Use AI SDK 7 to move TypeScript agents from demos to production

Vercel AI SDK 7 is a confirmed major release for teams building production agents in TypeScript. The release adds durable WorkflowAgent execution, harness adapters for agent runtimes such as Codex and Claude Code, typed runtime and tool context, approval controls, and a redesigned telemetry layer. The main caveat is migration cost: AI SDK 7 requires Node.js 22 and ESM imports.

What changed

Vercel published AI SDK 7 on June 25, 2026 as a major release for building, running, integrating, and observing agents. The SDK now covers more than model calls and chat primitives: it adds production features for long-running agents, tool approvals, sandboxed execution, telemetry, speech, transcription, realtime voice, image generation, video generation, and richer file workflows.

The most important developer-facing shift is the agent layer. WorkflowAgent persists execution state between steps, so long-running jobs can survive deploys, restarts, delayed approvals, and interruptions. AI SDK Harnesses also let teams wrap external agent runtimes behind a common interface, including Codex, Claude Code, Deep Agents, OpenCode, and Pi adapters.

Key takeaways

• AI SDK 7 raises the floor for production agent work: durable execution, approvals, timeouts, sandbox hooks, telemetry, and typed context are now first-class concerns.
• The release is not a drop-in upgrade for older stacks because Node.js 22 and ESM imports are required.
• Harness adapters reduce lock-in at the agent-runtime layer, not just at the model-provider layer.
• Provider file uploads, skill uploads, MCP Apps, and broader multimodal APIs make the SDK more useful for real workflows than a text-only chat wrapper.
• Teams should test approval replay, telemetry redaction, and timeout behavior before moving sensitive automations onto v7.

Availability and access

AI SDK 7 is available through Vercel's AI SDK packages and documentation. Teams upgrading from v6 should plan a migration pass rather than treating this as a patch release. The two hard requirements are Node.js 22 and ESM imports; CommonJS require() is not supported.

Vercel also points developers to migration tooling and codemods for import and rename changes. That helps with mechanical edits, but the workflow-sensitive parts still need manual review: approval policies, tool context, telemetry settings, and long-running agent recovery.

Practical LinkLoot angle

The useful move is to treat AI SDK 7 as an agent operations upgrade, not just a framework release. If your team already has separate prototypes for chat, coding agents, MCP tools, and long-running automations, v7 gives you a clearer place to normalize runtime behavior.

Start with one bounded workflow: repository triage, release-note drafting, support-ticket summarization, or a sandboxed code-review assistant. Add timeouts, approval policies, and telemetry before expanding scope. For a broader automation map, pair this with LinkLoot's guide to AI workflow automation.

What to verify before you act

• Confirm your runtime can move to Node.js 22 without breaking deployment, local tooling, or CI images.
• Audit CommonJS usage and package boundaries before enabling ESM-only imports.
• Test whether your target harness adapter supports the permissions, sandboxing, skills, and session behavior your workflow needs.
• Review telemetry defaults so runtime context, tool context, API keys, and user data are not exposed.
• Rehearse approval replay and interruption recovery before using WorkflowAgent for destructive or customer-facing actions.

Source check

Confirmed by Vercel: AI SDK 7 is a major release, requires Node.js 22 and ESM imports, adds WorkflowAgent, typed runtime and tool context, approval controls, harness integration, telemetry changes, and expanded multimodal APIs.

Independent context: Releasebot tracks the Vercel AI SDK package feed and shows the v7 release activity around June 25 and June 27, including workflow and harness-related package updates. The AI SDK docs describe AI SDK Harnesses as a uniform API for running established agent harnesses with HarnessAgent.