AI Is Killing the Framework Era

Hinzugefügt: 2026-06-03

[00:00:00]Every year, Ryan Carniato, the creator of SolidJS, and the person most responsible for pushing the entire JavaScript ecosystem towards signal-based reactivity, publishes his take on the state of web development.

[00:00:14]It is one of the more honest and technically grounded reviews in the space, which makes it worth paying attention to. So, today we are going through the architectural shifts that are actually changing how modern web apps get built, why the framework wars are finally over, and why the conclusion of those wars is more depressing than anyone expected. Let's start with React, because everything starts with React, whether you want it to or not. React's problems are not exactly a secret at this point.

[00:00:44]What began as a simple UI library has grown into an ecosystem that requires a full mental re-education just to render a button without accidentally triggering a re-render cascade, a hydration bug, or a suspense waterfall.

[00:01:01]The virtual DOM, once considered an elegant abstraction, is now widely viewed as architectural overhead, a tax you pay on every interaction because of decisions made in 2013 that the framework cannot undo without breaking half the internet. An entire generation of front-end developers just felt their eye twitch reading that sentence. Welcome back.

[00:01:25]React server components were supposed to fix this. They mostly made it worse.

[00:01:31]The community's general verdict is that RSCs blur the line between server and client in a way that creates more cognitive load than it eliminates. You are no longer writing a front-end application. You are writing something that is partially a server, partially a client, and entirely confusing about which rules apply where. Then came the React compiler, introduced to automate the the memorization work that was driving developers insane. The pitch was reasonable. Stop writing use memo and use callback everywhere.

[00:02:04]Let the compiler figure it out.

[00:02:07]In practice, what happened is that the compiler became another layer of magic sitting underneath your code, making bugs harder to trace and behavior harder to predict.

[00:02:19]Ryan's take is direct. Instead of fixing React's architectural debt, the compiler is a bandage applied on top of it.

[00:02:27]Because React's unit of reactivity is the entire component function, the compiler has to perform increasingly heroic feats of aesthetic analysis just to prevent basic performance problems that other frameworks simply do not have. But here is where it gets genuinely interesting.

[00:02:45]React's biggest strength in 2026 might actually be becoming its biggest liability.

[00:02:51]We are now in a phase where a significant portion of code gets written by AI, not by humans.

[00:02:59]And because React has the largest training data set of any front-end framework in history, years of Stack Overflow answers, GitHub repositories, blog posts, tutorials, AI is extraordinarily good at generating React code. The problem is that React went through many architectural phases, had no strong imposed structure, and accumulated a decade of questionable patterns that were considered best practice at the time. The AI was trained on all of it, the good code and the bad code and the 2019 code that nobody should be writing in 2026.

[00:03:35]Ask an AI to build you a React component today, and depending on your luck, you might get modern hooks, or you might get use effect patterns from 4 years ago, deprecated dependencies, state management anti-patterns, or if you are truly unfortunate, a class-based component. The framework's dominance is causing a regression loop where AI confidently produces outdated React because outdated React is statistically the most common React that has ever existed.

[00:04:06]Ryan's argument is that the industry's attachment to React because AI can write it is actively holding back the web. If we stop innovating because the AI handles the complexity, we are building the future on top of a frozen mental model that no longer matches how browsers, networks, or users actually behave. One concrete signal of this shift is the Remix 3 rewrite, which is notable specifically because it moves in the opposite direction from where everyone assumed frameworks were heading.

[00:04:40]Instead of adding more abstractions to help AI reason about intent, Remix is removing them. They are stripping the framework back towards standard web primitives, which sounds like progress until you realize it is essentially a very expensive round trip back to where we started. 15 years of framework innovation and the destination is the web platform but angrier.

[00:05:03]This connects directly to what Ryan calls the isomorphic first phase. For the past few years, the dominant trend was pushing more logic to the server.

[00:05:12]Islands architecture, React server components, server-side rendering, all in the name of performance and lighthouse scores.

[00:05:21]The problem is that server-driven UI models fall apart when users actually start interacting with your application.

[00:05:28]Islands and RSCs work beautifully for static content. They crack under the weight of real interaction, long-lived client state, and complex user flows.

[00:05:39]Navigation becomes declarative again.

[00:05:41]Mutations become page-oriented.

[00:05:44]Every simple client-side interaction has to cross an invisible server-client boundary. So, after years of moving logic to the server, frameworks are now quietly reversing course and asking how to make the server and client feel like one coherent thing instead of two countries with an expensive border crossing. The patterns emerging across multiple frameworks right now, out-of-order streaming, server functions, granular optimistic UI, single flight mutations are all attempts to answer that question. Out-of-order streaming means the page renders as data arrives. Server functions let you call server logic directly from client code without pretending they are different worlds. Granular optimistic UI updates only the specific value that changed rather than re-rendering entire components. Single flight mutations deduplicate repeated actions so that a user clicking a button three times does not trigger three conflicting server requests. None of these ideas are new.

[00:06:47]Web development is, as usual, the last discipline to arrive at conclusions everyone else reached a decade ago. And then spends another decade arguing about them anyway. If you have ever watched a room full of front-end developers discover database transactions for the first time, you know exactly what this feels like.

[00:07:05]When modern frameworks advertise progress by deleting things rather than building things, it usually means the abstraction ceiling has already been hit.

[00:07:16]The historical footnote here is worth mentioning.

[00:07:19]The isomorphic JavaScript idea.

[00:07:22]Write your logic once, run it on both server and client, stop pretending they are fundamentally different.

[00:07:29]Was the entire pitch of Meteor back in 2012.

[00:07:34]It was genuinely exciting at the time.

[00:07:36]It also failed eventually because of tight coupling, global state problems, and scaling limitations that the architecture could not handle.

[00:07:45]Ryan might thread that needle where Meteor could not. His track record on signals suggests he is worth taking seriously. But after 15 years of watching the JavaScript JavaScript ecosystem reinvent the same ideas under different names, it is difficult it's difficult to work up genuine excitement about the next reinvention cycle. At some point, the pattern becomes the story. And that is maybe the most honest thing that can be said about the state of JavaScript in 2026.

[00:08:19]The tools we have right now are good enough. The frameworks are converging.

[00:08:24]The wars are over. The winner is a draw.

[00:08:27]And if you ever genuinely hit a limitation that nothing in the ecosystem can solve, the answer, apparently in 2026, is vanilla JavaScript, which is somehow making a comeback as a serious recommendation. We started the decade trying to abstract away the web platform.

[00:08:48]We are ending it by rediscovering that the web platform was fine.

[00:08:52]See you next time.