Original video in Chinese.
Key Takeaway
- Best for beginners: if you’re just starting to learn AI workflows, don’t jump straight into n8n/Dify/Coze (the barrier to entry is high and it’s easy to get lost). Start with Google Opal, which has both auto-generation and manual building modes, so you can quickly understand the logic of breaking down requirements.
- Opal’s core structure: a workflow consists of input (text/links/documents), generation (multi-node processing logic, such as research/outline/article), and output (assembly and presentation, such as web pages/Google Docs). Common functions are built into the node dropdown menus.
- Example and advantages: for example, input a topic → research materials → generate an outline → write the article → output to Docs. AI can be optimized at any time, the barrier is extremely low, and after trying it, your understanding of AI workflows improves significantly.
If you want to learn AI workflows as an ordinary person, don’t start with things like n8n, Dify, or Coze right away. It’s not that they’re bad—it’s just that the learning curve is a bit high. I don’t believe you’d look at that canvas plus a bunch of plugins and not get confused.
As a beginner, I recommend starting with Google Opal.
In the previous video, I introduced how to use the new GEM in Gemini to automatically generate workflows. The thing behind that feature is Opal. It supports both automatic generation and manual building.
For beginners, the benefit of expressing your needs through conversation and letting Opal generate it automatically is that you can see how AI understands and breaks down your requirements. That is very helpful for understanding the logic of workflows.
If you want to build it manually, that’s fine too. Opal lowers the difficulty a lot. It breaks the whole process into three parts: input, generation, and output. A classic workflow really is made up of these three parts.
At the very beginning, in the input stage, either the user provides a block of text requirements, or they provide a link or a document.
When it gets to the generation stage, AI starts processing the user’s input. In this stage, more nodes start to appear, because we want AI to generate according to the logic and steps we envision.
The final output stage is actually the assembly stage. AI takes all the things passed over from earlier and assembles them, then outputs them in the required format.
There are many forms of workflows, but logically they are roughly these three stages. Let me demonstrate it for you.
I want AI to generate a blog post based on the topic I give it.
At the very start of the workflow, I add a user input node. In this node, I provide the topic of the article.
Once there’s a topic, you can’t just generate the article right away—that would definitely not work well. I want AI to do two things:
First, research the topic. I want it to break down the topic and then search the web for various materials.
Second, combine the topic and the research results to generate an outline for the article.
So, I first add a node for topic research. In the dropdown list, we can choose Google’s preset models and functions. Here I choose Plan and Execute. This function is very useful; it breaks down our requirements.
The prompt part is simple: just explain the requirement clearly in plain language. If you’re not sure, you can click the magic wand button and let AI help you edit it.
Inside the node content, besides the prompt, we can also choose tools from the toolbox. For example, I added the tools for searching web pages and fetching web page content, because I’m doing topic research.
After the topic research node is done, I add a node to generate the article outline. This node has two inputs: the original topic and the research. With these two pieces of content, AI has enough background information to generate the outline.
OK, everything is ready. It’s time to generate the article content. I add another generation node. To generate the article content, I need to feed this node two things: the outline and the materials, which are the results of the topic research. So I connect those two earlier nodes to the new node.
Finally, in what form should this article be presented? Add an output node. In the node’s dropdown list, we can choose to make no changes, or choose to present it as a web page, or import the result into Google Docs.
That’s how simple it is to build AI workflows with Opal. Google split the entire process into three parts and put common functions like requirement breakdown, image generation, and video generation into the node dropdown menus, all to lower the barrier to building as much as possible. So as long as you sort out the requirements and logic, you can definitely build it.
And you can also let AI get involved at any time to help you optimize it. Workflows, which used to feel pretty far away from ordinary people, are really much closer now.
If you haven’t tried this before, I recommend giving Opal a shot—it will be very helpful for understanding AI.
OK, that’s all for this episode. If you want to learn about AI, want to become a super individual, and want to find like-minded people, come join our newtype community. See you in the next episode!