In 2026 the TypeScript toolchain finally standardized. TypeScript 7 (the Go-native compiler) shipped, often ~10× faster. Vite 8 defaults to Rolldown, a Rust bundler on Oxc. Vite+ unifies Vite, Vitest, Oxlint, Oxfmt, Rolldown and tsdown on that same foundation. The "which bundler" debate is basically settled — and the answer is fast.
Which makes the next question more interesting than the last one: execution is solved, so what's the layer above it?
All of these wins are on one axis: speed of execution. And that axis is nearly saturated. Meanwhile your build still does this every single day:
- restart dev server → re-resolve + re-optimize deps
- switch branch and back → re-derive
- CI run → start near zero
- prod build → redo semantic work dev already proved
That's recomputing work that was already verified correct. The fastest tool in the world is still just fast at redoing proven work.
The distinction I'd push on:
- A cache asks: have I seen these inputs before?
- An authority asks: who owns the correct answer — and is it still valid?
A cache optimizes recomputation. An authority owns truth. Different jobs.
So the ladder I care about is execution → persistence → authority:
- Execution — compute fast (2026 nailed this)
- Persistence — don't discard the verified answer
- Authority — one owner per category of verified knowledge (dependency identity, export/ABI, transformed artifacts, CSS, production readiness), allowed to publish it and say if it's still valid
That's what kills "works in dev, breaks in build": dev, CI and prod consuming the same owned truth instead of each recomputing their own.
I'm building this in the open as Ionify — a persistent Build Authority engine in Rust. Honest status: warm builds are validated (incl. a ~11k-module / ~25k-dependency stress test); the cold first build is still improving. Not a launch — research in the open.
Curious what you think: in your projects, how much build time is spent recomputing things you already proved correct?
Top comments (0)