Making AI Coding Work for Enterprise-Grade Browser Extensions
summary
Defines browser extensions as Distributed Systems in a Box and prescribes four concrete practices: pick a foundation (WXT vs Plasmo or PixieBrix), put 90% of your UI belongs in "boring" React/Vue, instrument with TypeScript and tag types, and create fast feedback loops using Playwright, Rainforest QA and Canary testing. Uses AI-generated lint rules written for each preventable bug and AI-generated pages that reproduce Shadow DOM leaks and hostile host-site race conditions so regressions vanish.
AI-written summary · about
chapters (12)
- 0:00Intro: Permissionless innovation in extensions
- 1:15The Extensibility Spectrum: From user scripts to low-code
- 2:41The AI Gap: 3 reasons why standard tools fail at SDLC
- 3:33"Distributed system in a box": Decoding hidden architecture ⚡️
- 4:08Managing async messaging and complex component lifecycles
- 4:35Solution 1: Choosing the right foundation (WXT vs Plasmo)
- 5:50Solution 2: Isolating "magic" to maximize AI context
- 6:37Solution 3: Fast feedback loops and AI-generated lint rules
- 7:19Shifting left: Building a robust testing hierarchy for extensions
- 8:11Solution 4: Using AI to generate test pages for Shadow DOM leaks
- 9:05Beyond Code: Enterprise-grade requirements and trust centers
- 9:42Hands-on with the PixieBrix platform
related talks
32mConversations Without Prompts: AI in the Web
10mWhy are Web Extensions fantastic for AI?
20mDevelop your own Browser extension
21mand AI
33mBecome a wizard and make the browser your own
22mExtending the Web - explore new UI surfaces with WebExtensions
29mEverything You Ever Wanted To Know About Cross-Browser Web Extensions
01mEntrevista