Migration Guide

Key takeaways

• Migrate from prototype AI features to a governed platform in stages, moving the most risky and valuable flows into platform primitives first rather than rebuilding everything at once.
• The stages run inventory, standardize, isolate, durable, async, and govern, each with a concrete output like a Gateway route, Sandbox boundary, or Workflow execution.
• Prioritize flows that are customer-facing, high-cost, high-volume, high-risk, or hard to debug.
• A flow is done migrating only when it has an owner, route policy, runtime boundary, logs, evaluation, fallback, and incident path.

Migration sequence for current versions

1. Confirm Node.js 24 and ESM, then review AI SDK 7 imports, callbacks, streams, and usage shapes.
2. Register telemetry and verify span receipt. Keep the golden set while separating SDK and model changes.
3. Rehearse Workflow 5 async bootstrap, hook metadata, removed APIs, and deployment pinning.
4. Move Edge runtime assumptions, including Next.js runtime = 'edge', to Node.js.
5. Edge Config is now Global Config. Use @vercel/global-config and GLOBAL_CONFIG, verifying store bindings and environment. Do not assume the older SDK supports newly connected stores.

Design rollback separately for code, configuration, active runs, and completed external side effects. This handbook revision does not upgrade an existing application's packages.

AI SDK 7 migration · Workflow 5 changes · Global Config rename · Edge→Node.js.

Migration should reduce operational risk in stages. Do not rebuild every AI feature at once; move the most risky and valuable flows into platform primitives first.

Migration Stages

Prioritization

Start with flows that are customer-facing, high-cost, high-volume, high-risk, or hard to debug.

Exit Criteria

Each migrated flow has owner, route policy, runtime boundary, logs, evaluation, fallback, and incident path.