Nolan Lawson’s October 3 post, “Why don’t more developers ‘use the platform’?”, sat at 259 points and 262 comments on Hacker News by the weekend — for a question about web standards, roughly a papal conclave. Lawson has spent years as one of the advocates he’s now interrogating, and he doesn’t flatter his own side. His pessimistic branch: because LLMs love duplicating code — cheerfully ignoring the helper function that already exists in favor of writing their own for the umpteenth time — the volume of custom, non-idiomatic code will skyrocket, as developers commit the agent’s first draft without asking it to test alternatives or try a second approach.
Every comment thread under that post is having the same argument about awareness. Somebody needs to teach these kids about the platform. That argument is a decade stale, and it is aimed at the wrong side of the ledger.
The Platform Stopped Shipping the Layer People Write In
If the problem were ignorance, it would have been solved. The documentation exists. Can I Use exists. The syntax is on MDN, in the browser, free, forever. Nobody uses a framework because they’ve never heard of a <dialog> element.
They use it because that’s where the spending is. Next.js alone is used by over four million developers, according to a March writeup at Digital Applied, and the ecosystem around it employs people whose entire job is making the first fifteen minutes pleasant. Meanwhile, as Lawson notes on his own blog, the standards conversation has quietly drifted from “use this newer syntax” to “here are these emerging capabilities” — which is what advocacy sounds like when it has stopped asking you to change your code and started asking you to be aware of its roadmap.
The platform’s problem is not that developers won’t read. It’s that the platform has largely stopped shipping the layer developers actually write in, while the framework layer ships it weekly and sends someone to your conference to explain it.
Reuse Is a Deflationary Act
Now follow the money, because it explains the LLM behavior Lawson is worried about better than any theory of developer laziness.
Sacra, which models private-company revenue, estimates Vercel at $340 million in annual recurring revenue as of February 2026, up from roughly $305 million at the end of 2025, with a $9.3 billion valuation. Digital Applied’s writeup on the Series F notes the previous round priced the company at $2.5 billion in 2021. That is a five-year re-rating from a frontend host into, in the company’s own framing, an AI cloud — Fluid Compute, an AI Gateway, model providers behind a single bill.
Here is the part the standards crowd hasn’t metabolized. A shared helper function is a public good: written once, cached, imported a thousand times, billed never. A regenerated utility function is a transaction. An agent that writes its own date formatter instead of importing one isn’t making a mistake that costs its owner a few tokens — it’s making a purchase. Deduplication is deflationary. Generation is metered.
No one has to intend this. That’s what makes it durable. The default of a metered pipe is to bill every attempt; the default of a shared module is to bill zero. Point a generation engine at a codebase and let the market choose, and you will get exactly the outcome Lawson predicts — not because developers are incurring a knowledge debt, but because the incentives are aligned, and they are not aligned with reuse.
An engineer I know on a mid-size hosting vendor’s platform team — mid-level, not a founder, the kind of person who reads changelogs for fun — put it to me while we were both waiting at a gate in Denver after a developer summit: “Our migration guides get fewer reads every quarter. Nobody wants to learn our abstractions anymore. They want the agent to produce something that compiles, and they’ll pay per attempt until it does. That’s the product now.”
Your Framework Is Next, and It Knows It
The flattering reading of the last decade is that frameworks beat standards and won. The less flattering reading, available in any competitive breakdown — one aggregator has Vercel at roughly $200 million in revenue against Netlify’s $75 million — is that frameworks won a war whose prize is already being repriced.
If agents write from scratch, an API surface stops being a moat. A framework’s strategic value was always that developers memorized it, defended it in code review, and hired for it. An agent has no muscle memory and no loyalty; it has a context window and a latency budget. The framework’s most valuable asset — the thing four million developers have memorized — is precisely the thing a model can memorize faster and better, and then reproduce on demand, for a fee.
Which is why the smartest company in the category stopped describing itself as a framework company. Vercel’s own pitch has migrated to the metered layer. Sacra attributes $863 million in total funding to the company; the interesting question is not how much of that goes into making Next.js better, but how much goes into making generation faster. Those are different bets with different winners.
The Tell to Watch
I’m not proposing a rule here, and I’d be suspicious of anyone who does. Standards bodies have never been able to outspend anyone.
The tell is simple. Watch whether the next browser capability is designed to be imported or designed to be inferred. If it’s the former, the platform has a future as a commons. If it’s the latter, the platform has a future as a feature of somebody’s inference bill — and “use the platform” becomes advice that costs money to follow, which is a strange fate for a slogan whose whole appeal was that it was free.
Lawson asked why more developers don’t use the platform. The honest answer is that the platform stopped being the cheapest thing to use. Reuse is cheap for everyone, which is exactly why nobody is building a business on it.