The asteroid currently hitting frontend web development

Frontend educators are pulling back, and Nolan Lawson thinks that’s not a random coincidence. He names people like Axel Rauschmayer, Salma Alam-Naylor, and Josh W. Comeau as examples of folks easing off, while others such as Kent C. Dodds, Addy Osmani, Rachel Nabors, and Lydia Hallie have shifted their attention toward AI instead of the old frontend beat.

Lawson’s own example is telling. He used to write and talk about browser performance — CSS internals, shadow DOM, style engines, the usual slippery stuff that trips people up — and then asked Claude Sonnet to inspect a Chrome trace with high Style cost and low Layout cost. The model gave a solid answer: look for selector complexity, broad invalidation from changes near the root, repeated recalculation from JS loops, inherited property churn, and shadow DOM or scoped-style repetition. In other words, it handled a genuinely tricky frontend debugging prompt about as well as many humans would.

That matters because it hints at where AI is most likely to flatten frontend expertise. Lawson argues that frontend work is easier to hand to an agent unsupervised than, say, a database migration. A React component can be yolo’d into production with less fear, even if it still might break accessibility or create an infinite loop. And once agents are writing the code, “agent experience” starts to matter more than developer experience. That, he says, helps explain why React keeps showing up in migrations from Solid and Lit: the agents know React better.

He also thinks the center of gravity in standards work is shifting. Polishing syntax for the sake of developer ergonomics may matter less than features that actually change what browsers can do. Shadow DOM and custom elements, for example, change where code is written more than what the browser can do. That’s a pretty brutal summary of a whole category of frontend hope.

Lawson isn’t fully doomposting, though. He sees some room for frontend education in teaching agents the big picture, not just how to spit out SPAs for everything. He also thinks there’s real value in pushing sites toward server-rendered content, accessibility, and page speed, especially as more people wrap “AI” around the basics to make them sound new. And then there’s the mess of vibe-coded sites: if the machine writes the frontend and the machine also breaks it, somebody still has to clean up the wreckage.