Lightnews — Scholar-powered news

Kenton Varda

@kentonvarda.com

1.8K followers 130 following 180 posts

Tech lead for workers.cloudflare.com Also made: capnproto.org, lanparty.house, sandstorm.io @KentonVarda on Twitter.

Posts Media Videos Starter Packs

Pinned

Kenton Varda @kentonvarda.com · 23d

I don't have a problem with making new RPC systems! I can stop any time I want! But I don't want to. So I made a new one for you. It's called Cap'n Web. It's like Cap'n Proto RPC, except focused on working great in browsers and JavaScript/TypeScript servers. blog.cloudflare.com/capnweb-java...

An image of a circular fabric patch showing some sort of superhero in a nautical hat? Cap'n Web: Protect & Connect, it says. Looks like AI slop.

11 19 72

Kenton Varda @kentonvarda.com · 1m

Where, to be clear, "they" is Theo and his team, not me or Cloudflare. But indeed, "vanilla" was much faster than any framework.

Kenton Varda @kentonvarda.com · 19h

Something like that. Not sure if it scales linearly.

Kenton Varda @kentonvarda.com · 20h

The vanilla case actually generates 3x more HTML compared to next.js because it was too fast to measure reliably otherwise.

The SvelteKit case is supposed to do the same work as the next.js case though.

The react case produces less HTML than the others, dunno why.

1 4

Kenton Varda @kentonvarda.com · 21h

Correction: I'm told that the SvelteKit benchmark does actually perform the same work as next.js. The "vanilla" benchmark, though, generates like 3x the HTML.

1 5

Kenton Varda @kentonvarda.com · 21h

The benchmark cases are not comparable to each other, they all do different stuff.

Kenton Varda @kentonvarda.com · 21h

The benchmarks render 5MB to 15MB of HTML. They are not normal workloads.

1 3

Kenton Varda @kentonvarda.com · 1d

Those benchmarks from @t3.gg revealed some interesting issues in Workers. Happy to say it's all fixed now, save for some lingering next.js-specific stuff we're continuing to work on. Thanks for the reproducible test cases, @t3.gg.

Excessive details in blog post: blog.cloudflare.com/unpacking-cl...

6 9 85

Kenton Varda @kentonvarda.com · 4d

Personally, I don't use the "for you" tab, I only use "following", and I basically don't see any politics at all. (I tend to unfollow anyone who posts politics TBH.)

I was actually doing that even before the acquisition, so didn't see much change after.

1 4

Kenton Varda @kentonvarda.com · 10d

Exactly as I remember it, in excruciating detail. Why does my brain continue to store this while forgetting people's names?

Kenton Varda @kentonvarda.com · 11d

Wait... I think someone made, like, a flash animated version of this?

I can still hear it in my head, including the heart beating at the end.

Kenton Varda @kentonvarda.com · 11d

Wow I remember this one. Despite not thinking about it once for 25 years. Why is this occupying space in my brain?

1 3

Kenton Varda @kentonvarda.com · 11d

Is it not Durable Objects?

1 1

Kenton Varda @kentonvarda.com · 14d

I don't personally know much about atproto or next.js, but this all sounds easily doable in Workers. @dom96.picheta.me built @listifications.app on top of Workers, FWIW.

2 3

Kenton Varda @kentonvarda.com · 19d

The ideal API surface to expose to a dynamically-loaded Worker sandbox is a Cap'n Web API. blog.cloudflare.com/capnweb-java...

Cap'n Web: A new RPC system for browsers and web servers

Cap'n Web is a new open source, JavaScript-native RPC protocol for use in browsers and web servers. It provides the expressive power of Cap'n Proto, but with no schemas and no boilerplate.

blog.cloudflare.com

Kenton Varda @kentonvarda.com · 19d

That's the next step I think. MCP is convenient because it's designed for the whole end-to-end experience of connecting an agent to an arbitrary MCP server including auth, etc. But once we fully solve those for e.g. OpenAPI, we should be able to go direct.

2 2

Kenton Varda @kentonvarda.com · 19d

Sorry by "the readme" I mean: github.com/cloudflare/a...

github.com

Kenton Varda @kentonvarda.com · 19d

No, the example should work correctly if you follow the instructions in the readme. It works for me. But only in local dev. To deploy it to prod you would need to be in the closed beta for worker loaders.

1 1

Kenton Varda @kentonvarda.com · 19d

Yes exactly. It's cheaper sandboxes. Also, the capability-based model is well known to be a great fit for sandboxing -- much cleaner than trying to filter and block arbitrary network requests.

1 2

Kenton Varda @kentonvarda.com · 19d

Anyway, please don't murder me for the quip that kinda sorta compares LLMs to Shakespeare.

/me runs away

Kenton Varda @kentonvarda.com · 19d

Dynamic Worker loading, of course, has applications far beyond just MCP or AI. It's a general-purpose capability-based sandboxing mechanism. If I had more time I'd build Sandstorm.io 2.0 on this...

1 5

Kenton Varda @kentonvarda.com · 19d

This is all based on a new Workers Runtime feature called Dynamic Worker Loading, which lets you dynamically load arbitrary code into a fresh isolate. Isolates are wayyyyyy cheaper and faster to load than containers. Perfect for executing one-offs. developers.cloudflare.com/workers/runt...

Dynamic Worker Loaders

The Dynamic Worker Loader API, which allows dynamically spawning isolates that run arbitrary code.

developers.cloudflare.com

2 3

Kenton Varda @kentonvarda.com · 19d

This works better than direct tool calls becaues LLMs actually have WAY more training data for writing code than for calling tools. After all, tool-calling is a sort of artificial thing just for LLMs. Code is everywhere.

2 1 11

Kenton Varda @kentonvarda.com · 19d

Hey Bluesky, I know how people here just LOVE everything AI, so I wanted to show you what @threepointone.bsky.social and I have been up to. We made an agent that calls MCP not as "tools", but by converting the tools into a TypeScript API, and writing code against that. blog.cloudflare.com/code-mode/

Code Mode: the better way to use MCP

It turns out we've all been using MCP wrong. Most agents today use MCP by exposing the

blog.cloudflare.com

2 1 27

Kenton Varda @kentonvarda.com · 19d

Oh! The stub is being disposed because you are returning it which transfers ownership to the caller. Try `return customerSession.dup()` instead. See the docs in the readme about automatic disposal.

1 2

Kenton Varda @kentonvarda.com · 19d

Sorry, I meant, "not one per RPC".