Vercel Ship 2026 recap: Agent Stack, Connect, and Services launched

For a decade, Vercel has shaped how the web is built. Now, we’re doing the same for agents. The companies that win the next decade will build on infrastructure designed for agents from the start, and thousands of people gathered across London, Berlin, and New York to do just that at Vercel Ship 2026.

Guillermo kicked off Ship by sharing his vision for Vercel: a true full-stack platform where you can deploy anything, including software that can think.

has three parts: Agentic infrastructure

In the next keynote, Tom walked through each primitive in the , an end-to-end set of building blocks for agents. Hedi introduced , a secure way for agents to connect with external systems. And Shar gave a full demo of , Vercel's new agent framework that implements the Agent Stack in minutes, in a single directory. Agent StackVercel Connecteve

No matter the workload, every agent needs to connect to models, run workflows across many steps, and connect to the systems and people that make it useful. Vercel gives you every primitive you need to build all three capabilities.

launched at Ship 2026 in London as the newest building block in the Agent Stack. Vercel Connect

Connect provides agents with secure access to the tools, data, and services they work with, without requiring a long-lived provider token stored in your environment variables. Your app or agent requests a temporary credential scoped to the one task in front of it, so there’s no standing secret left to leak.

After building hundreds of agents at Vercel, we noticed that the architecture underneath had the same shape. is that shape as an open-source framework, and it's how we build, run, and scale production agents at Vercel. eve

eve Agents live in a single directory, with instructions in markdown and tools in TypeScript. Durable execution, sandboxed compute, approvals, subagents, and evals are already wired in.

Guillermo also highlighted Vercel's ability to host full-stack applications, with support for additional backend frameworks, databases, and now, microservices.

Malte announced one of our most requested features. You can now build, store, and run Docker images on Vercel with and . Bring a Dockerfile and run it on Vercel Functions or Vercel Sandboxes, and push, pull, and manage your images. VCR is OCI-compliant and works with the commands you already use, , , and , so there's nothing new to learn.Dockerfile supportVercel Container Registrydockerpushpulltag

A project can hold unlimited repositories, created in the dashboard or on the fly when you push, all under the same OIDC and access-token controls as the rest of Vercel. Every image is optimized in the background for Fluid compute and stored as a precompiled snapshot, so it spins up fast in Sandboxes and Functions.

Earlier this year, Vercel brought on some of the in the world to extend support for backend workloads. best Python developers

Guillermo announced the launch of Vercel Services, available July 1. With Vercel Services, microservices are now a first-class citizen on Vercel. You can develop and deploy your frontend and backend together, and backend-only changes still build the app in a full preview environment. And for the first time, those services can communicate with each other without touching the public internet.

For us, Next.js and eve made building agents the easy part. The hard part was everything around them: access, authentication, integrations, and proving all of it to the security team. Jeanne told the story of how we built an enterprise security platform for ourselves, then made it available to our customers.

brings the Vercel developer experience to everyone at your company, with the identity, access, and governance needed to deploy safely. Vercel for Enterprise Apps and Agents

Malte announced (now in Public Beta), the intelligence layer for shipping on Vercel. Built on eve and the Agent Stack, it monitors your production deployments, autonomously investigates alerts and anomalies, and opens fixes in PRs for you to review and approve. Vercel Agent

Vercel Agent's first-of-its-kind permissions model combines plan mode with granting permissions, making it safe to use for both developers and the enterprise. Rather than asking you to approve actions one-at-a-time, Vercel Agent plans what permissions it will require to complete a task and then asks you to approve them in a single, coherent step. It runs as its own identity, is read-only by default, and asks for narrow, temporary permission before it touches production.

Our first Ship outside the U.S. ran five days of events leading into the keynote, and an ice cream truck spent the week working its way across the city.

Two hundred builders shipped agents on Codex and Vercel in a hackathon with OpenAI. Three projects stood out. Stella hunts down unclaimed grants for London SMBs, Oscar is a co-pilot that sharpens your prompts before you send them to a model, and Phone Jail blocks distractions and roasts you when you reach for your phone anyway.

Reporters traded story ideas over a media dinner. Twelve founders talked through the future of AI in EMEA at a VC dinner. And a dozen CTOs from teams including Marks & Spencer and Currys went off the record on what it takes to build and scale agents in the enterprise.

Day Zero, the day before the keynote, turned into a 400-person builder night. Enterprise teams Mentimeter, Okta, and MongoDB joined Vercel in a packed room alongside partners DeepMind, ElevenLabs, and Cursor.

A 160-deep waitlist formed for Malte's talk , followed by Guillermo and Deliveroo's Will Shu's chat on building the agentic enterprise.A New Stack for a New Era of Software

Jeanne Grosser, COO of Vercel, moderated a panel on taking agentic workloads from prototype to production with some of the EU's top CTOs and AI leaders from frontier labs. A few highlights:

Autonomy should track risk, and the better an agent can assess the risk of a change, the more you can let it run on its own. Shopify and Amplitude already auto-review and merge 60-70% of low-risk PRs with no developer time, while a two-line change to authentication still goes to a human. (Cursor):Arthur Viegers

: To decide what an agent can do on its own, judge each action by how reversible it is and how large its blast radius is, then contain it accordingly, for example inside a self-hosted sandbox. As you grant more autonomy, observability and evals become the things that keep it safe.André Balleyguier (Anthropic)

The predicted death of stores never arrived, and Currys leads its markets by betting on humans. The open question for agentic commerce is which shopping journeys customers want an agent to handle and which they still want a person for.Ryan den Rooijen (Currys):

AI-native customer experience replaces navigation with intent. Reaching it takes three layers: a way to capture true customer intent, a brand brain that stores brand and product data for AI systems to retrieve, and generative UI that composes pages on the fly instead of serving static ones.Nicolas Le Pallec (AKQA):

The day ended with a fireside chat between Guillermo and from 20VC. Below are excerpts from the lightning round at the end of the discussion. Harry Stebbings

Guillermo Rauch: What are the attributes of a founder you want to back?

Harry Stebbings: The profile has changed completely. It used to be a sales-led CEO who could fundraise and grow customers, where personality and sales carried the company. Now I want an engineering- and product-led founder.

Guillermo Rauch: Can you teach someone to become a voice for their industry?

Harry Stebbings: You can, but it's hard. You have to do it for six months straight. It's a muscle you build, and you have to commit to it every day and every hour.

Guillermo Rauch: When anyone can produce content and software, how do you stand out?

Harry Stebbings: You come up with a new format, a new intro, a new way to tell the story and shape the work. The content itself has shifted too. We used to do shows about leadership and hiring, and nobody cares about that anymore. What lands now is a founder and AI. Those are the two things.

The Berlin event opened at the Old Mint, “Alte Münze”, where coins were once hammered by hand, before machine automation took over. It is also where the JavaScript logo was born, and where Guillermo Rauch (CEO) met Malte Ubl (CTO) for the first time at JSConf EU.

Software is moving the same way, from hand to machine, and Vercel is paving the way for building in the agentic era. Over 500 people gathered from across Germany to do just that at Ship 2026 in Berlin.

(Head of Forward Deployed Engineering @ Vercel) moderated a CTO panel on building and scaling as the cost of creation trends toward zero. A few highlights:Chris Williams

AI won't fix a broken business model, so the work is still operational excellence. He compares ecommerce to a decathlon, you win by being top-tier across every discipline at once. What's changing is discovery, which is moving into chat, so Flaconi is investing in the data foundation that makes its catalog retrievable by agents, betting that buying and checkout eventually happen there too. (Flaconi): Sven Rosemann

venture funding is no longer a prerequisite. For many ideas you ship, find distribution, and iterate instead. Astra reached 10 million users and $25M ARR fully bootstrapped, with a small team kept fast by a stable foundation and agents given a lot of autonomy while kept as safe as possible. (Astra AI): Tjaž Silovsek

Vercel presented its first-ever DACH Partner Award to Valtech, for the exceptional impact and delivery, strategic thinking and team enablement it brought to clients. The presentation also thanked the event's sponsors: Auth0, Storyblok, Cursor, SAP, Accenture Song, Blazity, and JVM.

Ship NYC was the most in-demand of any stop yet with over 3,000 people registered. Founders, partners, leaders, and builders gathered for five days of hackathons, executive receptions, workshops, and more, with Ship day at the center.

The week opened with Built in NYC, a hackathon co-hosted by Vercel and Anthropic alongside Sentry, Auth0, Resend, and Supabase. More than one hundred people joined us for an afternoon of building and demoing agents on Vercel.

Jeanne DeWitt Grosser and Nick Bogaty gathered the leaders shaping AI-native development at a VIP reception. Founders traded notes on the future of agents over a VC dinner. At Per Se, Guillermo, Jeanne, Tom, and Malte took CTOs and CIOs for a closed-door conversation on where traditional architectures start to break in the agent era.

AI executives and senior leaders assembled at Moss for a fireside chat led by Tom Occhino with (CTO @ Bloomberg Media), (VP Engineering @ Serhant), and (Staff Design Program Manager @ DoorDash). They discussed what's working, what's not, and where agentic infrastructure is driving meaningful outcomes for enterprises. William AndersonJeremy BuntingAli Patterson

Jeanne moderated a panel on building and deploying internal agents with technical leaders on who should build, what governance looks like, and how teams can move from experiment to production. A few highlights:

The way to win the shadow-IT battle is to constrain the stack and stay permissive inside it. His AI Accelerator trains non-engineers, marketers, and early-career staff to build internal apps on a deliberately tight stack of Claude Code, Vercel, and GitHub, with ephemeral preview deployments as the guardrail. As he put it, "If it becomes the Department of No, you will have people working around you." (MJH Life Sciences): Andy Martin

The cleanest way to structure agents is to mirror a human software team. Product-manager agents write requirements in Linear, engineering agents generate specs, and build-and-QA agents take it from there, so engineers become leads running multiple streams. To keep output safe, SERHANT runs evals for hallucination and compliance, including a custom model for Fair Housing Act rules. (SERHANT): Greg Chan

Embedding v0 inside your own product lets your customers build their own workflows. At ZoomInfo, 40,000+ of them do, turning the voice of the customer into a pull request. Agents analyze recorded calls and open PRs, so the product team decides what ships instead of synthesizing feedback. They pair Vercel's deployment guardrails with ZoomInfo's context-graph permissions, so each agent only touches data its user can see. (ZoomInfo): Victor Oliveros

Tom sat down with (VP of Product @ Shopify) to discuss where commerce and software go next. Below are highlights: Vanessa Lee

Tom Occhino: Vercel and Shopify have shipped something new together. What is it?

Vanessa Lee: We boiled Hydrogen down to the parts of commerce that are genuinely hard (analytics, optimistic cart UI, variants) and made them work with any framework. So if you're on Next.js on Vercel, Shopify's commerce primitives now drop in, with an integration in the Vercel Marketplace. Read more .here

Tom Occhino: You also opened the Catalog API. What does it unlock?

Vanessa Lee: There's never been a good open product-search API. We have billions of products and millions of merchants, so we built one, the only widespread shopping API, made for commerce. It lets anyone embed shopping anywhere. Someone in my Toronto running group used v0 to build a site with a leaderboard and the weather, and the Catalog API surfaced products from nearby stores. Commerce stops being a destination and becomes part of what you build.

Tom Occhino: Shopify and Vercel both obsess over reducing friction. Why does that matter so much?

Vanessa Lee: Early on, Tobi's insight was that tools shape the market. The TAM was never limited by demand, it was limited by how hard the tools were to use. Make building easier and more people build. There's resourcefulness in the general population that's just untapped, and lowering the friction is what releases it.

We closed out Ship in New York with a conversation between Guillermo Rauch and (CEO @ Notion). A few highlights:Ivan Zhao

Guillermo Rauch: You've started telling designers to design for the API first and the UI last. Why?

Ivan Zhao: Because your next customer isn't only a human, it's a human and an agent, and agents are blind. They don't see your interface, they read your semantics and your API. The most enduring software has always been a data structure anyway, the spreadsheet, or our block, a data structure that happens to have a UI. So design the structure and the meaning first, the interface last. A great tool for developers turns out to be a great tool for agents.

Guillermo Rauch: Vercel and Notion are both aggressively model-agnostic, Switzerland in the model wars. Why?

Ivan Zhao: The models leapfrog each other every few weeks, so picking a side is a losing game. We route automatically by the intelligence a task needs, frontier models for coding, smaller and faster ones for support and summarization. What excites me most is that as more use cases stop needing frontier intelligence, you can shape smaller models to your product and win back speed and cost. And speed creates discontinuities, a model five times faster can change the kind of interface you'd even design.

Guillermo Rauch: When a billion knowledge workers are agents, how does that reshape us?

Ivan Zhao: We can't fully see it, and that's the point. First-order effects are easy, the iPhone was an iPod plus a phone plus a browser. The second-order effects are the ones that surprise you, like what social media did to teenagers. We shape our tools, and then our tools shape us. With software that can think, the second-order effects are the ones we won't predict.

Guillermo Rauch: Taste or velocity, keep one.

Ivan Zhao: Taste. It's the distillation of wisdom from a lot of shots on goal.

Thanks for joining us for Ship 2026 in London, Berlin, and New York. We will see you in Sydney in July and San Francisco in October. If you haven't signed up yet, . there's still time to get a ticket

Read more

Agentic infrastructure

Vercel is for shipping agents

Vercel is for shipping full-stack apps

Vercel is for shipping in the enterprise

Vercel Agent

Ship 2026 in London

Ship 2026 in Berlin

Ship 2026 in New York

Ship what's next

• . When you ask Claude Code or Codex where to deploy, you get Vercel, because Vercel is built for the way agents work.Vercel is where coding agents deploy software

• . We give you every tool you need to build and run apps and agents in production, securely, at scale. Vercel is where you build and deploy your own agents

• . Vercel runs your apps in production, handling traffic, traces, observability, and anomalies. That data gives our agents the context they need to investigate autonomously, and then surface pull requests, not just alerts.Vercel itself is automated by agents

• gives you one API to call any model, so streaming, tool calls, and structured output work the same across every providerAI SDK

• routes your requests across hundreds of models from one endpoint, with automatic failover when a provider goes downAI Gateway

• makes every run durable with automatic retries, state persistence, and observability built inWorkflow SDK

• gives each agent an isolated microVM to run and test the code it writes before it ships to productionVercel Sandbox

• lets you deploy your agent across Slack, Discord, GitHub, and more from a single codebaseChat SDK

• You can now run , , , , and at scale on Vercel. FastAPIFlaskExpressHonoother backend frameworks

• Backend-only services are also supported, so you can host REST APIs, durable workflows, queues, cron, and . MCP servers

• The Vercel Marketplace added access to databases like , , and directly from the Vercel dashboard. Amazon AuroraAurora DSQL,DynamoDBOpenSearch

• keeps internal apps internal, private by default behind your IdPVercel Passport

• The (Private Beta) is one place to see the security posture of every account and project on Vercel. It flags misconfigurations that can lead to breaches, shows you members without MFA, alerts you about secrets shared across projects, and identifies long-lived credentialsSecurity Dashboard

• Bring your own cloud (BYOC) on AWS (in Private Beta) gives you the ability to run Vercel in your own AWS tenant, including your Vercel functions.

• from Anthropic showed Claude Managed Agents on Vercel, where Anthropic hosts the agent loop while every command the agent runs executes inside a Vercel Sandbox you own, keeping its filesystem, processes, and network egress in your environment.André Balleyguier

• from Currys/Elkjøp showed how his team evolved a Nordic retailer's ecommerce stack from storefronts toward shopping agents in three phases, anchored by a Next.js migration that cut time to first byte by 40%. The end state is one where customer intent replaces navigation.Tomas Jansson

• from Vercel broke down how Vertex, Vercel's support agent, now automates 91% of support tickets and saves 5,000 engineer-hours a month without degrading as its context grows. Matan Kushner

• and from Auth0 made the case that getting an agent into production depends more on identity than on model quality. They demoed four standards-based identity patterns on the Vercel AI SDK, Token Vault, CIBA approvals, agents as first-class principals, and on-behalf-of delegation, so every action is scoped and traces back to a human decision instead of a shared API key.Jas SagooSam Bellen

• from Vercel went deep into d0, Vercel's internal data agent built on eve, which runs analyses in a sandbox and now gets 45% of its questions from other agents, rather than people.Abhi Sivasailam

• from OpenAI showed developers how to ship faster by delegating whole tasks to Codex Cloud Agents on Vercel, and made the case that as agents become users of your product, you design for agent experience with scoped API keys, markdown docs, and plugins.Romain Huet

• from Callstack demoed a QA agent that runs mobile apps on real devices, explores user flows on its own, and posts screenshots, recordings, and logs to pull requests on every PR.Michał Pierzchała

• from ElevenLabs paired the ElevenLabs Speech Engine with the AI SDK to build a voice agent that can reply in under a second, take interruptions mid-sentence, and switch language mid-conversation.Alex Holt

• from Vercel showed the mechanics of how software can ship itself, extending the agent loop past development into testing, observability, and experimentation with primitives like Vercel Flags and Skills.Malavika Balachandran Tadeusz

• from Vercel announced . The AI SDK is now downloaded over 16 million times a week. Version 7 turns it into a toolkit for building and running agents that reason, call tools, run across many turns, and work across files and sandboxes, all behind one provider-agnostic interface with no lock-in.Suyog RaoAI SDK 7

• from SAP showed how the company rebuilt its design system to be composable, so people and agents assemble interfaces from the same parts, with v0 running across its build, deliver, and decide stages.Philip Miseldine

• Building with AI SDK 7 brought three speakers on stage. , the primary engineer behind the update, went deep on the release, with the new harness layer and generation beyond text as the headline additions. from AKQA demoed the Starbucks ChatGPT app as a "generative store" that composes branded results from Starbucks' own products and visual identity, flipping the funnel so customers pull the brand into their conversation instead of visiting its site. from Black Forest Labs showed where the Flux image models fit in an agentic stack: sub-second generation and editing today, heading toward real-time generation and world models.Lars GrammelCarsten HøyerStephen Batifol

• from Cursor showed how coding agents move from autocomplete to working across an entire codebase on their own.Arthur Viegers

• from SERHANT. walked through S.MPLE, a real estate agent built on Vercel that drove a 144% increase in commission income in its first full year.Ryan Coyne

• from Brex gave its expense-audit agent a -only shell so it manages its own context instead of drowning in tool responses, cutting tool calls and token usage by 75%.Brandon Bloombash

• and from Auth0demoed how to give an agent scoped, revocable power instead of a master key, with standards-based identity patterns like Token Vault and out-of-band human approval, live on the Vercel AI SDK.Ari SchapiroChris Sev

• from Alta and from PUBLIC SCHOOL NYC showed how Alta's AI styling app runs on Vercel, generating an avatar a second and 250M+ outfits so far. Together they built a custom version for PUBLIC SCHOOL's New York Fashion Week relaunch.Jenny WangMaxwell Osborne

• from FLORAshowed three creative-production workflows running on Vercel Sandboxes. They collapsed 18 bespoke tools into two by letting the agent run and write its own code against the AI SDK.Charlton Roberts

• from Stripebroke down how to give agents spending power and capture agent-driven revenue using emerging protocols like MCP, UCP, and x402 for the machine-to-machine economy.Prasad Wangikar

• from AWSdiscussed how Vercel and AWS work together, with native AWS databases and OpenSearch available in the Marketplace and models through the AI Gateway, proven out by The Weather Company's 350M+ users.Shaown Nandi

The Agent Stack

Dockerfile support and container registry

Backends and databases

Vercel Services

Vercel for Enterprise Apps and Agents

Ship week London

Ship day sessions

Fireside chat with Harry Stebbings

Ship day sessions

Panel: CTOs on AI and growth

DACH Partner Award

Ship week New York

Ship day sessions

Vercel Connect

eve

Built in London hackathon

Media, Founder, VC, and CTO dinners

Day Zero builder night

AI Social Club

Panel: Agents in production

Built in NYC

VC, CTO, and VIP dinners

AI Social Club

Panel: Internal apps and agents in production

Fireside chat: Building agentic storefronts

Closing keynote with Ivan Zhao