內建持久性：Vercel 推出 Workflow 開發套件（WDK），解決 Serverless 逾時與狀態管理難題

### Background & Challenge: The Last Mile of Serverless In serverless architectures, developers frequently run into two major pain points: **execution time limits (Timeout)** and **state management**. When tasks require minutes or even hours to complete — such as generating long-form AI content, processing large video transcodes, or running subscription flows that wait for user confirmation — traditional serverless functions get cut off by timeouts. In the past, developers had to build their own complex queue systems, maintain databases to record each intermediate state, or adopt heavyweight external frameworks like Temporal. ### What Is Vercel Workflow? The **Workflow Development Kit (WDK)** officially launched by Vercel is a native solution designed precisely to address these pain points. It allows developers to write multi-step workflows with built-in **durability** directly in Next.js or Node.js environments using pure TypeScript/JavaScript. The core mechanism of Vercel Workflow is to decompose a large task into multiple independent **Steps**. Each step's result is automatically recorded and persisted as it executes. If a step fails, the system retries only that step rather than re-running the entire workflow, which greatly improves reliability. ### Four Core Advantages of Vercel Workflow 1. **Built-in Durability & Retries**: Every step in the workflow is fault-tolerant. If an API call fails or a network anomaly occurs, Vercel Workflow automatically performs exponential backoff retries according to configuration, ensuring the task ultimately completes successfully. 2. **Automatic State Persistence**: Developers no longer need to manually write intermediate results to a database. Data passing between steps is automatically handled and encrypted by Vercel — the next step can directly read the output of the previous one. 3. **Flexible Sleep & Delay**: Using the simple `step.sleep` command, developers can pause a workflow for hours, days, or even weeks (for example: send a promotional email 3 days after a welcome email). During the pause, no serverless compute resources or costs are consumed. 4. **Completely Bypasses Serverless Timeout Limits**: Because the workflow is decomposed into multiple independent serverless invocations, each step has its own execution time. This means the total execution time of the entire workflow is virtually unlimited, completely eliminating the 15-second or 60-second timeout curse. ### Killer Use Case for the AI Era: Building Reliable AI Agents With the rise of AI Agents, the timing of Vercel Workflow's launch is perfect. AI Agents typically need to perform multi-step reasoning, make multiple calls to different large language models (LLMs), execute tool calling, and even implement human-in-the-loop review mechanisms. Combined with the **Vercel AI SDK**, developers can easily build highly flexible AI workflows. For example: - **Step 1**: Call an LLM to generate an article outline. - **Step 2**: Send the outline to an editor (pause via email trigger, waiting for a human to click a confirmation link). - **Step 3**: Upon confirmation, continue calling the LLM to generate the full article body and auto-publish. In the past, this kind of AI process — requiring human response and spanning multiple days — was extremely difficult to implement. Now, with Vercel Workflow, it can be done in just a few dozen lines of code. ### Conclusion The launch of Vercel Workflow marks serverless development entering a new era of **Durable Computing**. It not only simplifies the development of traditional background tasks and automation workflows, but also provides robust infrastructure support for the deployment of next-generation AI applications and Agents. For developers in the Next.js ecosystem, this is undoubtedly an important tool not to be missed.