Everything Vercel Shipped at Ship 26 (June 2026)

The agent-era Ship

Vercel held Ship, its annual conference, in London on June 17, 2026, and the theme was unambiguous: this is the agent era, and Vercel wants to be the place agents deploy. The framing came with a number that explains the urgency. At the start of the year, fewer than 3% of deployments on Vercel were triggered by coding agents. Now agents account for more than half of all commits, and token volume through the AI Gateway has grown from roughly 2 million to 20 million over the same window.

"For the agent era, that's Vercel," said founder and CEO Guillermo Rauch, per SiliconANGLE's coverage. Here is what shipped that backs up the claim.

1. eve: an open-source agent framework

The headline launch is eve, an open-source framework for building, running, and scaling AI agents in production. Vercel calls it "Next.js for agents," and the design is filesystem-first: you define each agent with files under an agent/ directory, and eve compiles them into an app that runs on Vercel Functions.

A minimal agent is two files. agent/instructions.md holds the system prompt, and agent/agent.ts holds the config:

import { defineAgent } from 'eve';

export default defineAgent({
  model: 'openai/gpt-5.4-mini',
});

Tools are just files in agent/tools/, with the filename becoming the tool name. The reason eve is more than a wrapper is that production concerns ship in the framework: durable execution (sessions checkpoint each step and survive crashes and deploys), sandboxed compute, human-in-the-loop approvals, subagents, and evals. Vercel says it already runs more than 100 production agents on eve, so this is dogfooded infrastructure, not a demo. eve launched as a public preview and is still in beta.

We did a full walkthrough in our deep dive on the eve framework, including the quickstart and real code. The short version: npx eve@latest init my-agent and you have a durable, sandboxed agent scaffold.

2. Vercel Drop: deploy by dragging a folder into the browser

Vercel Drop is the antidote to "I just want this live." You deploy a file, folder, or .zip by dragging it into your browser. No Git, no Vercel CLI, no local setup.

The flow, per the Drop docs: drop a project onto vercel.com/drop, pick a team and project name, and select Deploy. Vercel creates a new project, uploads your files, and publishes straight to production with a shareable live URL, in seconds. It handles more than static files too. If Vercel detects a framework like Next.js, it builds it; files with no framework deploy as-is with no build step.

It is purpose-built for the moment where the artifact already exists and the friction is the pipeline. Drop is positioned for prototypes, one-off sites, and shipping exports from tools like Webflow, Claude Design, Google Stitch, and Bolt.new without wiring up a repository first. In an era where a generation tool can hand you a finished folder, "drag it to the browser to go live" is exactly the last mile that was missing.

3. AI Gateway: spend caps per API key

Cost governance is the quiet anxiety of running agents, and Vercel addressed it directly with budgets for AI Gateway API keys. You can now set a spend cap on any AI Gateway key, and the gateway rejects further requests on that key once the limit is exceeded, until the budget resets or you raise it. The cap applies across all providers and models running through the key, so it is a single governance lever over your whole model spend.

You can pair a key with an optional refresh period (daily, weekly, monthly, or none) to scope the limit to a window, and you can create a budgeted key from the dashboard or programmatically via the CLI:

vercel ai-gateway api-keys create --name my-key --budget 50 --refresh-period monthly

One honest detail worth knowing: the budget is a soft cap, not a hard limit. The check runs at the start of each request, so the request that crosses the limit still completes, and total spend can end up slightly over the budget. For anyone who has watched an agent loop chew through a token budget overnight, even a soft cap per key is a meaningful guardrail.

4. AI SDK 7: HarnessAgent unifies Claude Code, Codex, and Pi

The most interesting developer-facing API is HarnessAgent in AI SDK 7, a single interface for running established agent harnesses. The AI SDK has always let you switch models without rewriting your agent. Now you can switch the entire harness the same way.

Harnesses manage everything above the model call: skills, sandboxes, sessions, permission flows, compaction, runtime config, and subagents. AI SDK 7 normalizes access to those capabilities behind one abstraction. The initial experimental adapters are Claude Code, Codex, and Pi, with more coming, and every harness runs the agent in a sandboxed workspace to keep the host environment safe.

The API looks like this:

import { HarnessAgent } from '@ai-sdk/harness/agent';
import { claudeCode } from '@ai-sdk/harness-claude-code';
import { createVercelSandbox } from '@ai-sdk/sandbox-vercel';

const agent = new HarnessAgent({
  harness: claudeCode,
  sandbox: createVercelSandbox({
    runtime: 'node24',
    ports: [4000],
  }),
  tools: { /* pass custom tools */ },
  skills: [ /* pass custom skills */ ],
});

const session = await agent.createSession();

try {
  const result = await agent.stream({
    session,
    prompt: 'Check the test failures and fix the production code.',
  });
  for await (const part of result.fullStream) {
    if (part.type === 'text-delta') {
      process.stdout.write(part.text);
    }
  }
} finally {
  await session.destroy();
}

Both HarnessAgent.generate() and HarnessAgent.stream() return AI SDK-compatible results, so a harness drops into the streaming patterns AI SDK developers already use. Swap claudeCode for codex or pi and the rest of your code stays put. For anyone who has thought about meta-harness agent orchestration, this is the standard library version of that idea. It is an experimental release, so expect breaking changes between versions.

The throughline

Stack these four launches and a single thesis emerges. eve standardizes how you build an agent, HarnessAgent standardizes how you drive existing coding harnesses, AI Gateway budgets standardize how you bound the cost, and Drop standardizes the fastest possible path from artifact to live URL. Together they extend Vercel's agentic infrastructure stack from "deploy your app" to "build, run, govern, and ship your agents," which is the platform Vercel clearly intends to own.

Plenty of the launch was still beta or experimental, so treat the API surfaces as moving targets for now. But as a statement of where the platform is heading, Ship 26 was the most cohesive agent-era roadmap Vercel has put on stage.

Sources

• Introducing eve - Vercel blog
• Introducing eve, an open-source agent framework - Vercel changelog
• Vercel launches a new framework and enterprise controls for agentic AI infrastructure - SiliconANGLE
• Introducing Vercel Drop - Vercel changelog
• Deploying with Vercel Drop - Vercel docs
• Budgets for API keys on AI Gateway - Vercel changelog
• Program Claude Code, Codex, Pi and other agent harnesses with AI SDK - Vercel changelog