Gumloop

AI workflow automation is the hottest topic right now because you can automate really anything. You can let it, for example, read your blog post and automatically generate social media posts from it. Or you can let the AI organize your email inbox, draft replies to emails, and apply tags. You can connect virtually any app to any other app and use AI in the middle to execute different steps intelligently. But these tools are still difficult to set up because you have to configure and connect these different nodes. You often have to convert data types between these nodes so that they are compatible. Sometimes you have to add if-else branches. So it's almost like programming. It's still very technical and it has a learning curve, but I don't want to have to learn this stuff. I want convenience. The cool thing is more and more of these workflow builders are now adding their own AI agents that can build the workflow for you. And today we will take a look at Gumloop's agent called Gummie. Together we will build a complete workflow start to finish. Whenever we create a new meeting in Google Calendar, the flow will automatically extract the person we are having this meeting with. Then it will search all information it can find about this person and their company. And then it will draft a personalized welcome email and pull it into our Gmail account. And this is just one example of the thousands of different things you can build. We will see how easy this actually is. Can we do it just with the AI agent without having to mess around with these different nodes and learn the stuff? Will we run into any problems? Let's find out together. My name is Florian Walther and this is the AI tool corner where I review the latest AI software to find out which ones can actually improve our lives and businesses. I will put a link to Gumloop into the video description. You actually get a ton of free credits in the free account, which is more than enough to create a lot of different flows and try this out extensively. So, let's go ahead and create an account. We click on Get started and then down here in the bottom left on Get started again and then create your account. In the dashboard, we have a bunch of different links. Here we have Gumloop University which contains a lot of different tutorials. But again, we don't want to study this stuff, right? We want to use the agent to build this with just natural language. So let's ignore the university for now. You can also earn additional free credits by doing different tasks like following Gumloop on different social media platforms. You can check this out. And we also have a templates gallery here that contains different workflows. And this will give you an idea of what you can build with Gumloop. So there are tons of different workflows in here, but we want to build our own one. Here in the hub in the top right we can create an agent or a flow. A flow is a predefined number of steps. You have an input then you do something which can include AI steps and then you get an output. An agent is an AI that you can give a task and then it decides itself which steps it should take. So a flow is more deterministic and an agent makes decisions by themselves. Agents are a pretty new feature. I will make a separate video about them soon. So stay tuned and subscribe to the channel. In this video, we want to create a flow. Then we get to this canvas. And here we could now build this workflow ourselves with these different nodes and triggers. But this is very technical. When we close this, we can also use this chat box to tell the AI to build this workflow for us. I've already prepared one. Whenever I create a meeting in my Google calendar, research the attendee and their company online, then draft a warm personalized welcome email in my Gmail account. Let's send this and then the AI will get to work and build this flow for us. So the AI created a plan. It came up with all the nodes that we will need and now it step by step goes ahead and adds all these nodes to the canvas and we don't have to do anything. This is super cool. And by the way, I had a cool idea for a workflow that can actually make money. I will show it to you at the end of the video. So make sure to watch the whole video. From time to time the agent will pause because it needs input from you. For example, here in this calendar step, we have to select the calendar that we want to use as the trigger here. Right? So I select it here and then we click on continue and then the agent gets back to work. This can go on for minutes and minutes. The AI adds these different nodes. It connects them. You just have to be a bit patient. Sometimes the AI creates these custom nodes with code in them. And whenever it creates code, it wants us to confirm it. So we just click this button and it will go ahead. And this is another custom node it created. Again, we just have to confirm this and this will go ahead. So right now we have three different nodes. It's still working. These two nodes here are already connected. We have this Google calendar event reader which is the trigger of this whole flow and Apollo is a service that can find information about people and companies for you. And this is used for the data enrichment step. So we can already see it extracted the email addresses of the attendees and put it into this Apollo step. So this is how such a workflow builder works. You get different inputs and outputs and you connect these steps and decide what output goes into which input. This is how you build these different step-by-step workflows and at the end you get a result. Doing this manually can be quite complicated, but until now I didn't have to connect any of this manually yet. It inserted all these variables at the correct place automatically. Looks like our flow is finished. Let's take a look at these different steps. So it starts with the calendar event reader that is triggered whenever a new event was created. So this is how the flow starts. Then it actually branches out. So this one step finds information about the attendee and this other step first extracts the domain name of the email address and then again it uses Apollo to find information about the company. So we have information about the person itself and about the company. Then it also does a web search. So it generates a search query and again we have this output data which is a search query and it passes this on to a Perplexity step. So, Perplexity is an AI alternative to Google search that uses AI to search for information online. Again, I didn't have to configure any of this. This was all configured by the AI. And here we have an AI step. So, this sends a prompt to GPT-4.1. We can also change this for example to GPT-5 because it's better. Let's do this. And this is the prompt. Create a warm personalized welcome email for an upcoming meeting. And then it inserts all these different variables. person details, company information and so on. And here at the end of our workflow, it drafts an email in Gmail and it inserts the body and the subject. So let's try this out. First, we have to save this workflow. So that this will then trigger whenever we create a new event in our calendar. When we click on run here, I think this will actually not run because it needs a calendar event created to run. Yeah, there we go. We get an error message. We have to run this indirectly by creating a calendar event. This is why the AI tells us good news, your flow is actually configured correctly, but we have to trigger this with a calendar event. So here I'm in my calendar and let's say we have a meeting coming up here at this time discussing video ideas with Florian from codingflow.com which is me. Let's create this event and see if this triggers our flow. This will take a moment, but when we look into our dashboard and scroll down here, we can see that this is actually running because it was triggered by this calendar event. Let's wait until that has finished. So, it completed successfully. And indeed, when we look into our Gmail inbox, we now have this draft here looking forward to our meeting discussing video ideas. And here is our generated email. You can also let the workflow send the email automatically, but it's usually a better idea to have it generate a draft so you can check it and send it yourself. But there are a few things I don't like about this yet. First of all, it put the subject line here into the body. We should use this for the title of the email, right? And then we have this placeholder here. And also, the email is very long. I want something very short. We can just go back to Gummie and tell it the changes we want. The flow works, but the email is way too long. Make it two to three sentences at most. Don't put any placeholders like your name into the email. Inside the generated email body, there's also a subject line. Let's use that subject line for the draft email subject. Let's send this. And then Gummie will get back to work. So, a few minutes later, the AI updated our workflow with a bunch of new steps that pass the data out of the AI response. It also updated the prompt to make the email more concise. don't include any placeholders and then again into a draft in Gmail at the end. So let's save this and try this out again. So I create a new event in Google calendar brainstorming video ideas and collaboration opportunities again with Florian from Coding Flow. Let's save this again. It triggers the workflow automatically. Let's wait for it to finish. Okay, this time it failed. Let's click on this run and on view run to see the details. We got an error at the end. This is not good. But we can tell Gummie to fix it by clicking this button. This will now analyze the error message and then figure out a solution. And from my experience, Gummie eventually always figures out the error, which is really good. It doesn't have this tendency to paint itself into a corner like AI agents sometimes do. I was always able to get my workflows to work eventually, but sometimes I had to let Gummie fix errors multiple times in a row.

So, it says that it fixed the error. Let's save this again and create another event in our calendar. So I created a new event in the calendar with the same name. And while we are waiting, we can also give this workbook a name. A workbook is basically this whole workflow. Let's call it welcome email generator. The workflow is running. If you ever need to reset the chat to a previous state, you can do this over this roll back button, but usually that's not necessary. So this time our automation succeeded and in our email account should be a new draft. This time with a personalized subject line that fits to the topic of the calendar event and it created a much shorter email. Hey Florian, looking forward to our session to brainstorm Coding Flow video ideas with the latest US–China tariff news. This is a bit weird. So it read the latest news and it tried to find a topic related to the news to talk about. But it also figured out that Florian from Coding Flow is a developer because on Coding Flow I make programming tutorials and I also focus on React tutorials. So this is relevant information just the other stuff is a bit weird but we can fine-tune this by improving the AI prompt. The important part is that it figured out my name, my company address, it found the information about my person and it drafted this short welcome email. Actually let's try to improve that right now. We go back to our workflow. We click on ask Gummie for help to open a new chat. The automation worked, but it made a weird reference to the current news which didn't really fit the topic. Can we make the email focus on the attendee and the area of expertise? Let's send this. And the AI actually figured out the problem because here in the prompt in this prompt step, which is this one, it inserts the recent news for some reason. We don't want to have this in there. So now the AI will remove this from the prompt and then our email should be better again. Let's save the workflow. Create the same event again and now the email looks much better. Hey Florian, excited for our chat today to brainstorm video ideas and collaboration. I've been impressed by Coding Flow's web focus shift with Next.js and React, which is correct. This is what I focus on on this channel and the open source AI resume builder which was a recent video I released. I have a couple of concepts that build on that momentum. This is perfect. So, we needed a few iterations until our email was perfect, but we didn't have to configure any of this manually. This was all built by the AI chat and all the errors were fixed and all the changes were made by the AI agent as well. Of course, you can always go in here and change this manually. If you want more fine control, we have a bunch of different nodes and triggers. Some particularly interesting ones are the web scraper to scrape any website. You can also use AI image generators, for example, for social media posts. And if you are more technical, you can also connect different tools via MCP servers. And there's also a webhook node. This is more technical, but if you have your own app and a developer, you can connect your app via a webhook to a workflow. For example, when a user signs up to your app, you could trigger a workflow that, similar to our workflow we just built, gets information about the person that signed up and either sends them a welcome email or puts that information into your database, whatever you want to do with it. Gumloop also has a browser extension which you can use to pass the website that you're currently on and then send all of this information to a workflow to do whatever you want with it or you can even automate different browser actions. So you do different steps in your browser and then the AI can replicate them in the future. Another really cool feature I want to show you are interfaces. So up here we can click on add interface which puts this node into our workflow. And here's an example of how this could look. So here we have this YouTube video analyzer workflow and at the beginning we have an interface. We can edit this interface by putting different input fields and other elements into it. And this is very useful if you want to share a workflow with someone who is non-technical and doesn't want to have anything to do with all the stuff that happens in here, right? Because this can be very confusing for a person that has never used Gumloop. So with an interface, you can hide all of this because here you get a link where you only see the interface and this looks like a regular website with different input fields. So for example, for this video summarizer, we can insert the link for a YouTube video. Select the summary depth and again this can all be customized and then trigger this workflow. Here we see the progress but we don't see the workflow itself. And after this has finished we see the output here at the bottom which is a summary of this video and the possibilities what you can build this way are really endless. So I promised you to also show you a workflow that can make money. So a friend of mine resells gaming consoles. He searches for consoles and games on eBay Kleinanzeigen, which is a German classifieds site. He looks for the ones he can resell for a profit and then he resells them. And I thought, couldn't we build an AI that automates the search process? And that's what I've done here. So it starts with the website scraper which scrapes this exact website with these search filters here. We search for consoles in my state. Then it asks AI to extract the information like the pricing information, the condition, product name, title and so on. Then it runs a web search to find the market values of these different consoles. Right now this should be 2025 but we can change this manually. And then at the end it sends me an email with an AI generated response that contains these different products but only the ones that we can realistically resell for a profit. And then it sends this email to me. Let's actually run this. Here we see the details of this run. You can even look at the input and output at each step like this AI response. This way you can see what exactly happens between these different steps if you want to. You don't have to. And at the end I get this email. Isn't this cool? And here it lists the items that we could realistically sell for a profit with the most important information. Current listing price, estimated profit condition, and a direct link. And I think this is really cool. I'm going to give this to my friend to make his life a bit easier. Again, I will put the link to Gumloop into the video description. You get a ton of free credits when you sign up, so you can build a lot of different workflows and try this out. And if you have built something cool, please let me know in the comments below. I would love to hear your automation ideas. Also, stay tuned for my Gumloop Agents video. Subscribe to the channel to not miss it. And then I hope I see you in the next video. Have a nice day. Take care.