Vercel Ship 2026 Recap: Every Talk and Launch Explained

Vercel Ship 2026 was a keynote event in San Francisco streamed live for about six and a half hours. The company framed the day around agentic infrastructure as the place where people and coding agents build, deploy and run software.

Speakers came from Vercel, Anthropic, Meta Superintelligence Labs, Shopify, AWS, SAP, Notion, ClickHouse, Cerebras, Slack, Mercor, Ora and Adapta. The number of developers building on Vercel tripled over the last 12 months. This recap walks through the launches and then each talk in order.

What Vercel launched at Vercel Ship 2026

• Python on Vercel: generally available, standard CPython with PyPI packages and Django, FastAPI or Flask support
• Containers on Vercel: available, Docker containers with custom runtimes running on fluid compute
• Vercel Daemons: public beta, a compute primitive for always-on background workers
• Vercel Connect: generally available, more than 1,000 connectors
• Vercel Drives: generally available, persistent workspaces that keep code, data and memory across sandbox runs
• Token Pass: $80 of model usage for $30 a month on Vercel AI Gateway
• Bring Your Own Cloud on AWS: generally available, Vercel running inside the customer's own AWS account
• Amazon S3 integrations on Vercel: generally available for S3 buckets, S3 Tables and S3 Vectors
• Vercel on AWS Marketplace: Vercel products can be bought through AWS Marketplace
• Vercel Shop: generally available, an open-source Next.js storefront template for Shopify
• Shopify Hydrogen 3 and the Universal Commerce Protocol: for agent checkout and catalog access
• ClickHouse on the Vercel Marketplace: with $300 in free credits
• Claude Managed Agents dynamic workflows: from Anthropic, running up to 1,000 agents
• Muse Spark Open: from Meta Superintelligence Labs, an open-weight model under Apache 2.0 that runs locally on a Mac Studio

Guillermo Rauch, Vercel CEO: the agentic infrastructure bet

"The haters said AI would take your job, but instead you gave AI a bigger backlog." Guillermo Rauch, founder and CEO of Vercel, opened the keynote 11 years after starting the company and cited growth over the last 12 months. The number of developers building on Vercel tripled. The number of new projects launched on Vercel doubled, and deployments triggered by coding agents grew 28 times. It took Vercel 10 years to reach its first billion deployments, and the platform now sees one billion every four months.

His argument was that Vercel turns software development into a loop in which developers and agents build, deploy and iterate. A developer can deploy an end-to-end software factory on Vercel in less than three minutes. The platform exposes more than 200 CLI commands so coding agents can orchestrate platform resources.

Malte Ubl, Vercel CTO: Python, containers and Daemons

Python on Vercel is generally available, using standard CPython and PyPI packages with Django, FastAPI or Flask. Malte Ubl, CTO of Vercel, said fluid compute lets teams run full backends, Python apps, containers and always-on workers without managing servers. Python functions can run for hours or days, with single requests lasting up to 30 minutes. More than 150,000 teams already run Python backends on Vercel.

Containers on Vercel run Docker images with custom runtimes on fluid compute, and since June 5,000 teams shipped 150,000 container deployments. Vercel Daemons is in public beta as a primitive for continuous background work. Delphi's team of 10 engineers deploys to production on Vercel more than 100 times a day. Vercel processes two trillion requests a month, and he noted "We process over one billion API requests every month on fluid compute for ourselves and that number is growing fast." The company runs its own control plane on fluid compute to handle that volume.

Tom Occhino, Vercel CPO: the agent stack

Vercel AI Gateway serves over 4 trillion tokens a day for more than 400,000 production customers. Tom Occhino, chief product officer of Vercel, walked through the building blocks Vercel offers for production AI agents: compute, a model gateway, integrations and storage. The open-source AI SDK, Vercel's TypeScript library for calling models, defining tools and streaming responses, gets over 30 million downloads a week. Vercel Connect is generally available with more than 1,000 connectors.

Token Pass gives $80 of model usage for $30 a month on AI Gateway. Vercel customers start over six million sandboxes a day. Vercel Drives is generally available, giving agents persistent workspaces that keep code, data and memory across sandbox runs.

Jeanne Grosser, Vercel COO: governance and Bring Your Own Cloud

At Klaviyo, 512 employees deployed 356 internal apps in their first two weeks on Vercel. Jeanne Grosser, COO of Vercel, presented enterprise apps and agents as a central control plane to govern apps, agents, users, budgets and tokens. Enterprise Managed Authorization lets IT grant team-wide agent access to services such as Slack through the company's identity provider. Bring Your Own Cloud on AWS is generally available for Vercel enterprise customers.

Mercor served 63 billion tokens through Vercel AI Gateway in its first 10 days without hitting rate limits. Vercel's own support agent resolves 91% of Vercel's customer support cases by itself.

Assaf Elovic, Ora: the web is not ready for agents

Web traffic from AI agents grew by almost 8,000 percent in the past year. Assaf Elovic, co-founder and CEO of Ora, argued that websites need a front built for AI agents because agent traffic is overtaking human traffic. He said "We've learned that only 4% of the web is barely usable by agents." Ora's research covered 120,000 websites.

Agents reached an llms.txt file in only 32% of the cases Ora evaluated. Agents working from screenshots instead of HTML used 45 times more tokens and took 23 times longer. In 35% of scenarios, agents guessed URLs before running a web search.

Katelyn Lesse, Anthropic: Claude Managed Agents

On a test codebase with 70 planted bugs, Claude with a dynamic workflow found 66 of them across three runs while a single agent found 14 to 27 per run. Katelyn Lesse, head of platform engineering at Anthropic, described Claude Managed Agents as separating the agent's reasoning from the execution environment, so long tasks run securely with execution on Vercel Sandboxes. She said "Dynamic workflows can run up to 1,000 agents." Up to 64 sub-agents can run at once.

In Anthropic's financial-task experiments, every multi-agent strategy beat a single agent on the same token budget. Claude Max and Team plans now include API credits for building with the API and Claude Managed Agents.

Prashant Ratanchandani, Meta Superintelligence Labs: speed and open weights

Meta went from deleting its codebase and starting over to running Meta AI on Vercel in 95 days. Prashant Ratanchandani, VP of engineering at Meta Superintelligence Labs, in conversation with Tom Occhino, said the lab prioritizes development speed and open distribution. He said "We don't believe in centralizing superintelligence to very few." The lab has released 11 AI models and products since April.

The team on Meta AI grew from four to 97 engineers within three months of moving to Vercel. Muse Spark Open is an open-weight model under Apache 2.0, built with Apple to run locally on a Mac Studio. Muse Wonder gives agents active memory and focus for tasks that run over several days.

Mala Neti, Slack: Vercel's internal agent lives in Slack

V is Vercel's internal company agent and lives inside Slack as the single entry point to all internal sub-agents. Mala Neti, senior software engineering manager at Slack, showed how Vercel built and deployed V. That entry point includes a data agent maintained by Vercel's data science team.

Vercel Connect issues short-lived, scoped tokens on demand to query Snowflake, so no static credentials are stored. Slack's developer tools for agents include the Bolt SDK, Block Kit Builder and the Slack CLI.

Adapta: model routing for small businesses

Adapta is a bootstrapped, profitable Brazilian business that passed $70 million in annual recurring revenue in 2026 and won 25,000 customers within six months of launching its knowledge-work platform in 2023.

Adapta, a Brazilian AI company, was an early Vercel AI Gateway customer and routes each request to a model using a golden dataset it updates daily to cut costs while keeping frontier quality. It launched Maestro for building company-wide agentic workflows. One deployment engineer supports up to 60 mid-sized business clients, and a custom agentic CRM for an agricultural client doubled its service capacity and saved over $200,000.

Sarah Sachs, Notion: Notion as the context layer

About half of all content written to Notion is now written by AI agents and MCP connections. MCP traffic grew 70-fold over the past six months.

Sarah Sachs, AI lead at Notion, positioned it as an open context layer where people and agents work together, exposed through MCP. Notion workers run stateful scripts on Vercel Sandboxes, which cuts token spend for users. Notion built separate search for people and for agents, because agents read and write differently and need less re-ranking.

Olawale Oladehin, AWS: agent workloads inside your own cloud

60% of Vercel deployments in 2026 are triggered by coding agents, up from 3% six months earlier. A typical multi-agent run uses 50,000 or more tokens.

Olawale Oladehin is managing director at AWS. He argued that AWS infrastructure plus Vercel's control plane gives enterprises compliant agent hosting inside their own VPCs. Amazon S3 buckets, S3 Tables and S3 Vectors are generally available as native Vercel integrations. Vercel products are now available through AWS Marketplace, so enterprises can consolidate billing. Bring Your Own Cloud runs Vercel workloads inside the customer's AWS VPC with the same developer experience.

Sophia Levens, SAP: generative UI across SAP

Joule Work generates interfaces at runtime from what the user intends, using a compositional design system. Sophia Levens is SVP and head of SAP Design System. She said SAP is rebuilding its interfaces with a compositional design system on Vercel that generates UI at runtime. She said "84% of all the world's global transactions and commerce hit an SAP system."

SAP software reaches 440,000 customers and 350 million users. SAP customers and partners ship over 35,000 customized Fiori apps a month on its design system. SAP cut its client rendering primitives to nine core elements to give generative UI more room.

v0 is the main prototyping and vibe coding tool across SAP teams. SAP distributes its AI skills to coding agents through skills.sh.

Aaron Brown, Mercor: identity and AI Gateway at scale

Batch LLM grading runs dropped from 12 hours to two hours at Mercor. The company runs 2,000 active projects, up from fewer than a dozen a year earlier.

Aaron Brown is head of security, IT and GRC at Mercor. He explained how Mercor secures heavy AI usage with Vercel Passport identity defaults and AI Gateway governance. Mercor was the first customer to deploy Vercel Passport, across hundreds of internal projects. It served 18 million requests and roughly 63 billion tokens through AI Gateway in 10 days.

Yury Izrailevsky, ClickHouse: from database to data platform

ClickHouse launched a managed Postgres service on NVMe storage for fast transactional work. Yury Izrailevsky is co-founder and president of product and technology at ClickHouse.

He described ClickHouse growing from an open-source analytical database into a data platform covering analytics, observability, managed Postgres and AI engineering. ClickHouse open source has shipped over 120 releases since 2016, with updates from nearly 3,000 contributors. Over half of ClickHouse Cloud revenue comes from customers outside the US, across 38 cloud regions. Anthropic and OpenAI run ClickHouse for data management. ClickHouse is now on the Vercel Marketplace with one-click provisioning and $300 in free credits.

Andrew Feldman, Cerebras: inference fast enough to sit inside the loop

CrowdStrike runs Cerebras inference in-line to inspect and clean live network traffic. Small fast models used for real-time classification answer in under 100 milliseconds.

Andrew Feldman is founder and CEO of Cerebras. He argued that very fast inference lets models run directly inside real-time software loops. Cerebras makes AI chips that run inference 15 to 20 times faster than GPUs. At that speed an application can now call a model dozens of times a second.

Tobi Lütke, Shopify: commerce built for agents

More than 50% of new commits to Shopify's main monorepo come from its AI agent, River. Shopify handles 20% to 25% of all human web traffic during Cyber Week.

Tobi Lütke is CEO of Shopify and appeared in a fireside chat with Guillermo Rauch titled The Third Customer. The pair discussed agentic commerce, headless storefronts and verifying AI-written software. Shopify launched Hydrogen 3, a framework-agnostic SDK that works with Next.js and Vercel Shop. Shopify introduced the Universal Commerce Protocol so AI agents can use open checkout and catalog capabilities. Vercel Shop, an open-source Next.js storefront template for Shopify, is generally available. Shopify is rewriting its storefront rendering engine from Ruby to Rust with AI coding, aiming for a 10x performance gain.