Skip to main content
Back

Vercel's AI SDK 7 Closes the Gap Between Agent Demos and Production

Vercel AI SDK 7 ships durable agents, harness abstraction, MCP Apps, and full observability for TypeScript AI apps in production

Repo, Full Access
  • AI SDK 7 is live: Vercel's open-source TypeScript AI toolkit ships its most agent-focused release, targeting production reliability over new features.
  • Durable agents: New WorkflowAgent persists agent state across crashes, restarts, and browser disconnects — no custom queue infrastructure needed.
  • Harness abstraction: HarnessAgent lets you run Claude Code, Codex, or Pi through a single unified API and swap them without rewriting agent logic.
  • MCP Apps, tool approvals, and timeouts: Agents can now render custom UIs via MCP, require human sign-off before tool execution, and enforce per-step and per-tool time limits.
  • Full observability: Register OpenTelemetry once and get traces, lifecycle events, and per-step performance stats across all SDK functions; integrates with Datadog, Langfuse, Sentry, and more.
  • Free and open source: Install with npm install ai; existing SDK 6 users can auto-migrate with npx @ai-sdk/codemod v7. Docs at ai-sdk.dev.

Vercel's AI SDK just shipped its biggest release yet. Version 7 is not a new feature or a polished API surface , it is a direct answer to the wall every team hits when they try to move an agent from a demo into production. The SDK already had 16 million weekly downloads and was the foundation under tools like Thomson Reuters' CoCounsel and Clay's Claygent. Now it is tackling the hard problems: state that survives crashes, tools that need human sign-off, secrets that should never reach the model, and harnesses like Claude Code or Codex that you want to swap without rewriting your agent logic.

The gap between prototype and production

Where earlier versions laid the groundwork for multi-step tool calling and streaming, version 7 tackles the hard problems that show up when you try to run agents in production: state that survives crashes, tools that need human sign-off, secrets that should never reach the model, and harnesses like Claude Code or Codex that you want to swap without rewriting your agent logic. The release organizes all of this into five pillars: develop, run, integrate, observe, and go beyond text.

Agents that survive the real world

The biggest addition is WorkflowAgent, a native integration between the AI SDK and Vercel's Workflow SDK. The core issue it addresses is simple but painful: a standard ToolLoopAgent runs entirely in memory, so if the process crashes, all progress is lost. For any agent doing meaningful work , filing a ticket, running a code review, waiting on a human approval , that is a non-starter.

WorkflowAgent solves this by running inside a Vercel Workflow, where each tool execution is a durable step with automatic retries. Durable streams persist agent output, and getWritable() gives you a persistent stream that multiple clients can connect to, disconnect from, and reconnect to later. The workflow keeps running even if the user closes the browser, and when they come back, the client resumes exactly where the stream left off , no Redis or custom pub/sub required.