Guillermo Rauch

“Feedback is a gift” has always been a favorite mantra of mine and a strongly-held belief. Now it’s fact. Each piece of feedback is someone gifting you a prompt for you to give your agents to improve your product ☺️ Thankful for everyone who takes time (or tokens) to critique our products, write to us or forward your agents’ transcripts.

If you intern at Vercel, you can work on incredible projects like this one. Improving Next.js chunking can yield massive efficiency improvements at internet scale. Excellent writeup by @sam_poder

Run ▲ ~/ vercel ai-gateway coding-agents setup Points all coding agents to AI Gateway, getting you 100% uptime, observability, budgets, and ease of switching.

It took blood, sweat, and tokens

Follow @eve

Everything is code

It's truly the best time to build… 50% off on Vercel AI Gateway right now: ▸ 𝚐𝚙𝚝-𝟻.𝟼-𝚜𝚘𝚕 ▸ 𝚐𝚙𝚝-𝟻.𝟼-𝚜𝚘𝚕-𝚏𝚊𝚜𝚝 ▸ 𝚐𝚎𝚖𝚒𝚗𝚒-𝟹.𝟾-𝚏𝚕𝚊𝚜𝚑 [𝚗𝚎𝚠] ▸ 𝚐𝚕𝚖-𝟻.𝟹-𝚏𝚕𝚊𝚜𝚑 ▸ 𝚖𝚒𝚗𝚒𝚖𝚊𝚡-𝚑𝟹 ▸ 𝚖𝚒𝚗𝚒𝚖𝚊𝚡-𝚑𝟹-𝚖𝚊𝚡 https://vercel.com/ai-gateway

Excited to partner with and support @tannerlinsley and team, champion of React, open source, and the web What this means for our customers: whether you choose @nextjs or @tan_stack, we are committed to the highest quality service and support.

Fable 5.1 now on @vercel AI Gateway Try it with http://fx.sh. Every model at your fingertips 🥹

Fluid is the thing that has enabled Vercel's: ◾ Industry-leading Build performance ◾ Sandbox reliability & massive concurrency ◾ 30min Function durations (& soon longer) This unification makes the best DX possible. Shared 𝙳𝚘𝚌𝚔𝚎𝚛𝚏𝚒𝚕𝚎, security perimeters, networking, file systems, across all compute products. You already see this in some pretty unique Vercel capabilities, like our Builds and Functions being able to participate in the same Secure Compute boundary and config. Imagine the same egress Firewall features for Sandbox that prevent agent escapes, applying to your Builds, Functions and Servers. The future is one global and unified compute platform, with only known physics as the limiting factor 😊

Vercel 🤝 AWS

Your next design system is… Markdown. We wrote about how 𝙳𝙴𝚂𝙸𝙶𝙽.𝚖𝚍 is helping solve the hardest problem in AI today: slop. And how you can truly, finally scale design taste within a large organization.

Coding tokens are basically infrastructure. Yet, the way companies have approached them amounts to: "Hey all, here's an AWS key. It can spin up anything from a 𝚝𝟹.𝚗𝚊𝚗𝚘 (＄𝟹.𝟾𝟶/𝚖𝚘) to an 𝚙𝟻.𝟺𝟾𝚡𝚕𝚊𝚛𝚐𝚎 (＄𝟺𝟶𝚔/𝚖𝚘). Go wild" And obviously the reason companies don't do this for their core infra is that you need to govern, optimize, and observe usage and costs. The intern might decide to host a static website in that 192 vCPU machine, and you won't be very happy about the bill. AI Gateway fixes this for all your tokens, with per-key, and now per-user budgets.

Digital abundance

Cool

When something goes viral on the internet, especially on X, it attracts all kinds of attention. Including some of the largest botnets on the internet launching targeted DDoS, from hijacked devices and residential networks. Extremely proud of the Vercel CDN & Traffic Security teams, who protect 400,000+ customers and tens of millions of Vercel users 24x7x365.

Work so hard you can buy a Pikachu Illustrator PSA 10

Two days ago JFrog dropped CVE-2026-82329, a critical authentication bypass in Artifactory. It’s a CVSS 9.8, a disastrous vulnerability score. It’s like a 9.8 earthquake on the seismic scale. It affects default configs, requires no auth, no user interaction. It’s an RCE bomb because Artifactory hosts binaries, so you can basically poison everything, but an admin escalation can cause damage even beyond that. When the OpenAI / Hugging Face news came out of agents discovering zero-days, I was wondering if it was marketing-speak or reality, because I hadn’t seen a CVE filing. Now it’s here: https://www.cve.org/CVERecord?id=CVE-2026-82329. I don’t see any official confirmation that it’s indeed the case, but one can speculate this is what the agents discovered and exploited. I’d previously written that it was obvious agents could help in finding serious vulnerabilities *alongside humans*, but exploiting them autonomously was a bridge not yet crossed. It seems like we’re now there. Our guidance for this new world: assume everything hackable will get hacked. And it will get hacked autonomously. You must also defend yourself autonomously, because your surface of attack is likely bigger and your code more vulnerable than you expect: https://vercel.com/blog/everything-hackable-will-get-hacked

I read every DM people send me, even if I don’t reply to all. I’m constantly learning about new ideas, ways to improve our product, where we’re falling short, who could join our company next, where to invest… very grateful for the @x platform.

http://fx.sh 0.0.7: - hugely improved MCP support, eval'd with top MCPs like @context7ai, @datadoghq, @mongodb, @linear, @notionhq, @supabase, @neondatabase, @stagehanddev, @usekernel… - more minimalist: fewer tools and improved shell execution. One of the trickiest things about CLI agents is that running commands is actually not that simple. e.g: commands might hang or run for very long, their output needs to be monitorable by the agent, they might produce huge amounts of output, or none. This is an area where 𝚏𝚡 really shines and will continue to get better.

I see the web evolving in two extremes: 1️⃣ ultra-fancy human experiences, in the service of delight, exploration, brand, entertainment. http://vgpu.sh / threejs / typegpu are enabling this 2️⃣ agent-centric content, data and APIs. the opposite end of the spectrum: pure, lossless signal for machines. The rise of markdown and MCP are the tip of this iceberg The 'in between' is likely to be subsumed by agents producing just-in-time UI. When you need utilitarian content or data, it'll be generated just-in-time for you. Think of the agent as the new browser in this model.

MCP is growing explosively. And it's not just Vercel MCP… Look at 𝚖𝚌𝚙-𝚑𝚊𝚗𝚍𝚕𝚎𝚛 npm downloads, which you can use to implement MCP servers

There are many “easy” ways to create agents out there. But how many that give you a Git repo where you actually *own* the entire intelligence stack? Runtime, model choice, skills, tools, connectivity, sandbox… Easiest way to start with eve: https://vercel.com/new/agent

As models get more performant, UI gets more generative and dynamic. Excited for this trend to continue

Deploy to Vercel

https://vgpu.sh and shaders are further proof that “everything is computer”. 2D, 3D, geometry, light, materials, textures, shadows, reflections, particles, post-processing. Just programs evaluated massively in parallel over vertices and pixels. All computer.

Our best products have been exports of internal technology we gain strong conviction in. @matinotfound and team have been shipping extraordinary WebGPU creative productions, and now http://vgpu.sh is here. It’s a fully agent-native devtool: it was designed for agents, not humans alone. Much like 𝚊𝚐𝚎𝚗𝚝-𝚋𝚛𝚘𝚠𝚜𝚎𝚛, it’s part of a new exciting generation of tools to serve the new world.

CMA is a remarkably elegant architecture. The 'hard parts' of the agent are managed, but it doesn't take away from your freedom to customize. e.g: you can have a fully custom UI for your agent on @vercel with Chat SDK. Or handle custom tools with your own functions on Vercel.

So cool. Global compute at your (agents') fingertips. Multiple regions, failover, up to 10,000 concurrent sandboxes & 5,000 vCPUs/min ramp by default, with more regions to come 🫪

We're shipping a cool security dashboard, but also the 𝚟𝚎𝚛𝚌𝚎𝚕 𝚜𝚎𝚌𝚞𝚛𝚒𝚝𝚢 𝚌𝚑𝚎𝚌𝚔 CLI. Much like 𝚒𝚜-𝚊𝚐𝚎𝚗𝚝𝚒𝚌, you can call on agents to improve your security posture with human-in-the-loop, or run in crons. Who would have thunk CLIs would save the world 😁

The hardest problem in building agents is secure connectivity to services and data. Excited for Vercel Connect to be GA. Run e.g: 𝚟𝚎𝚛𝚌𝚎𝚕 𝚌𝚘𝚗𝚗𝚎𝚌𝚝 𝚌𝚛𝚎𝚊𝚝𝚎 𝚗𝚘𝚝𝚒𝚘𝚗, then get an MCP client you can query on behalf of the authenticated user.

AI Gateway can now call webhooks after long generations 😍 ① Lower compute costs (no long-running process waiting idly) ② More reliable. Gateway can take more of the processing burden ③ You can 𝚊𝚠𝚊𝚒𝚝 𝚑𝚘𝚘𝚔() in "𝚞𝚜𝚎 𝚠𝚘𝚛𝚔𝚏𝚕𝚘𝚠" to sanely consume these

One of the coolest things about gateways: managed tools. If you add 𝚎𝚡𝚊_𝚜𝚎𝚊𝚛𝚌𝚑 to your 𝚝𝚘𝚘𝚕𝚜, you give zero-setup, fast Exa search to your agents. No credentials, accounts, connections to set up, plus unified billing.

We’ve seen ISPs and telcos unjustly censor domains and even entire TLDs. A big carrier recently deemed the entire .𝚝𝚘𝚘𝚕𝚜 gTLD is “sus” so they broke every TLS handshake, and thus, all websites on it. Vercel now has another tool in fighting these blocks: SNI encryption.

It's faster, cheaper, more capable, and… smaller. As software evolves, it tends to get slower, more bloated, buggier, and bigger. https://fx.sh is architected from the outset to prevent this.

If your terminal ever gets into a 'bad state', like garbled outputs, you run 𝚛𝚎𝚜𝚎𝚝. I noticed it was… oddly slow. Turns out in 1979 3BSD's 𝚝𝚜𝚎𝚝 had a 𝚜𝚕𝚎𝚎𝚙(𝟷) to let mechanical printer-and-ink terminals 'settle down' 😂 I asked 𝚏𝚡 to write me a faster alternative in Zig. It takes 1ms instead of 1s. It was really cool to dive deeper into 𝚗𝚌𝚞𝚛𝚜𝚎𝚜, 𝚝𝚜𝚎𝚝, 𝚝𝚎𝚛𝚖𝚒𝚗𝚏𝚘, and the myriad ways in which your terminal can get… cursed. It's legit fascinating how far back our technology stack goes, and how enduring good system design is, even with the occasional 'spider webs' you can find. Check out http://github.com/rauchg/rst for a fun explainer (and some alternative solutions.) ps: once you start using 𝚏𝚡, you'll notice slowness like this everywhere :D

Intelligence is getting cheaper. @OpenAI Sol's price reductions & discounts on Vercel AI Gateway have made Sol our fastest-growing frontier model. This shows ① that the demand for intelligence is highly elastic: as inference costs fall, usage grows rapidly. ② If you're not using a gateway, you're missing out on this incredible price volatility, which lowers your operating costs and increases your margins. It's no wonder the router space has heated up… gateways are inevitable.

Our philosophy on extending 𝚏𝚡: open protocols. ∙ MCP (modelcontextprotocol⁠.io) ∙ Skills (agentskills⁠.io) ∙ Plugins (agent-plugins⁠.org) And the best one, Unix: ① Small programs that do one thing well and compose by calling other programs. ② 𝚕𝚒𝚋𝚏𝚡 that enables embeddability into more complex programs. You should be able to build your own CLI, background agent, software factory. Local or cloud.

AI can do many wonderful things and grant many wishes, but we have only one earth, one California, one Patagonia, one Mendoza. USA and Argentina are two of the freest countries that also happen to have the most glorious land and geography. I'm very long 🇺🇸🇦🇷. We're so early.

I used 𝚏𝚡 to build Mini, a personal web browser. Many times I need to screen share, stream, or just test something without the 'bloat' and state of my browser. So I go Mini. This was a cool test for http://fx.sh and open models. It was the first time I experienced how cheap, but also how *good* these things can be. 'Personal software' is a hot topic. "I spent $25,000USD to cancel my $10 XYZ subscription" is a real thing 😂 (especially in Enterprises these days). But I think open and local models will change this dynamic very profoundly. Software will really become spontaneous. ps: shoutout to Orion browser who inspired me with 'focus mode'. I wanted my browser to be even more minimal, hackable, and run on Chromium.

http://fx.sh now supports your @Grok & Codex subs Test it out on a sandbox. Instant to install:

This was wild to watch unfold. We ran 𝚒𝚜-𝚊𝚐𝚎𝚗𝚝𝚒𝚌 in a loop against http://is-agentic.com until it got to 100/100. It made us close quite a few gaps. We worked hard to make sure the criteria is high quality and worth your time & tokens.

Our Python team is making rapid progress

https://fx.sh 0.0.5 gets even smaller fits in 2 floppy disks 💾💾😁 (𝚡𝚣-compressed) shipping tomorrow with our most-asked feature!

We’re building AWS for agents

Bun’s pursuit of simple, fast & open is the perfect match for Vercel and how we think about the world. Congrats @jarredsumner on shipping!

http://fx.sh is 6.3mb. It starts up in 10µs¹. It's a Zig-compiled static ELF binary, or an even smaller 𝚕𝚒𝚋𝚏𝚡.𝚠𝚊𝚜𝚖² AI will make most infrastructure natively optimized. Case in point: fx can literally finish tasks faster than other agents can boot. Fast is a one-way street. ¹ that's microseconds ² fun fact: the wasm build is actually *smaller*. A lot of binary weight comes from the TLS/HTTP stack. In WebAssembly, fx delegates 𝚏𝚎𝚝𝚌𝚑() to the JS runtime instead. You can try this on http://fx.sh/try

I love @𝚟𝚎𝚛𝚌𝚎𝚕 on Slack. Incredible ship

SF has breathtaking views

PSA: http://fx.sh has free GLM-5.2 for 2 weeks from @blackboxai, and 50% off @ChatGPT Sol for a month if you use ▲ AI Gateway. Sol is a brainiac, GLM-5.2 is a workhorse. Usage is much higher than we anticipated. We're working on adding more GLM capacity!

I’ve been using http://fx.sh as my daily driver and it’s a one-way street. It’s 10-20x smaller than the major coding CLIs. It starts up instantaneously. It feels more like using 𝚣𝚜𝚑 than an IDE in your terminal. It’s embeddable anywhere, even your browser via WebAssembly. And of course it’s open source and model-agnostic. It’s experimental, but give it a shot and let us know how it goes!

Your software factory should be a monorepo. All your company context (design, marketing, sales, engineering, support…) in one place for agents to build upon

We are putting $1M towards verifying the security of Vercel Sandbox, in the open. You're free to test any model in the world to try and find an escape. I'm looking forward to bringing transparency to what frontier models can and cannot do in terms of real-world guardrail exploitability. If escapes are discovered, we'll be ready to patch, iterate, and share our findings with the broader community to strengthen global cybersecurity.

Every idea you shelved as "too ambitious" is now back in play

You can now host your repos in Cursor Origin and deploy to Vercel via Cursor Origin which is itself hosted on Vercel. And unlike GitHub, it's online 😁

Wow

https://xn--c1h.to

We ran evals on GLM 5.3 cybersecurity capabilities. It's the new open frontier. Given its lower costs, I expect this to be a boon for defensive security work. e.g., it means you can run https://deepsec.sh at least 3× more often!

Recursive self improvement

Impressive

I think people don’t realize how much of the success of React is actually @shadcn. I once referred to React as the actual “LEGO brick for adults.” But in reality it was more the description of the geometry of the bricks. The spec. Shadcn is what people actually wanted from React. Reusable high quality components that also tunable. It’s a pseudo-library. There’s code but it’s actually meant to be digested into your context window and remixed.

I’m interested in finding and funding the next Disney. True to its original values, focused on the American spirit. Merit, family, overcoming of adversity, optimism, the pursuit of happiness. AI makes everyone able to create. What will define our next generation is *what* we create with it. The next icons. The next unique brand that goes further than how it’s drawn or rendered. Mickey Mouse survived several technological revolutions because of what it represented. I think most big entertainment companies have forgotten that. If you’re creating something cool in this space and need help, technologically or financially, slide into my DMs.

I’m interesting in finding and funding the next Disney. True to its original values, focused on the American spirit. Merit, family, overcoming of adversity, optimism, the pursuit of happiness. AI makes everyone able to create. What will define our next generation is *what* we create with it. The next icons. The next unique brand that goes further than how it’s drawn or rendered. Mickey Mouse survived several technological revolutions because of what it represented. I think most big entertainment companies have forgotten that. If you’re creating something cool in this space and need help, technologically or financially, slide into my DMs.

Vercel is the fastest [AI Gateway] infrastructure in the world

One command to rule them all (tokens). I predict that this will become the default way of using coding AI at scale. Uptime, model choice, lower costs, observability, ZDR… The cool thing is that this command configures every *existing* coding harness, like Claude Code & Codex!

Try https://eve.dev with 🆓 GLM 5.2 @ up to 500TPS from @blackboxai

The best things in life are free

𝙷𝚊𝚛𝚗𝚎𝚜𝚜𝙰𝚐𝚎𝚗𝚝 gives you the freedom to run and swap any brain. Or make a bunch of brains compete or collaborate on a problem. It’s now one line of code to switch to @grok.

Endless opportunity everywhere you look

This is such a nice improvement. Try 𝚗𝚙𝚡 𝚜𝚊𝚗𝚍𝚋𝚘𝚡@𝚕𝚊𝚝𝚎𝚜𝚝 𝚜𝚑 – it's mind-blowing. It feels faster than your local machine. Sandbox now comes with a reasonable default set of pre-installed tools, and you can fully customize them.

Seedance 2.5 on @vercel AI Gateway 🐇

Digg was by far my favorite site of the web 2.0 era. Getting my tech essays onto "the digg front page" was one of the coolest memories of my teenage years 😁 So happy to be powering the new, agent-native http://digg.com on @vercel. Full 🕳️

Towards a future of self-maintaining software. Must read of how we balance quality and autonomy for the development of @aisdk, one of the most popular open source AI libraries in the world.

Deployments on Vercel are up 3x since the start of the year. And 2x the 6 months prior. Agents love deploying. We supported this growth with excellent uptime, while also shipping 19 build & deploy performance improvements on our public changelog in this period. Redis served us well since I first created the z̵e̵i̵t̵ Vercel infrastructure 10 years ago, in numerous places. But as many such cases, Dynamo is the right beast for this size of demand.

Reply with your critical Vercel dashboard feedback and we will fix!

Many people are saying

The growth of @aisdk has been astonishing. ~80.5M downloads every 30 days, and growing faster than all AI labs SDKs. Most importantly: open and provider-agnostic.

Keep an eye on this view http://vercel.com/ai-gateway/leaderboards/models

Cool

RT Malte Ubl Given the state of AI we must internalize: Everything hackable will get hacked. Two forces are at work: ▪︎ Near-frontier open weight models hack without refusal ▪︎ Frontier models help you defend yourself How we're approaching this new reality ↓ https://vercel.com/blog/everything-hackable-will-get-hacked

We were promised flying cars and all we got is infinite superintelligence for everyone

There’s a cool meta to the “you’ve got this, try harder, keep going” prompts. That’s engineering management. You know the outcome is within sight, so you decide to spend more time and money chasing it. You encourage. And you also need the wisdom to know you’re chasing a deadend and have to pivot and cut losses. Agent management.

This tool has become so valuable for cybersecurity defensive work, that it’s become a verb within Vercel. “Did you deepsec it?” “@𝚟 can you deepsec” “/𝚍𝚎𝚎𝚙𝚜𝚎𝚌” It’s a bit like /𝚝𝚑𝚎𝚛𝚖𝚘-𝚗𝚞𝚌𝚕𝚎𝚊𝚛-𝚌𝚘𝚍𝚎-𝚚𝚞𝚊𝚕𝚒𝚝𝚢-𝚛𝚎𝚟𝚒𝚎𝚠 but for the security of your code. Must have in your software factory.

Was so fun to speak to the @speedrun @a16z startups while enjoying ▲ 🍵. Great founders and great vibes

Vercel Sandbox isolates both ① compute and ② network. Kimi's paper shows container-based isolation is not enough for frontier models. Vercel Sandbox uses strong microVM isolation to address ①. OpenAI's escape was on ② the network path to Artifactory. Our egress firewall is now free so everyone can constrain misbehaving agents' network activity further. ¹ "in our early experiments with traditional container-based sandbox runtimes, we observed several kernel panics and deadlocks caused by unintended agent operations." 🔗 github⁠.com/MoonshotAI/Kimi-K3/blob/master/k3_tech_report.pdf ² "to gain Internet access, the models identified and exploited a previously unknown zero-day vulnerability in Artifactory⁠, a package registry cache proxy" 🔗 openai⁠.com/index/hugging-face-model-evaluation-security-incident/

At Vercel every week is ship week

Full Bun support on Vercel

Hermes + Vercel = 🖤

If you’re not reading the code, whether explicitly or through agentic inquiry, one or more of these is true: ○ You’re a beginner ○ Software is throwaway ○ You’re prototyping ○ You have no users / revenue ○ You’re taking on debt & risk ○ Your problems are basic And btw. All of this is fine. But the reality is that models are still not at the “full autonomy” stage yet. They make rookie mistakes, they go down bad architectural paths. I just had the best model in the world add a nonsensical 700ms delay to “settle” something and it told me “you’re right, I was cargo-culting” 🤨 I am on the camp that this need will diminish more and more. Most code is indeed going to be assembly-like. But we also have the global internet and software infrastructure riding on these models and narrative, and we have to respect that.

Dreamcore

Grok Imagine Image 2.0 on Vercel AI Gateway Excellent 🖼️ model, #2 already on http://Arena.ai

How Vercel helps prevent surprise cloud bills: ◾ Soft & hard caps¹ ◾ Anomaly alerting² ◾ Recursion protection for Functions³ ◾ Billing usage APIs your agents can query⁴ ◾ Always-on DDoS L3/L4/L7 mitigation on all plans⁵ A ton of work went into the streaming data infrastructure to support all this functionality. Vercel ingests a superlative amount of data that needs to be analyzed in realtime to detect anomalies, threats, dispatch alerts, and more. Super thankful to the teams that worked so hard on this for years! Much more to come. ¹ vercel⁠.com/changelog/improved-hard-caps-for-spend-management ² vercel⁠.com/changelog/anomaly-alert-configuration-now-available ³ vercel⁠.com/changelog/automatic-recursion-protection-for-vercel-serverless-functions ⁴ vercel⁠.com/changelog/access-billing-usage-cost-data-api ⁵ vercel⁠.com/docs/vercel-firewall/ddos-mitigation

Fun day at ▲. Excited to go home and deploy to ▲

SITUATION DETECTED: Herdr joins YC, gains Vercel Sandbox plugin

“The others make the easy part easier. Vercel makes the hard part easy” – direct quote today from tech lead for AI agent platform built on http://eve.dev at 55,000+ person company. They wanted to have their company's all-knowing agent, like our @𝚟. Tried @aisdk, good but too low level. Tried off-the-shelf solutions and ${⁠lab} Enterprise products, expensive and inflexible. Tried agent frameworks, which led to the quote above… didn't hit the spot. @evedev_ team cooked. It's hard to find the abstraction that's both easy and scales with sophistication.

Choose your (free) fighter: ○ .online ○ .site ○ .space ○ .store ○ .tech ○ .website

Excited to welcome the legendary Amit Agarwal, fmr President and CPO of Datadog, to the Vercel Board of Directors. He built the ubiquitous observability platform for developers and enterprises in the cloud. Together we'll scale Vercel to the greatest level.

Skill𝑠𝑒𝑡𝑠

Next.js is the Next.js for SPAs