Vue 4 Promises No Breaking Changes, ESM Only — And a Toolchain Built for AI Agents

On October 24 at Vue Fes Japan 2026, Evan You presented three confirmed directions for Vue 4: no large-scale breaking changes, ESM-only distribution (no more CommonJS builds), and several new-feature RFCs still under discussion. The first two are commitments; the third is intentional openness. The most anticipated question — whether Vapor Mode will become the default rendering path — received no answer.

The Weight of "I Promise"

The Vue 2 to Vue 3 migration serves as the backdrop. Vue 3 launched in September 2020 but did not become the default version until February 2022; Vue 2's end-of-life was extended to October 31, 2024. Over four years, a large portion of projects remained on 2.x. Migration pain was concrete: Options API gave way to Composition API, filters were removed, $listeners merged into $attrs, and the entire ecosystem had to upgrade in lockstep. Therefore, "I promise" guarantees not syntax stability but the most expensive asset a framework owns — migration cost.

A counter-argument in the community asks: if a major version breaks nothing, does it change anything? The answer may lie in Vapor Mode. If it becomes the default rendering path, the transformation happens in the compiler and runtime, not in your script tags. Code stays untouched while performance rises automatically.

ESM Only: The Only Certain Change

const { createApp } = require('vue') // still works today

import { createApp } from 'vue' // after Vue 4, only this works

Modern projects using Vite or webpack 5 are largely unaffected because they already use import. The impact falls on legacy SSR services, older Electron main processes, and test tools like Jest that have historically struggled with ESM support. This is not a syntax upgrade but a module-spec cleanup: it won't error in component code, but it will error in build configuration.

What Matters More Than Vue 4 Is the Toolchain

Half the presentation focused on the JavaScript toolchain, with the keyword "agentic." For the past decade, tooling optimized for human coding speed: cold-start time, HMR latency. Now the target is different — the probability that an AI agent modifies code correctly. Humans watch HMR speed; agents need structured error messages, statically inferable types, and dev servers that can be programmatically driven. This explains Vite's multi-year investment in Rolldown and Oxc: rewriting bundling and parsing in Rust fundamentally swaps "output for humans" with "output for machine verification." For everyday developers, the implication is clear: the configuration line you write tomorrow may be read by a machine, not a person.

Closing Thoughts

Although Vue 4 has no public release timeline, it has already begun proactively absorbing migration costs on behalf of users.