Vercel Ship 2026 made one argument across the whole event: that the platform known for deploying front ends should now be where developers build and run AI agents. Nearly every release in the recap ties back to that pitch, from a new way to structure backend code to open-source tooling for the agents themselves, and it is a notable reframing for a company whose reputation was built on shipping websites.
The most consequential piece is Vercel Services, which the company says launches July 1 and makes microservices a first-class citizen on the platform. Frontend and backend can deploy together, backend-only changes get their own full preview environments, and services can talk to each other without routing over the public internet. That last detail matters for anyone who has stitched private networking onto a serverless setup after the fact; it moves a common piece of production plumbing into the platform rather than leaving it to the developer.
The agent tooling arrived in a cluster. Vercel open-sourced eve, a framework for production agents that comes with durable execution, sandboxed compute, approvals, subagents, and evals already wired in, organized as a single directory of markdown instructions and TypeScript tools. Vercel Connect reached general availability, handing agents temporary, task-scoped credentials to reach outside systems instead of a long-lived token sitting in an environment variable. Dockerfile support and a Vercel Container Registry also went generally available, so teams can build, store, and run standard OCI images with the docker commands they already use.
Underneath sits what Vercel calls its Agent Stack, the set of primitives it presented as the through-line for the event: a provider-agnostic AI SDK, an AI Gateway that routes across hundreds of models from one endpoint with automatic failover, a Workflow SDK for durable execution, and the isolated microVMs of Vercel Sandbox. The company also pointed to AI SDK 7, which it says is downloaded more than 16 million times a week and adds a harness layer for agents that reason, call tools, and work across files, plus generation that goes beyond text.
A second track aimed at larger customers. Vercel Passport entered public beta to keep internal apps behind an identity provider, a Security Dashboard landed in private beta for account and project posture, and Bring Your Own Cloud on AWS reached private beta so customers can run Vercel functions inside their own tenants. Vercel Agent, in public beta, watches production and proposes the permissions it needs up front while staying read-only by default, an unusually cautious posture for a tool that could otherwise be handed the keys to a live environment.
The gap worth watching is between what is generally available and what is still gated. The image and container primitives and Connect are shipping now; much of the enterprise governance layer, and Vercel Agent itself, remain in public or private beta. A platform bet on agents lives or dies on the parts that keep an autonomous system from doing damage, and for now those are exactly the pieces Vercel is still testing behind closed doors.
