Guillermo Rauch

Glimpse of a world of fully generative interfaces. AI → JSON → UI: http://github.com/vercel-labs/json-render

We're encapsulating all our knowledge of @reactjs & @nextjs frontend optimization into a set of reusable skills for agents. This is a 10+ years of experience from the likes of @shuding, distilled for the benefit of every Ralph

If you're an engineer, 2026 is the year when you return to the fundamentals. Unix, CLIs, tests & types, markdown. Great and accessible documentation. Did I say markdown. There'll be lots of markdown. You can now 𝚌𝚞𝚛𝚕 -𝙷 '𝚊𝚌𝚌𝚎𝚙𝚝:𝚝𝚎𝚡𝚝/𝚖𝚊𝚛𝚔𝚍𝚘𝚠𝚗' the @vercel docs: Which agents like Claude Code do by default, saving lots of data transfer and tokens in the process:

There's an app I use regularly that's defective. I'm weighing in my mind whether I should send feedback and hope it gets attention from its creators, or re-build it with AI from scratch. This is a tiny app so it's plausible for me to do it. But if you're in the business of selling software, this is how your every customer is thinking now, or how they'll be thinking soon. Iteration velocity matters more than ever before. How quickly you fix, improve, and ship is your counter-signal.

"What did your agents get done this week?" is the new perf review

There's definitely a non-zero chance we get "generate your own GTA6 in a few minutes" before GTA6

Beautiful Rust CLI for browser automation by @ctatedev. Agents love CLIs so I expect this to gel really well with Claude Code, Codex & OpenCode for OS ai automation. https://github.com/vercel-labs/agent-browser

10 days into 2026: - Terence Tao announces GPT & Aristotle solve Erdős problem autonomously - Linus Torvalds concedes vibe coding is better than hand-coding for his non-kernel project - DHH walks back “AI can’t code” from Lex podcast 6 months later An acceleration is coming the likes of which humanity has never experienced before

2026 is the year of Brazilian Jiu-Jitsu

In a world of self-improving software, nothing feels more dated than "click here to update the app". This will become more obvious as the rate of software production continues to increase. The only viable software delivery model is the web's: "Hot app replacement".

2026 off to an awe-inspiring (and vertiginous) start with @OpenAI GPT 5.2 Pro and @HarmonicMath near-autonomously generating a proof to an Erdős problem https://mathstodon.xyz/@tao/115855840223258103

Betterment's email notification system was hacked, and they issued a "Bitcoin & Ethereum giveaway". Spam and phishing filters need to get more sophisticated with AI. The email indeed comes from Betterment, but it's *obviously illegitimate*. If we give Claude Code the prompt to analyze the raw source of this email, without any additional hints… it perfectly diagnoses it: > "This is a SCAM email. Do not send any cryptocurrency to those addresses." And it even makes a very sound hypothesis of how the hack happened: > The email passes DKIM/SPF/DMARC for e⁠.⁠betterment⁠.com uses Braze/Sparkpost infrastructure, suggesting either: - Betterment's email marketing account was compromised - A third-party with access to their marketing tools sent this Truly amazing.

2026 is the year of the @x Article

I like that the word 𝑡𝑒𝑐ℎ𝑛𝑖𝑞𝑢𝑒 in French means both a skill and a technology. Because with artificial intelligence, every human skill, like drawing, programming, story-telling, is destined to become a technology.

Very proud of the @nextjs and @reactjs communities. To power the dissemination of a guideline that will impact the lives of every American… feels special. And to do it in a way that's fresh, beautiful, and iconic. Kudos @ndstudio on http://realfood.gov 🇺🇸🫡

The "Delete and Report Junk" button on iMessage

Vercel will be officially sponsoring http://tailwindcss.com. That's a given. We as a community and industry owe @adamwathan and team a lot. Tailwind is foundational web infrastructure at this point (it fixed CSS 😉). I've also reached out to Adam to explore how we can make this a longer-term commitment.

All software will be generative and generated. Adjust accordingly.

RT Daniel Lurie 丹尼爾·羅偉 This week I stopped by Vercel’s expanded headquarters on Mission Street to mark a major investment in downtown San Francisco.   After sitting vacant for more years, this space is now home to growing teams. Vercel’s expansion reflects that San Francisco remains the epicenter of innovation—and why companies are choosing to grow, hire, and build the future right here. Original tweet: https://x.com/DanielLurie/status/2009085138223952151

Coding agents make a lot of mistakes. They're smart enough to correct them, but they waste time & tokens. @v0 is focused on extremely fast iteration loops and minimizing the number of errors we render. We're sharing the model architecture to fix this:

Chat with the cloud

The primary lesson from the actually successful agents so far is the return to Unix fundamentals: file systems, shells, processes & CLIs. Don't fight the models, embrace the abstractions they're tuned for. Bash is all you need.

The new onboarding ux: copy a prompt. Paste it to your agent. http://vercel.com/sandbox

wow

What are the most precious attributes of a production system? Security & privacy, durability, availability. (Not being hacked, not losing data, not going down) AI is an enormous opportunity, but also a threat. To answer it, Vercel is building anti-fragile self-driving infrastructure. You should embrace that many people at your company will be generating and giving new programs every day, likely multiple times a day. But you need a secure environment to deploy them to. That’s @vercel. We’re dogfooding this every day because every single employee is now a net producer of software. From the HR team to finance to data to PM. They can ship with confidence thanks to secure defaults, like internal SSO & zero-trust and pervasive encryption, serverless cost efficiency for dormant apps, AWS VPC integrations, and more.

As technologies like v0 and Claude Code continue to proliferate, expect the rise of latent technology traditionally encumbered by the UX layer. Linux desktop, crypto systems, decentralized infrastructure…

iPad games for kids are now ad-ridden brainrot. But AI can now give us unlimited quality game assets and code. Would happily pay a subscription to a universe of games that teach children math, logic, language in engaging ways, and crucially, AI and prompting. The best way to equip future generations is to make them AI-literate from the outset, so they can understand what’s possible and how it shapes the world around them.

🌉 We were just joined by Mayor of San Francisco Daniel Lurie in opening our new (and 4x bigger) Vercel SF HQ We're very thankful to get to live and work in this glorious city, that's been at the frontier of invention time after time Signed, an eternal SF supporter & optimist!

How does this game *run forever* on Vercel, automatically retry failures, time out unresponsive LLMs, stream updates…? Enter: Workflows ▪️Each ♛ game is a "𝚞𝚜𝚎 𝚠𝚘𝚛𝚔𝚏𝚕𝚘𝚠" ▪️Each ♞ move is a "𝚞𝚜𝚎 𝚜𝚝𝚎𝚙" It's just JavaScript™. Opus picked up immediately, even with no exposure during training (given its recency). Great validation of its design. The local DX, testing story, and the built-in observability with tracing is *chef's kiss* 👇

Deployments on Vercel on December 2025 traffic grew at nearly 18x the rate of Dec 2024. Any guesses why?

· ✶ ✢ ✽ Discombobulating (on @vercel AI Gateway)

This is fun. Left the autonomous chess match (http://v0-chess-match.vercel.app/) running overnight between @xai Grok 4 and @openai GPT 5.2. Grok's 𝚐𝚛𝚘𝚔-𝟺-𝚏𝚊𝚜𝚝-𝚛𝚎𝚊𝚜𝚘𝚗𝚒𝚗𝚐 won 19 of the last 20 matches 😅 What models should battle next?

RT cam worboys Fresh http://cash.app website Original tweet: https://x.com/camworboys/status/2008002646410703282

No one could have predicted this is what coding AGI would end up looking like

It's an amazing time if you know what you're doing and if you don't know what you're doing

2026 will mark the year when humanity achieved digital abundance. Any thought and desire instantly transmogrified into software. Apps, games, images, music, docs, sites, every one generated, by anyone.

17 rules for successful agentic coding (also known as the unix philosophy)

I've got Claude Code running through of @vercel AI Gateway 🤩 This not only gives me failover (e.g.: to @awscloud Bedrock), but also means your company can manage all models & spend in one place. How? Set 𝙰𝙽𝚃𝙷𝚁𝙾𝙿𝙸𝙲_𝙱𝙰𝚂𝙴_𝚄𝚁𝙻 and 𝙰𝙽𝚃𝙷𝚁𝙾𝙿𝙸𝙲_𝙰𝚄𝚃𝙷_𝚃𝙾𝙺𝙴𝙽 to your AI Gateway key, then /𝚕𝚘𝚐𝚘𝚞𝚝. e.g.: ANTHROPIC_BASE_URL="https://ai-gateway.vercel.sh" ANTHROPIC_AUTH_TOKEN="vck_…" ANTHROPIC_API_KEY="" It's experimental. Let me know if you run into any issues!

A 10x-30x faster 𝚜𝚟𝚎𝚕𝚝𝚎-𝚌𝚑𝚎𝚌𝚔 in Rust just dropped, built with Claude Code and Codex CLI by @pheuter. The motivation was to make agents themselves more efficient. 2026 is about to be wild 😄 https://svelte-check-rs.vercel.app/

The fact that a Google engineer is able to appreciate other AI labs' tooling is actually a bull signal. The worst attitude a company could have in this AI race is "my way or the highway". My message to Vercel right now is "try everything, don't get too attached to last week's tools and models".

I expect the web to become a lot more whimsical in 2026. Cost of personalization is plummeting with AI. Your wildest dreams and visions can be turned into pixels now. A renaissance is imminent.

What would you like to see in the world this year?

I built an infinite AI chess game, powered by: ▪️ AI SDK for a single DX to all frontier models ▪️ AI Gateway for seamless access with 1 API key ▪️ Workflow to keep the game going forever Right now Anthropic is playing OpenAI: http://v0-chess-match.vercel.app

The fundamental coding agent abstraction is the CLI. It’s not a UI or form factor preference, it’s rooted in the fact that agents need access to the OS layer. Coding agents are, at their core, computer-use agents. They run programs, create new ones, install missing ones, scan the filesystem, read logs. More than text editors, they’re *automating your computer* at a low level. It’s more accurate to think of Claude Code as “AI for your operating system” than a continuation of the copilot autocompletion → IDE AI assistance paradigm. Another consequence is that it works for any computer. Your desktop computer, but crucially, also the ones running in the cloud. 𝚗𝚟𝚒𝚖 nerds like yours truly always liked that it’s an editor that runs on your Mac but also any machine you 𝚜𝚜𝚑 into. Imagine if you could launch a million Claude Codes concurrently to tackle a bug, find cybersecurity threats, process your issue backlog, build features based on user feedback, run QA… That’s why we’re building http://vercel.com/sandbox as the infinite compute layer for agents.

Happy new year! 🖤 May the new year bring wonderful ships

Someone named "Vercel" was "building" my "repo"

🚠

Narrative violation: AI coding will result in the most rigorously tested, type-checked, and provably correct code in the world. First, the testing and verification feedback loop helps agents and prevents slop. Second, tests have always been a chore, but who doesn't love ✅ a high test count and checkmarks? Agents are glad to write as many tests as we want.

First wave of AI: know it all Second wave of AI: do it all

Beautiful 𝟻𝟸𝚜 → 𝟷𝟽𝚜 and very significant perf wins coming across the board (filesystem, network, compiler, and compute)

SF when the sun comes out

Holiday hack: Typing Stats on my menubar ▪️ SwiftUI w/o Xcode ▪️ iCloud persistence ▪️ CRDT GCounters for convergence Motivation: how much I input on a given day is a good proxy for productivity / focus. I look forward to looking back on how much I type over the years 😄

Merry Christmas! 🎅🌲🎁 What are you hacking on during the holidays?

We've completed the acquisition of the 𝟷𝟺𝟹.𝟷𝟹.𝟶.𝟶/𝟷𝟼 and 𝟷𝟻𝟻.𝟷𝟸𝟷.𝟶.𝟶/𝟷𝟼 IPv4 blocks for @vercel CDN and Fluid compute. Fun fact: these are a piece of internet history, which we acquired from P&G, as part of the "legacy space". ARIN legacy IPv4 ranges are address blocks originally issued before the inception of the regional Internet registry system in 1997. These larger blocks were originally given to corporations like Procter & Gamble and Ford (!) with fewer restrictions and regulations. These days the world is transitioning to the much larger IPv6 space, but IPv4 is still incredibly valuable for interoperability with billions of devices and huge amounts of infrastructure. High quality and high reputation IPs like these are scarce and expensive. They're like a rare, borderless PSA 10 Charizard. Each day there's fewer of them. Security and availability are the top priority at @vercel. Our teams do a lot of work behind the scenes to make sure your visitors get the best global access to your projects, with your infra on autopilot 🫡

RT Pontus Abrahamsson — oss/acc Midday codebase now ~11× faster compile time thanks to Next.js 16.1 Original tweet: https://x.com/pontusab/status/2003518282377372142

Our d0 agent has completely change the way I work. I can't imagine doing my job without this capability. Imagine: ⟩ 𝙷𝚎𝚢 @𝚍𝟶 𝚝𝚎𝚕𝚕 𝚖𝚎 𝚝𝚑𝚎 𝚅𝚎𝚛𝚌𝚎𝚕 𝚛𝚎𝚐𝚒𝚘𝚗𝚜 𝚠𝚒𝚝𝚑 𝚖𝚘𝚜𝚝 𝙵𝚞𝚗𝚌𝚝𝚒𝚘𝚗 𝚒𝚗𝚟𝚘𝚌𝚊𝚝𝚒𝚘𝚗𝚜? The agent is extremely effective thanks to our semantic layer defining the 'vocabulary' of how we reason about our business. If you don't have something like this in your @slackhq, read on, or reach out to @andrewqu to learn how we built d0 on top of Vercel's AI infrastructure. 📃 http://vercel.com/blog/we-removed-80-percent-of-our-agents-tools

Re Companies like @Reuters and @Clay have adopted @aisdk as their foundation to build agents

Within minutes of @zai_org announcing their frontier coding and reasoning model, 𝚐𝚕𝚖-𝟺.𝟽 is available on http://vercel.com/ai-gateway

A CDN that you can program with TypeScript

Amazing accomplishment by the @nextjs Turbopack team with 16.1

Following React2Shell, we had the privilege of working with the world's best security researchers on strengthening our network and runtime defenses. WAFs are broadly considered a 🩹 "bandaid" because they can be trivial to bypass, or have no insight into application and framework level protocols. Not only did we mitigate an extraordinarily broad range of attacks, we taught our WAF to understand the Flight protocol, and shipped defense-in-depth mechanisms to patch our compute layer automatically. In other words, even if you bypassed our WAF, our customers were protected. As a result of this incident, we now have the infrastructure in place to mitigate extremely complex and sophisticated CVEs. React2Shell is informing design improvements to our Firewall and Fluid that we think will push the frontier of defensive cybersecurity.

Incremental caching is now enabled by default for 𝚗𝚎𝚡𝚝 𝚍𝚎𝚟. When agents create new projects or open existing ones, Next will feel instantaneous. For tools like @anything, @orchidsapp and @v0, this means scaling from prototypes to projects of any size.

RT Belce Doğru Pattabi Grok is now the top model in @vercel's AI gateway, beating the next model by 48%! 🚀 Original tweet: https://x.com/belce_dogru/status/2001288752518328671

Cline is now powered by Vercel AI Gateway. Compared to their previous router, they're now seeing 44% fewer errors and 14% faster streaming. A few days ago, a major LLM provider had a severe outage. We automatically routed requests to healthy endpoints and recovered precious tokens. https://vercel.com/blog/cline-on-ai-gateway

Check it out here: http://localhost:3000

There are no limits anymore. Anyone can do anything. The only limiting factors are agency and ambition. Never has a college degree, work experience, network, even the accumulation of knowledge been worth less. You can just ship things.

RT shadcn Introducing shadcn/create – Build your own shadcn/ui Customize Everything. Pick your component library, icons, base color, theme, fonts and build something that doesn’t look like everything else. Now available for Next.js, Vite, TanStack Start and v0. Original tweet: https://x.com/shadcn/status/1999530406744293593

RT shadcn Launch day. It's a special one. Original tweet: https://x.com/shadcn/status/1999467787669164285

Cool @vercel × @shopify store 👀