The blog explains how dropping a column mid‑deployment can crash rolling updates because old pods still expect the column. Archestra’s Drizzle migration linter flags such unsafe expand/contract patterns in CI, enforcing zero‑downtime deploy rules for Postgres. Adding the check early saves costly rollout failures.

Massive PostgreSQL DELETEs leave dead tuples, tax replication, and never return disk space. Using DROP TABLE or TRUNCATE removes the underlying files, scaling independently of row count. Redesign schemas to treat bulk removal as table drops and you’ll avoid the hidden cost of large deletes.

Even after two decades, developers still assume perfect networks, zero latency, and unlimited bandwidth. APNIC’s retrospective shows those mistaken beliefs drive bugs, security gaps, and performance woes, urging engineers to redesign systems with real‑world network constraints in mind.