Windsurf has released a new feature!
In version 1.8.2, it introduced the workflow feature. Although workflow is not new, products like dify and coze launched workflow capabilities long ago, and it's one of their biggest features. However, Windsurf's workflow is significantly different in form from the past drag-and-drop interactions. It uses natural language to define workflows and is deeply integrated with the development IDE. The author believes that this form is the result of model capability evolution and a more advanced form of workflow. For various popular orchestration tools and automation scripts in the past, this not only points the way forward but will also be a huge challenge and a 'dimensional reduction' strike.
What is Windsurf Workflows?
Windsurf Workflows allow users to define a sequence of steps to guide Windsurf's core AI engine, Cascade, through repetitive tasks.
Key Features include:
Structured Step Sequences: Workflows allow users to define clear prompts at each step level, breaking down complex tasks into a series of interconnected subtasks or actions. This is in line with Windsurf's existing "Rules" (providing persistent, reusable contextual guidance at the prompt level), but Workflows extend this concept to the entire task 'trajectory' level.
Markdown Definition: Creating a Workflow is very simple: users can create a new one via the Windsurf interface under "Customize" -> "Workflows" -> "+Workflow", and then write a series of steps, titles, and descriptions using text (Markdown format). These Workflow files are stored in the project repository's .windsurf/workflows/ directory, making them easy for team sharing and version control.
Convenient Invocation: Once saved, users can execute them directly in Cascade via the slash command /[workflow-name].
AI-Assisted Generation: Users can also directly ask Cascade to help generate Workflows! This is especially efficient for scenarios involving multi-step operations with specific CLI tools.
Built-in Error Handling: Windsurf Workflows can even handle errors during execution. For example, if a formatting step fails, it reads the error message, attempts to fix it, and runs again.
As can be seen, Workflows have very rich potential application scenarios: for daily repetitive work that previously required manual execution based on Runbooks (operation manuals) or completion through automation orchestration tools, it now only requires a question; integrating with MCP, it can easily achieve many tasks that were previously time-consuming and labor-intensive.
Potential "Dimensional Reduction Strike" on Application Orchestration Tools like Dify
Seeing this, we can't help but think of the mainstream AI application orchestration tools represented by dify, fastgpt, etc., which primarily focus on building large model applications through visual drag-and-drop methods, greatly lowering the development threshold and bridging the gap between model capabilities and practical needs.
Workflow orchestration is currently the best compromise method to balance model capabilities and production requirements. Through orchestration, open-domain, complex multi-step problems can be broken down into multiple sub-problems to be solved separately. Models are used where possible, and workflows are used where they aren't, even adopting a "Human in Loop" approach. This white-boxes the entire workflow, limiting large model capabilities within closed problems, making large models more controllable, and improving the interpretability and robustness of the entire application. This approach has become an industry consensus.
On the other hand, we say that workflow is a transitional method because as model capabilities gradually strengthen, many problems that were previously difficult to handle will be solved. In addition, this component stitching method has, to some extent, led to the emergence of "Frankenstein monsters," and the overall system stability will relatively decrease.
However, with the advent of Windsurf Workflows, the past "sweethearts" are about to become "old ladies"; compared to the former, Windsurf Workflows have the following advantages:
More Native Developer Experience:
Text-as-Code: Workflows are defined using Markdown, which is naturally suitable for version control (like Git), easy for code review, and team collaboration. For developers accustomed to codifying everything, this is more appealing than configuring via a GUI interface.
IDE/Editor Integration: Windsurf itself seems to be a tool deeply integrated with editors or IDEs. This means developers can define and invoke Workflows in their most familiar environment without switching to a separate orchestration platform. This "no context switching" experience is a huge efficiency improvement.
Lightweight and Suitable for High-Frequency Scenarios:
Many developers' daily automation needs, such as deployment, log checking, code formatting, PR assistance, etc., may not require a full-featured standalone orchestration platform. Windsurf Workflows provide a lightweight, fast definition and invocation method that just fits these high-frequency but relatively simple automation scenarios.
For CLI toolchain automation, Windsurf Workflows' AI generation capability and direct execution capability may be more direct than configuring API calls or code nodes in platforms like Dify.
AI-Driven Workflow Self-Creation:
Although tools like Dify also utilize AI, they are more of an "execution node" in the orchestration process (like calling LLM). Windsurf Workflows go a step further; AI (Cascade) not only executes the process but can also help create and optimize the process itself. This meta-level AI application is its unique advantage.
Potential Shift from "Auxiliary Tool" to "Workflow Core":
Traditional runbooks or scripts are often isolated and require manual triggering or integration with other systems. Windsurf Workflows directly integrate these processes into the developer's core toolchain and, through natural language interaction (slash commands, AI generation), make them a dynamic, intelligent "living document" and "automation assistant.".
Product Positioning and User Group Matching
Windsurf Workflow itself is part of the programming IDE. For developers, it aligns with their working habits and allows for good management of code, environment, and versions. Drag-and-drop application development platforms have long had an awkward problem: business people don't use them effectively, and developers are reluctant to use them, leaving them in an awkward middle ground; often, upon entering the deep waters of real business, they have to be dismantled and revert to a high-code state.
Summary
Although Windsurf is in the second position, its product innovation is outstanding: first the Cascade mode (Agent), then full integration with MCP, and now the Workflows feature is another major innovation from Windsurf, which will lead the direction of programming tools. From a certain perspective, recent popular general agent products like manus are just wrappers around this model.
At the same time, from the perspective of industry technology development, this native workflow method based on large model capabilities is a dimensional reduction strike against traditional drag-and-drop workflows. Of course, products like Dify have accumulated many users and capabilities during the model capability gap period. Facing the challenge, the author believes they can develop in two directions: Downward, leveraging their accumulation to position themselves as atomic capability providers and empowering developers to integrate them; Upward, quickly learning from this idea to supplement capabilities and provide even lower-threshold AI application building tools for upper-level users (business people).
Overall, as model capabilities continue to strengthen, how platform tool developers avoid being swallowed up is something they must consider.
From another perspective, should platform tool developers transform, change tracks, and open up new avenues? Those interested are welcome to join the group discussion.
Reference: https://docs.windsurf.com/windsurf/cascade/workflows#workflows
Reply "join group" on the official account to join the discussion.