Last week, it came to light Cursor had mostly finished migrating from SolidJS to React. This migration happened about seven months ago. But it became a central focus of discussion following the Solid 2.0 RC release. Then yesterday, a week later, it came to my attention that the Anthropic docs example command for their large-scale migration feature is:
I admit that my gut reaction was not great. Out of all the examples they could have chosen... Years of my work became a canonical example of the thing you migrate away from — in the same week we shipped the biggest release in the project's history — stung in a way I won't pretend it didn't. My second reaction was to assume that, like the other trickle-down posts I'd seen this week, this rode the same week-old news cycle.
Then I checked the Internet Archive and realized this has been there since at least April 2026. Four months before the Cursor story broke. At this point, the whole public footprint was a mention of an experiment sandwiched between bigger updates in a Cursor blog post posted in January. The kind of thing that no one outside the industry would even really pick up on. No reasoning, no benchmarks, no argument.
Stop to think about what that means. I should be careful here because I can't prove anyone at Anthropic ever read that Cursor post. Nobody can. Maybe a docs writer saw the experiment. Maybe Claude drafted its own example. But think it through. Either it traveled from a buried line in one company's release notes into another company's official docs, or it needed no origin at all. It was already assumed before any public migration existed.
Our industry has quietly started broadcasting conclusions where it used to transmit arguments. We couldn't have picked a worse time, because — as I'll get to — arguments are the only source that still matters.
It would be fair to ask, hasn't it always been like this? Teams cargo cult large players. Netflix or Facebook uses this predates AI by decades. Optics and politics have always beaten pure technical merit. The weight of Facebook definitely helped React's early propagation.
But a narrative used to come with friction. It was always in the race, but it couldn't outrun the argument. This has changed for two reasons:
First, execution cost has collapsed. Bun's rewrite from Zig to Rust — about a million lines — was executed almost entirely by Claude agents in days, not months. And before you start thinking "lock-in", going the opposite way, while not as easy, has never been easier. Migrating from Rust to Zig or React to Solid has never been easier.
That sounds like good news. And it is. When migrations took years of human effort, cost put a damper on fashion. Now execution is a lot cheaper in all directions. The only thing left is the reason. Which suggests, on the surface, these decisions should be rooted in technical merit. But when narrative is what sets the direction, you start to see how a line buried in documentation is worth an essay.
Second, verdict production has been industrialized. Cognition just migrated its marketing site from Astro to Next.js. A content site. Astro's home turf, by near-universal consensus, including from people with no stake in the fight. The migration was performed by Devin itself and published as a case study: the agent made the changes, tested them, recorded its own verification runs. No repository, no before-and-after numbers, no benchmark. The output wasn't the site. It was the story.
And believe me, there is no shortage of these stories. Migrations are now the demo genre of the agent industry. Theo Browne was trying to tell me this a couple weeks back with some very good advice. But it hits a lot harder when you feel it firsthand. Migrations that demo flawlessly are, by definition, the ones moving toward what the agent writes best today. Marketing departments are generating "X → dominant library" verdicts at a rate organic engineering decisions never would.
So you see the tension. Stacks have never been more sensitive to circulating verdicts, and verdicts have never been produced faster or with less reasoning attached.
When Bun moved from Zig to Rust, Jarred Sumner made the case in public, with receipts. He talked about bug classes and memory management issues. You can disagree with the argument. People have, point by point, in public. That's the value. An argument can be engaged, checked, narrowed, refuted, or strengthened. A verdict can only be repeated.
Or look at TanStack's journey with React Server Components on tanstack.com. They spent the year almost arguing with themselves. When tanstack.com adopted React Server Components, they wrote it up and measured it. When they stopped using them, they did the same: What changed, what replaced it, why the tradeoff flipped for their specific case — while TanStack Start went right on supporting RSC as an opt-in primitive for everyone else. The reversal post calling the decision "uneventful" is a tell. A public reversal on React's flagship architecture, and it generated insight instead of heat. Because they delivered analysis instead of verdict.
That's what evaluation entering the public record looks like. jQuery didn't lose to React because a training set said so. It lost an argument, in public, on merit over several years. Every library you respect got where it is by winning that kind of fight. The flow of events I described earlier doesn't hold fights. It holds direction, and compounds on it.
In all fairness, Cursor never issued a verdict. An experiment that "still needs careful review", then a congratulatory post on the Solid 2.0 RC release. But that post still characterized signals as "perf footguns" and "accidental fan out" — no numbers, nothing to engage — and reached straight for the React Compiler as the cure. No bad faith required. The diagnosis and the cure just came from the same place.
When I tried to reconstruct the case Cursor could have made, it turned out to be interesting. Far more interesting than "perf footguns."
The post obviously highlighted agents were "quite bad at writing good Solid code," everything "ended up being accidentally tracked." While my "Solid-brain" struggles a bit to imagine what they were doing to get there, I don't doubt it for a second. An agent that has internalized React's mental model will write Solid like React, and Solid written like React is bad Solid. I've also seen the reverse. The difference is when the agent fumbles React, the training data catches it. When it fumbles Solid, the training data shrugs. This is a pure numbers game.