Guillermo Rauch

⣴⣿⣿⣿⣿⣿⣿⣿⣿⣿⣦ ⣿⡏⠀⠀⣀⠀⣀⠀⠀⢹⣿ ⣿⡇⠀⠀⣿⠀⣿⠀⠀⢸⣿ ⣿⣇⣀⣀⣀⣀⣀⣀⣀⣸⣿ ⠻⣿⣿⣿⣿⣿⣿⣿⣿⣿⠟

Rebuilt the Roadster demo in a playable, highly accurate simulation that retains a lot of the IRL details: https://tesla-ramp.vercel.app

TypeScript is the new TypeScript 😂 When 𝚜𝚌𝚛𝚒𝚙𝚝𝚌-compiled, outperforms tsgo by 2-4x

Harness durability in http://fx.sh. The same 𝚌𝚛𝚎𝚊𝚝𝚎𝙵𝚡𝙰𝚐𝚎𝚗𝚝() is now *durable* on your laptop or in the cloud, via files, Postgres, or Vercel Workflows. Node.js or Bun. Agents need long-running compute. But they also need to tolerate restarts, redeploys, timeouts, crashes… Concretely, if a user sends a prompt and then throws their phone into a black hole, we ensure that turn runs to completion. If the process crashes mid-tool-call, it resumes from there. To this end, 𝚕𝚒𝚋𝚏𝚡 introduces a simple durability adapter API. On Vercel, it leverages all the infra we built for durable workflows and queues, giving you reliability with 1LOC. This is a critical aspect of production-grade agents (and something all @eve agents also get out of the box). Next, we'll show you how we bring the same durability to Pi, OpenCode, and even your own custom harnesses.

It’s time to 🇺🇸-max. Beam is an American-made open-weight model by Reflection. And it’s now free on @vercel AI Gateway. Plug it into deployable @eve agents, start coding with @fx, or build your own harness. Freedom. Never bet against America.

I think @tobi is the epitome of the modern entrepreneur. A CEO that's mission-driven, detail-obsessed, and a hands-on engineer who communicates openly to the world on @x. Yesterday we announced http://vercel.shop in GA, and http://vercel.store 👕, our own Vercel merch store dogfooding it. Vercel Shop gives you a production-grade, blazing-fast Next.js storefront that's agent-ready out of the box, with @shopify as its backend. Shopify and Vercel share the mission of enabling and powering the entrepreneurs of the future. Excited to see what people build with us 🫶

TLDR we got Vercel zeppelin and flying cars before GTA6

🤯

What a SHIP!

What should I ask @tobi?

Free AI

Part of the fam backstage 😁

I recently described the modern scalable agent architecture as: 🧠 brain, 🙌 hands, 🗃️ files. The files are here.

Vercel Daemons Coined in MIT in 1963, "we fancifully began to use the word daemon to describe background processes that worked tirelessly to perform system chores", inspired by Maxwell’s thought experiment of an imaginary agent in thermodynamics.

Vercel is the Vercel for Backends

RT Vercel https://x.com/i/broadcasts/1jGXgBlLjmoKZ

Fast approaching https://vercel.com/ship

Yes

See you tomorrow at https://vercel.com/ship… Back to where it all began

This is one of the most exhilarating times in technology, and this challenge was yet another jaw-dropping moment. When the details of OpenAI’s agent escape and Hugging Face hack started to emerge, it became clear that agent containment would be one of the most important priorities for any AI infra company. There are two classes of escapes: ① "escape the network" and ② “escape the matrix” 😁, as in, break the spell of virtualization. OpenAI was the former, where an agent not supposed to have internet access… gained it by hacking another system. Genuinely astonishing and worrisome. Through this challenge, however, we discovered something even more groundbreaking. A talented white-hat researcher, armed with a swarm of AI agents, expertise and perseverance, "pierced through" the Linux virtualization boundary that nearly every cloud sandbox provider in the world banks on today. I'm very thankful we got the report, were able to quickly move to protect our customers, found no evidence of customer data compromise, and collaborated with many industry partners to protect the wider internet. Our confidence and conviction in Vercel Sandbox are even greater now, after some of the smartest, most motivated researchers in the world, and their agents, gave it their best shot.

This is amazing 😂

Coding CLIs keep getting more complicated. Daemons, 3 layers of TUIs, client-server architectures. The whole point of a CLI is to deliver intelligence with zero friction, and be unix-like. 💾 Try http://fx.sh. It's 6mb, starts in milliseconds, model-agnostic. It's tastefully minimal: MCP, Skills, Subagents and ACP built-in.

Muse Wonder on Vercel AI Gateway Try with 𝚏𝚡 --𝚖𝚘𝚍𝚎𝚕 𝚖𝚎𝚝𝚊/𝚖𝚞𝚜𝚎-𝚠𝚘𝚗𝚍𝚎𝚛-𝟷.𝟶

We can do better than Rust

Incredibly fast progress by Meta

So cool. The future of learning

All plausible science fiction will be realized within our lifetime. If you’re reading this, you’re one of the luckiest humans to have ever been born (if we play our cards right.)

Some interesting machine & agent Vercel traffic stats. • Across the entire Vercel network, 58.18% traffic is bot-originated (last 30d). This was 32% in Jan 2024. • 60%+ of deployments on Vercel are now agentic, up from ~3% in Jan 2026. • Up to 83% of pageviews are now coming from agents to Vercel's own documentation sites. The % reliably increases the more we optimize content for them. I expect 'direct' human internet traffic to basically become a rounding error in coming years. The web will thrive, but it will be built for and by agents.

It's incredible to watch a new kind of economy be born. Agents purchasing infrastructure products and services. We've been shocked by how often they buy from our marketplace via our CLI already. We're now extending this to domains. Agents can now go full stack, from idea to online business, with a banger domain name 😁

Few things more enjoyable in work and life than discovering new talent. Seeing greatness in people, sometimes even before they fully see it themselves.

One thing the engineering community will (re)discover is that every program can be ① hardened (edge cases squashed, inputs tightened, errors handled) and ② optimized (profiled, benchmarked, rewritten) basically ad infinitum. There are real costs in every direction: time, attention, opportunity. You can literally explore input spaces that will never be hit or optimize performance of something that will never be used ("you have no users".) Even in an infinite token world, you need to know when to stop and what tradeoffs to accept. Agents will happily drill no matter what.

gm from the best city in the world

“Thinking fast and s̶l̶o̶w̶ a bit less fast under a confidence threshold.” A very simple feature that I think will be extremely impactful for at-scale AI decision-making.

At this rate AI will get us GTA 7 before GTA 6

Introducing gdp-ts: Ghosts of Departed Proofs for TypeScript. gdp-ts is a library, linter and AI skill for safer API design. Under this contract, sensitive functions require 'proofs' that the caller performed an authorization check. The typechecker verifies these proofs at compile time, preventing your team and agents from shipping catastrophic security (and other kinds of) bugs. While these patterns have existed for quite some time, especially in ecosystems like Haskell, ① human code review and ② cognitive and syntactic overhead made these solutions niche. The situation is now inverted. Agents are writing more code than we can review, and they *thrive* in tight loops with hard constraints that would frustrate us. We see this with the rise of Rust, borrow checker, code aesthetics debate and all. The README and examples model a real-world Vercel API product constraint: changing the password on a Project requires a proof of a certain role + a certain entitlement. Thanks to Matt Noonan and Ollie Charles for their research in this space. https://github.com/rauchg/gdp-ts

It’s hard to fathom http://fx.sh getting even faster, but it’s now… much faster 😂 As models speed up (as with Astra 𝚞𝚕𝚝𝚛𝚊𝚏𝚊𝚜𝚝), harness overhead matters more and more. Next release will feature even better performance for session storage and retrieval, and a big unlock for cloud durability in libfx.

DHH is fundamentally right about Rust. For context, Vercel has been undergoing a Rust-ification (carcinization, technically 🦀) for a while. One of the first projects we migrated was Turborepo, from Go to Rust¹. The migration completed, but the RoI was actually quite controversial internally. While Rust was in our eyes better for low-level OS access, something crucial for a build system like Turbo, the human migration costs were very sustantive. Go is very fast. It's beautifully designed. It's easy to iterate on. We were very conflicted about the migration, because it was *humans* writing the code, *even if we knew Rust was a better choice*. The calculus has now changed. What's "best for humans" is no longer necessarily "best for business". FWIW, it's also quite unlikely that Rust is the end-all-be-all toolchain. I'm quite certain there's greener pasture ahead, because Rust itself was designed before the 'supersonic tsunami' of agents hit. ¹ https://vercel.com/blog/how-turborepo-is-porting-from-go-to-rust

Working on a new little project. The 𝚁𝙴𝙰𝙳𝙼𝙴 is fully written by hand, because it's for human consumption. The documentation internals are AI English, because they're for agents. This little rule of thumb can make the world better. Blogs, tweets, READMEs: human communication. It's not even about "em dashes" or lack thereof. It's that it's the story-telling moment where you want to connect with other humans through their words.

Summer is finally here! 🌞⛱️

Security will become a larger and larger function in software companies. Security is verification engineering (eg: “my code is probably memory-safe”), as well as capital allocation (“what surface should I throw most tokens at?”) For startups, it’s a challenge and opportunity. As the world’s paranoia grows given ever-more-sophisticated AI adversaries, how can you trust a 2-person-and-a-dog company? But conversely, given the dire state of global cybersecurity and increasing reliance on centralization, as always, there’re vast areas that small teams can now disrupt that I’m excited to invest in.

AI will make everything free, including itself. The last domino to fall will be free energy, which is humanity’s final frontier.

We’ve confirmed a KVM 0day through our Vercel Sandbox bounty program. Affecting the industry’s gold standard solution for Linux virtualization. 2026 is wild! Thankful to Paulos and other researchers helping us make the most secure sandbox for agents. Full writeup coming.

Wonderful writeup on Jev + Python

Amazing. Love to see the https://muse.ai team ship so fast ▲

RT Vercel Engineers at @RogoAI build the company's internal apps on Vercel. They deployed over 73,000 times last month. Coding agents ship to production in 5 minutes, and run everything from churn risk analysis to the sales team's deal desk. https://vercel.com/blog/how-rogo-ships-agent-written-code-to-production-in-5-minutes-on-vercel

Re Here's the app. One thing I've been liking is to ask models to 'teach me back' what they did¹ by including "quines". Quines² are programs that output their own source code. So, in this demo you can reveal step by step the Svelte code powering it! https://sveltekit3-models-test.labs.vercel.dev/ ¹ Understanding is still great 😆 ² https://en.wikipedia.org/wiki/Quine_(computing)

SvelteKit 3 is a wonder. Congrats to the team. So excited for Async Svelte & Remote Functions to cook. I built a tiny app, it built & deployed in 15 seconds end-to-end, it’s instant, and fx + opus figured it all out effortlessly. https://svelte.dev/blog/sveltekit-3-is-here

The future is verification-engineering. Proofs, (e2e) tests, benchmarks, linters… Some tests will be deterministic, some agentic. This looks great.

The @MicrosoftAI team is training excellent models. Excited to bring them to @vercel, day zero.

Most token aggregators have extreme noise from ① token promos from/to providers that train on your data or ② providers that *claim* to be ZDR. Vercel AI Gateway is the largest trustworthy datasource to understand global AI token flows, because we have real usage, from real customers, an at-scale platform (400k+ paying customers and thousands of enterprises), and zero markup. Notably… every week we turn down "free token" promos from companies that come to us with dubious claims of Zero Data Retention (ZDR) and questionable reputations. We also don't get a kick out of having the "most providers". Most providers suck. You want the *best* providers. We put in as much work in engineering our gateway, as in our legal, compliance, privacy, and backoffice operations to responsibly serve trillions of tokens every day. We're here for the long haul.

If you run a service and want to get in front of 20M+ developers and the billions of agents they'll ship, add yourself to Connect. With code now being 'free', the final boss of building is connecting to stuff. Connect is both easier and more secure¹. For agents and for apps. ¹ The world of building agents is plagued with static credentials. You're supposed to give a bunch of keys to each agent or app you build. It's DX hell, but it's also a massive liability. Connect fixes this.

Faster

AI SDK has hit 30M weekly downloads. One-shotted this video in fframes + https://fx.sh + opus. I asked 𝚏𝚡 to look at https://npm.bet/ai and highlight the 5M/10M/20M/30M milestones. Yes, this video was written in Rust and rendered on the GPU. h/t @neogoose_btw

You can just ship things (and wallow in cartographic filigree)

This is just… beautiful

You can now search @vercel domains without auth. Especially great if you're an agent.

We migrated http://mercedesamgf1.com with as little intervention as possible, setting out to prove just how much faster @vercel is. ~70% faster builds and ~75% faster paints. We derived two AI skills from the migration that we'll be sharing back. It was a mature workload with a lot of "$formerProvider-isms", and yet the net of the work was done in under a week. If it's [built / rendered / shipped] fast, it's on Vercel.

I¹ ported my Mini web browser to Rust & Swift. It's faster, more secure, and shockingly, nicer to iterate on than Electron + Bun. I always wanted Safari-like UX but… Chrome 😁. Thanks to the 𝚌𝚎𝚏 crate, I can bundle up-to-date Chromium. 𝚏𝚡 𝚊𝚌𝚙 lets me embed an agent without bloating the app. It talks to my local fx CLI over ACP. fx can then manage the browser via an MCP server². By ditching Electron, I was able to get Liquid Glass, faster boots, and a more Mac-native feeling. Little things like fine-grained focus and input control, which are nightmare fuel in JS, just work®. Native is the future, both on the desktop and in the cloud. I suspect the entire software world will "nativify" faster than people realize, as DHH hinted at Rails World. Excited that Vercel's bet on Fluid will help support this, whether you choose to go Rust, Go, Zig, scriptc… ¹ Built on http://fx.sh with Opus 5.5 and Sol 6 ² In the process I learned about http://agentclientprotocol.com/rfds/mcp-over-acp, which I'm now really excited about. This will allow a more secure and direct way to expose MCP tools

Hello world, Lisa Rauch 👧. What an amazing and beautiful world to be born in.

Reject non-understanding. Slop grenades are not just code and PRs. We run a real risk that *reading* gets entirely discounted, because of the exhaustion of getting low-quality and unverified AI prose being thrown at you all the time. There's a thread going viral about a 'perf improvement' attributed to a compiler change. The PR *description* itself has the AI telling this person it's *not* due to the compiler change, it's how the algorithms and data structures have changed. And the implication is that a similar win could have been attained otherwise. I want AI in the service of understanding the universe and enhancing human cognition and creativity.

Cool https://jev-pokemon.vercel.app 😂

We're helping world-class organizations like @klaviyo build agentic deployment platforms. ① Connect every agent (claude, codex, cursor…) ② Configure SSO via their IDP (Okta, Entra) ③ Now everyone can cook, securely® What happens to SaaS once every company sets this up? I get this question a lot 😁 First, the data must come from somewhere. These apps only work if they can be infused with business data. This is why we see the big enterprise SaaS players suddenly prioritize shipping CLIs and MCPs. Or dust off APIs they were neglecting for years. The new 'procurement bar' will be how ergonomic your product is for *agents*, rather than humans. How easily they can navigate your ontology and work with your data. Once that's in place, there's a long tail of SaaS applications that I suspect will never be bought again. They'll be generated. They'll be more secure, more performant, more modern, and tailored to each company's and employee's needs.

The growth of https://skills.sh has been mind-boggling. I watched it grow from idea, to first ship, to 𝚗𝚙𝚡 𝚜𝚔𝚒𝚕𝚕𝚜 being on every README of the internet. We used to write code, now we write English.

Spend of OpenAI vs Anthropic vs Open (Vercel AI Gateway, last 2 months) • Anthropic still #1 in spend, but went 69% → 40% • OpenAI: 10% → 24% in spend • GPT-6 Astra + GPT 5.6 Sol are ripping • OpenAI now leads in tokens # • Kimi K3 + DeepSeek took ~half of Anthropic's loss • Opus 5.5 is up to 10% of spend in 2 days • OpenAI is 62% of image generations Watch here: http://token-race.vercel.app

ShellPerfBench. I'm really neurotic about the startup time of a new shell session. Opus 5.5 found a lot of really great optimizations other models missed. Anon, anytime you open a new terminal, you might be silently suffering. Tell your agent to optimize .𝚣𝚜𝚑𝚛𝚌 and friends

Muse, Instinct, OpenClaw, Claude Code… All successful agents have 3 key components: 🧠 Brain → model, harness (logic) 👐 Hands → tools, computer, browser 🗃️ Files → memories, skills, repos The 'easy' way is to throw all these in 1 stateful computer (a Mac Mini) Like, you run 𝚌𝚕𝚊𝚞𝚍𝚎 or 𝚏𝚡 in your mac, you keep it running all day with 𝚌𝚊𝚏𝚏𝚎𝚒𝚗𝚊𝚝𝚎, it has storage, and CLIs and apps installed. But if you want to cost-efficiently run agents in the cloud, you actually start breaking down these parts. 🧠 The harness can run in Fluid compute. To make it reliable across restarts, rollouts, crashes, you make its event log durable using Workflow. 👐 The hands can be a dedicated browser fleet like Browserbase/Kernel, a computer like Sandbox, and even more efficient lightweight tools like just-bash. 🗃️ 🆕 What was missing was a way to also decouple storage. Imagine you want to run a memory consolidation cron job every night ("dreaming"). You can read/write to the files directly without 'booting up' the agent's full computer. Today we're introducing the perfect companion to Sandbox: Drives. We shipped the computer for agents, now we're giving you the 'external disk' you can attach at will. It's early, and we'll be expanding capabilities here quickly. Btw, breaking apart the agent into these independent parts not only optimizes costs in a big way, it also *massively* improves security and auditability. I'd argue you can't even run a secure agent otherwise!

Software will never die again. You liked Google Reader? Cool, you can generate and deploy your own. Yours, forever.

We ran fresh Next.js evals. The tally: ① Opus 5.5 [𝟿𝟽%] ② GPT 6 Sol [𝟿𝟽%] ③ Fable 5.1 [𝟿𝟽%] ④ Grok 4.7 [𝟿𝟺%] Notably, Grok is 2x-7x cheaper

Love the tastefulness with which anthropic ships. Main reason I got excited about 'headless' for the web is that any given page could take whatever whimsical or unique shape you wanted. With AI, there's no excuse not to push the design frontier. This is @nextjs at its finest: http://anthropic.com/claude-opus-5-5

I'll be interviewing @tobi on stage at Vercel Ship SF on October 15. We'll be discussing agentic commerce, native rewrites, S3 WALs, slop grenades, and more 😂 Can't miss this convo with one of the all-time greats https://vercel.com/ship

In addition to our type-safe TypeSafe @aisdk API (get it?), you can now use Jev over HTTP in AI Gateway

Game recognizing game

Grok 4.7 on Vercel 40% off Fast, smart, cheap. Pick three Try it on Gateway with https://fx.sh or @eve

The thoroughness with which agents can test and QA software is unrivaled. I pointed out something wasn't rendering right in a mobile (in-app) browser. The thing goes to the depths of Mordor to reproduce, simulate, fix, deploy, verify. It created an ephemeral vercel deployment to throw at an iPhone simulator! The absolute grind on these things. Humans could simply never match this level of intensity. We'd run of energy, we'd extrapolate, we'd hope it'd work out. The software of the future will be of a quality and performance we've never experienced before. It's amazing to watch this shift.

https://v-prism.vercel.app

“I've spoken of the shining city […] a tall, proud city built on rocks stronger than oceans, wind-swept, God-blessed, and teeming with people of all kinds living in harmony and peace; a city with free ports that hummed with commerce and creativity. And if there had to be city walls, the walls had doors and the doors were open to anyone with the will and the heart to get here. That's how I saw it, and see it still.” – 🇺🇸 Ronald Reagan (January 11, 1989)

The coolest thing about the recent wave of AI is that everyone is excited about building. All short-term drama aside, this is an immeasurably good development for the world. The people want to create and ship, not just consume.

Looks like today may be a record day for token volume % of open models on Vercel AI Gateway: 🟦 Open 78.4% 🟨 Closed 21.6% While spend 💲 usually tells a different story, #3 and #4 today are Moonshot AI & DeepSeek. Adding Z⁠.ai, their combined spend surpasses OpenAI (#2). (Do note that's the spend for inference of the model across providers (mostly in the US), not revenue going directly to the open weight labs.)

Free Jev on Vercel I know what I’m doing this weekend 😁

The data and the anecdata on Jev's adoption are shocking. Everyone is adopting it. I think it's a great product, but this is also downstream of the "AI is too expensive/slow" zeitgeist. People are eager to optimize and put AI in even more places! https://vercel.com/blog/ai-gateway-jev-model-launch

Jev Generative UI has been achieved externally

Vercel is the Vercel for Java

It's likely that more software will be produced next year than in all of the history of computing. It took Vercel 10 years to get to 1 billion deployments, but then we added another 1.4 billion in the last 10 months alone. Some of this will be tiny, personal and even disposable software. An HTML file with some javascript ("artifacts"), a report, a pricing calculator, a slide deck, all the way to sophisticated apps, agents and platforms. Deployment speed is one of our obsessions. We've brought down the time to deploy, upload, assign a domain and globally propagate an artifact to 1 second. This is quite a feat considering we serve these artifacts with uncompromising performance and security. This sub-second deployments include our global CDN, Firewall, immutable deployments, domain assignment, observability, rollbacks… And we're not done yet. Follow @melkeydev for more.

This is really cool. When you pressure your agent to ship a hotfix really fast, it can now 𝚟𝚎𝚛𝚌𝚎𝚕 --𝚝𝚞𝚛𝚋𝚘 --𝚙𝚛𝚘𝚍 😊 It'll use the fastest available build machine. h/t @garberchov for tweeting this idea to us

RT TypeSafe AI Jev is now available on the @vercel AI Gateway https://vercel.com/ai-gateway/models/jev

We're seeing extraordinary results from @typesafeai. Default mode in 𝚏𝚡 is auto, with a safety reviewer analyzing every command. That reviewer runs on GPT Luna today. Jev is up to 18x faster (p95) *and* more accurate. It's coming to @vercel AI Gateway and likely new default.

Secure builds (in-VPC / VPN / static IP) are now 64% faster to start. Given how common data access at deployment time is on Vercel (e.g: for pre-rendering), we developed a capability for our build microVMs to join secure networks. This is now way faster!

Safari 27 supports JSPI, a magic WebAssembly capability for sync native code to suspend on an async 𝙿𝚛𝚘𝚖𝚒𝚜𝚎. On http://fx.sh, libfx uses the built-in browser 𝚏𝚎𝚝𝚌𝚑 stack to make API calls. As more code goes native, I expect WebAssembly to play a huge role in the future of the web.

Excited to formally introduce Vercel Labs. http://vercel.com/labs is the home for the in-public research and experimentation arm of Vercel. 247 million downloads in, we wanted to, also in public, share what we're supporting, what we're researching, and what experiments didn't pan out. Kudos to @ctatedev and @cramforce for shaping this great initiative and iterating in public.

The future is multi-model. Trying to hide the choice confuses and hurts customers, who can't participate in the upside of the most exciting market competition of our times, nor master the best tool for the job.

One of the coolest things about 𝚏𝚡 is that it auto-upgrades, and 𝚌𝚝𝚛𝚕+𝚐 restarts + resumes the chat. Also, that 0.0.10 is much, much faster for longer sessions 😊 https://fx.sh/changelog#v0.0.10

Agents are only as good as the proof-checkers, compilers, type systems, and linters you give them. 𝚜𝚑𝚊𝚍𝚌𝚗/𝚕𝚒𝚗𝚝 helps agents stay on track with the rules of your design system. Meta: verifiers + skills are the new 'frameworks'!

Welcoming Steren, creator of Google Cloud Run, to Vercel. He will lead the Fluid family of compute products (Functions, Containers, Sandbox, Builds). Serverless was the last chapter of the cloud, and Steren helped define the paradigm at Google. Agents are the next frontier, and they require new compute primitives designed for them. Excited for Steren to lead this transition once again.

Skill-driven development

You don't get it, I build software to relax

I’m seeing teams at Vercel iterate just as fast on Zig, Go, Rust projects as TypeScript & Python ones. The days of language or runtime choice based on human convenience are over. Agents are the new compilers. They compile intent into fast software.

http://fx.sh can now orchestrate subagents with different models & reasoning efforts. e.g: Fable planning and Grok executing. Fable is a genius, Grok is a fast workhorse. ① Simple. 𝙰𝙶𝙴𝙽𝚃𝚂.𝚖𝚍 or your prompt can indicate this preference. ② Steer as you please. It’s extremely enjoyable to use. You just talk to the agent and interrupt at will. Worth noting this works with any model, any gateway. No fancy server-side routing. Just harnessing (no pun intended) the model’s intelligence and ability to orchestrate.