/
Published on 14.03.2026
Temporal hits Stage 4 standardization — JavaScript finally has immutable, timezone-aware date/time handling with nanosecond precision after 9 years of TC39 consensus-building. Already shipping in Firefox 139, Chrome 144, and TypeScript 6.0 beta.
Vite 8 ships Rolldown bundler — A unified Rust-based bundler replaces the dual esbuild+Rollup approach, delivering 10-30x faster builds while maintaining full plugin compatibility. Real-world projects report 38-64% build time reductions.
TanStack launches generation hooks — React hooks for image generation, text-to-speech, transcription, summarization, and video, with type-safe streaming and server function integration. Same API across all generation types.
The problem was baked in from day one. In 1995, Brendan Eich had 10 days to create JavaScript. He pragmatically ported Java's Date implementation directly—a decision made under time pressure that echoed through 30 years of developer pain. Date is mutable (mutations happen in place), doesn't handle month arithmetic correctly (adding one month can silently roll into the next month), and parses "almost ISO" strings ambiguously depending on the browser.
By the 2010s, this became a critical pain point. JavaScript was powering banking terminals, trading systems, collaboration tools—all in different timezones on earth. Libraries like Moment.js (first released 2011) filled the gap, but at a cost: adding 100+ MB downloads of locale and timezone data wholesale, because most devs didn't know which timezones they'd need.
Temporal types:
Temporal.ZonedDateTime — exact moment + explicit timezone + explicit calendar + daylight saving correctness, fully immutableTemporal.Instant — exact moment in time, nanosecond precision (not milliseconds like Date)Temporal.PlainDate, PlainTime, PlainDateTime — "wall time" without timezone/calendar concernsTemporal.Duration — for arithmetic operationsThe implementation was itself a technical milestone. Temporal is the biggest addition to ECMAScript since ES2015, with ~4,500 tests in Test262. Instead of duplicating implementation effort across engines, Google and Boa collaborated on temporal_rs, a Rust library that V8, Boa, and other engines now share. This reduced friction, made the codebase easier to review, and created a maintainable home for fixes post-standardization.
Temporal: The 9-Year Journey to Fix Time in JavaScript
Real-world results:
The migration path was deliberate. A separate rolldown-vite preview package shipped first, letting teams test on real codebases before it became the default in Vite 8 beta. This feedback surfaced edge cases the team fixed before the stable release. CI validated key plugins and frameworks against the new bundler to catch regressions early.
Additional features in Vite 8:
resolve.tsconfigPaths: true.wasm?init imports now work in SSRThe trade-off: Vite 8 is ~15 MB larger than Vite 7 (Rolldown itself is ~5 MB, lightningcss now included is ~10 MB). This was a deliberate choice: better CSS minification and Rolldown's performance optimizations favor speed over binary size.
What's coming next:
Foreign JSON (APIs, databases): Parse it with validation schemas like Zod. Don't cast. The absolute worst lie is const user = (await res.json()) as User; because you're asserting the shape of data from across a network boundary. Use Zod's .parse() to get runtime proof. If the API changes shape, you get a clear error instead of "undefined is not a function" three frames down the stack.
Generic boundaries: One as cast in the utility, validation at the call site. Example: safeFetchJson<T> accepts the cast internally (boundary between any and the typed world), but callers must validate with Zod. This keeps application code cast-free.
Type guards: When you've already checked a condition but TypeScript can't narrow, write a type guard instead of casting. Example: function isImageNode(node: { type: string }): node is ImageNode { return node.type === 'image'; }. Reusable, and keeps the lie out of application logic.
Assertion functions: Like guards but throw on failure and narrow for the rest of the scope. Best for preconditions inside functions. For external data, still prefer Zod.
Catch clauses: catch (err) gives you unknown. Don't cast to Error immediately. Use Error.isError(err) ? err.message : String(err) instead. Works across realms (iframes, workers, vm contexts).
ORM JSON columns: Type at the schema level using $type<T>() in Drizzle or similar patterns in Prisma. Zero casts at call sites.
The hierarchy:
as in the utility, validation at call siteKey takeaway: The goal isn't zero as—it's zero unexamined as. Every cast should be a conscious decision, not a reflex.
Absorbing unknown Into the Type Realm
Three transport modes:
The genius is in the detection: if your fetcher returns a Response, it's treated as SSE. If it returns anything else, it's a direct result. No flags, no configuration, no separate hook.
Result transforms: Every hook accepts an onResult callback to transform results before storing. TypeScript infers the output type. Example: convert base64 audio to a blob URL for playback.
Video generation is special. Providers like OpenAI's Sora use a jobs-based architecture (submit prompt → get job ID → poll for status). useGenerateVideo() handles this transparently, exposing jobId and videoStatus as reactive state. Users see "pending", "processing", progress percentages, and finally the video URL—no polling loop required.
Adapter ecosystem: Generation functions are provider-agnostic by design. Swapping from OpenAI to Anthropic or a local model is a one-line change.
Key takeaway: 3 lines of hook setup gives you type-safe input, streaming progress, error handling, and abort support. That's the whole feature.
Generation Hooks: Type-Safe AI Beyond Chat | TanStack Blog
Speed tiers based on what changed:
Most teams start by offloading mobile builds to Workflows, then expand. You don't rip out your CI—integrate gradually. Instant jobs (~300ms) for Slack notifications, artifact lookups, lightweight tasks. Zero CI minutes consumed.
Operators report ~80% reduction in DevOps overhead, freeing engineers to focus on features.
Key takeaway: From indie devs to enterprises (HipCamp, PrizePicks, InfiniteRed), Workflows replaces sluggish, brittle build processes with something quick and dependable.
Mobile CI/CD built for React Native
What this unlocks:
pg_basebackup covers all three concerns (git mirror, artifact store, database replication).pg_dump: Backs up application code, git history, and application state all at once.The trade-off: The monolith only works if you accept that a bad deploy can take down your database. That's a management problem, not a technical one, but it needs more than a trigger and EXECUTE to handle safely.
Disclaimer: This is experimental and proof-of-concept. Large repos will be slow (packfile generator skips delta compression), deploy trigger isn't battle-tested with real applications, no auth, protocol v1 only, no concurrent push handling. It works for the happy path on small repos.
The insight: when everything is in one database, operational concerns like replication, backup, recovery, and monitoring become dramatically simpler because you're not coordinating across multiple storage systems with different consistency models.
This article summarizes technical newsletters and curated links for developers. All views and opinions expressed here are for educational purposes. Verify claims and evaluate tools based on your specific needs before adopting them in production.