Gumloop

AI agents can make you 10 times more productive by doing repetitive tasks for you. They are also smart. They can make decisions, learn skills, and use external sources. If you're not using them, you are wasting time and potential income. But these open-source agents like OpenClaw or Hermes are very technical. They are difficult to configure and they break all the time. I personally lost hundreds of dollars in OpenClaw with wrong configuration and I'm a programmer. Gumloop is the simpler alternative. It's just as powerful, but you don't have to install anything or manage a server or keep it updated. In this video, I will give you a full review and tutorial of the Gumloop agents. I will not only show you what they can do and how to configure them correctly, but also how to teach them skills, how to connect hundreds of tools, and how to use them from Slack or Microsoft Teams or via email. And at the end, you will know if these agents are the right fit for you or if you need something else. 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. Now, if you use the workflows feature in Gumloop in the past, they have been deprecated because agents can do all the same things, but they are easier to set up and more flexible. So, they decided to get rid of workflows altogether. It's all agents now. And to set up an agent, we click on this little plus and then we can see a bunch of example use cases for which we can set up agents. You can do pretty much anything with them. Those are just suggestions. We can click on a template and we get a prompt to set up this agent. You can type whatever you want in here. Just tell the agent what it's supposed to do and it will set up and configure itself. There's nothing you need to do manually. Let's look at some of my agents. For example, the AI tool corner blog post writer. The instructions for this agent were automatically generated from my prompt. I didn't have to write this myself. And what this agent does is it picks one random tool from my AI tool corner directory website and turns it into a blog post from the information it can find here and the video transcript. It then puts the blog post into my Google Drive and sends me a notification email. And for the triggers, I configure this to run once a week on Monday and pick one tool it hasn't covered yet. We could also run this daily or on demand. That's completely up to you. This is just an example. Every Monday 8:00 a.m. this will run. Pick a random tool and turn it into a blog post. Actually, I told it to pick the video with the highest number of views to make sure we cover a high performer. This runs on autopilot. Now, you can add additional triggers if you want, either at a certain time or from an event or even a web hook. Or you can just tell it to run right now in the chat. We can even use this agent from Slack or Microsoft Teams or even via email. I will show you how in a minute. Here's an example of the email I received about this new post it wrote with a direct link to the post. It even generated images using AI. So, this is no problem. And this is a well-written, well-structured post that we can put into our blog. And this is not AI slop because this is based on the content from my existing videos which I put a lot of time and effort in. And by basing this blog post on my manually created content, we make sure that it's high quality and not just AI slop. Now, of course, instead of putting this into Google Docs or Drive, we can also post it directly to any platform via a connector. I will show you the available integrations in a minute. We can also select any AI model from all the different providers including Deepseek, Anthropic, OpenAI, Grok, and so on. And of course, there's always a trade-off between intelligence and cost. So, the state-of-the-art models like Claude Opus, they are more expensive, but they are really good for complex tasks like writing a blog post. And then you have these budget models like DeepSeek, which are very cheap, and I use them for other tasks like classifying emails or sending me reports where the writing quality doesn't need to be at the very top. Or you can just keep auto mode selected, which picks the most fitting model for the task automatically. So, a cheaper model for simple tasks and a more expensive, more powerful models for the difficult ones. This agent will also on demand spawn sub-agents for specialized tasks which keeps the context window small and clean which is important for a high quality output. For example, these sub agents can do research, writing, SEO and all these little subtasks that are part of your bigger task. In the performance tab, we can not only see the credits and models this agent used. Here we can also find reflections. Those are learnings your agent came up with automatically based on your feedback for example or past mistakes. It will learn and it will put these learnings as text files into these reflections. And to make sure tasks are done properly, you can set up evaluations. Those are criteria for when a task is considered finished or successful. The way this works is that you have an AI model that checks the output of your agent and then decides if the task was finished correctly. This is a more advanced feature, but it's very useful. I will put a link to Gumloop into the video description below. When you sign up, you get a two week long free trial with credits, so you can try this out without any risk. Again, the link is in the video description. Check it out and create your own agents to make your life easier. Agents can also create skills, which are more specific instructions on how to do a task. These skills are not part of the instructions prompt that we saw earlier. Instead, they are managed separately to make sure we don't bloat the context window of the AI with too much information. Instead, these skills are loaded on demand only when the agent actually needs them. For example, this one fact checks a video before turning it into a blog post. And for this, it even does a web search. And these skills use the same specification as all the other agents like Claude Code with these skill.md files and optional scripts, assets, and references. So you can even take your existing skills for example from Claude Code and put them into your Gumloop agents. But the agent can always create these skills on demand. So if the agent makes a mistake, you can just tell the agent what it should do next time and then ask it to turn it into a skill or it will do it automatically if it's appropriate. This is how the agent learns. And as a tip, whenever you create a skill, use the most powerful model because a skill only has to be created once and you want to make sure that it's the highest quality. Here in the sidebar in the skills tab, we can manage all our skills. We can also share them with our team members and you can even synchronize them via GitHub if you use that, but it's not required. You can also manage them directly in the Gumloop interface. Agents also have a bunch of abilities by default that we saw earlier like searching the web or generating images for our blog posts. And if your agent needs to use any passwords or API keys, you can put them here into these secrets. So everything is taken care of. In the connectors tab, we can see all the tools we can connect to our agents. And this list is immense. There's basically every tool available. And even if there's no integration for it, you can always connect it via the API or via MCP. And your agents can use these tools. They can read data from there. Here you can see some examples what it can do with Gmail. But if you allow it, it can also create or update data or send emails or create drafts. It can do everything you allow it to, but you are always in control. For every connector used by this agent, we can configure more granular approval rules. For example, we can say the agent should ask us every time it uses this connector or only if it changes data, but not when it reads data and when we open this connector. Here we have even more specific rules and also access settings for our team members. So you can be extremely granular in what you allow and you can even give it rules as prompts or set specific rules for every tool of this integration. Always allow, ask each time or never allow. For dangerous actions like deleting an email, you can set it to ask each time. So you always have a human in the loop that has to allow this action before it happens. The brain are your knowledge sources. We can add a new source here. And again, there are a ton of integrations for this. We can even upload files. And in the agent configuration, we can then attach these as knowledge sources. For example, if I want this blog post writer to have my uploaded brand guidelines, I can add them as a source here. And it will always take these into account. Knowledge sources are read only. The agent can only pull information from there, but not change or delete anything unlike connectors where the agent can actually make changes. So in the brain, you want to keep everything your agents should know about like brand guidelines, documentation, support knowledge, and so on. Again, if you want to try this out yourself, the link to Gumloop is in the video description below. Artifacts are also really cool. Those are all the files our agents generated. This can be images, documents, reports, but also little websites. For example, I told my blog post agent to create a blog page and keep it in our brand design. So we use the same colors and it actually looks really good. The design is amazing. This is a fully blown blog. We can even turn this into a real website by sharing it with anyone. So we make it public. We can give it a custom domain and then anyone can reach this blog under this URL. So in other words, these agents can write code and they do it really well. But we cannot only talk to this agent via the web interface. We can also connect different channels. For example, the agent has its own email inbox and we can reach it under this address. Let's use my lead enricher agent as an example. So, this one finds new leads automatically and then it finds information about them, contact info and so on. And we can reach it via email. And you can do some cool things this way. For example, I can forward this incoming email to my lead enricher agent and tell it research this person and add them to our list of leads. 2 minutes later, we get a reply from our agent. And this agent has this lead review dashboard artifact where I have an overview of all the leads it gathered. Here's the new Christina. And I can mark her as qualified, which then drafts an email back to her. I have a separate video where I show this agent in more detail. You can also add your agents to Slack via this button. Add Slackbot. It's a simple setup. And then we can just mention our bots here via an @mention and give them tasks and they will reply directly in Slack and we can manage everything from here and even share these agents with your whole team if you configure it that way. It's your decision if only you can use them or your whole team. But this is extremely useful. You can also use Microsoft Teams if that's what your company uses. Your agents also have a hosted page which you can reach under this address. And this is basically a dedicated website to talk to this particular agent like you would do directly in the Gumloop interface. Down here we also have suggested prompts and they are configured here in the agent settings. You can add them here manually but actually the agent can generate them itself which is easier. You can also talk to your agents via an API endpoint if you are a developer or via MCP. You can also use your Gumloop agents from other AI agent tools like Claude for example. In the access settings, you can configure what your team members can do with this agent. For example, you can allow them to only use it but not make changes to it. That's up to you and what you want to allow. Gumloop also has a ton of compliance and enterprise controls. You have audit logging, optional VPC deployments if your company needs that, SOC 2 Type 2 GDPR. You can restrict models, roles and permissions. So this is a very mature tool for real companies. Now you might have noticed that here in the sidebar we have this Gumball agent. This is a special agent because this one is set up by default. It's basically your AI executive assistant. Every team member has its own Gumball and it manages your calendar, your emails. It can draft replies, prepare you for meetings, all these little daily tasks that everyone needs help with. I have a dedicated video on this Gumball agent. I will put it into the video description below. Check it out if you want. The link to a Gumloop is in the video description below. When you sign up, you get a two-week free trial and credits. So, you can try this out without any risk. Build your agents, see if it's useful for you. And if you want to see more advanced use cases and what's possible with these AI agents, check out this video here in the end card next where we build a special agent.