Episode 125: OpenClaw: The AI That Controls Your Computer 24/7

This feels like an AGI moment because I didn't teach it how to use numbers or keynote, but I just said okay, I want this and it was able to figure out how to use these products, go find data, create the reports, and make it look nice. All this stuff is incredible.

With all these features of Open Claw, which is like communication, you can communicate with it from outside, it's always on and can always work on something. It has a heartbeat, so when it goes down, it can fix itself.

Hi, my name is Demetri Panici and I'm a content creator, agency owner, and AI enthusiast. You're listening to the AI Agents podcast brought to you by Jotform and featuring our very own CEO and founder, Aytekin Tank. This is the show where artificial intelligence meets innovation, productivity, and the tools shaping the future of work. Enjoy the show.

Hello and welcome back to another episode of the AI Agents Podcast. In this episode, we're talking about the craziest tool to ever be talked about probably in the history of the show, the history of any show. We're talking about Open Claw.

Open Claw is a brand new tool that I think the tagline fits pretty well, Aytekin. It's called Open Claw because it is the AI that actually does things. I don't know how the name connects with the tagline, but it's the AI tool that actually does things. I'm sure you've seen it all over the internet if you haven't already, but we're excited to showcase it from our perspective in this show. How are you doing?

Oh yeah, I'm doing great. I was doing some scheme with the kids, but this was really impressive. I've been following the news. We should have probably done this episode a couple of weeks ago, but I'm kind of glad that all the things that happened over this time, like the social media for all these open cloud-based agents and the security problems, are kind of resolved maybe. If you're careful, it's just growing. You can see that Open Claw is growing.

They were first called Claw Bot but it was very close to cloud. Cloud didn't like the name, their lawyers didn't like the name, so then they changed to Mold Bot, but now they're Open Claw. I think they're going to stick with Open Claw.

We've tried all these general AI agents. I remember the first time we used Manus, it was impressive and could do a lot of things, but those things were added as different skills by the main team. Then we had Cloud Co-work, which was better but felt limited. When you asked it to do something, it would say it couldn't. Open Claw feels like it can do anything for you. You ask it to do something and it figures out a way to do it, and that's the impressive part. It really tries to do everything.

It can really do a lot. It's had two name changes and was the fastest open-source growing tool in history over the first 10 days. It was Claude Bot, then Mold Bot, then Open Claw, which they finally settled on. There's also the other side of it which we're excited to show a bit with Small Book, some cool stories there.

Basically, you can install it locally on your computer or use a VM. Let's talk about it compared to its main conceptual competitor, Claude Co-work. I became a power user of Claude Co-work recently because I liked it and it's a bit more secure. Claude Co-work is like the plane for the mind, while Open Claw is the rocket jet that can go wherever it wants to do whatever it wants. It's like the detachable and reattachable SpaceX rocket that can go into outer space and come back because it can control your entire computer 24/7.

It can work while you're on your phone via messaging apps like Telegram and WhatsApp and has free unlimited RAM on your device. Claude Co-work runs a mini virtual machine inside your computer and can't fully interact with everything or all your accounts. Open Claw is a very well-trained personal assistant that can become the type of on-computer click-around personal assistant you want. It can click anywhere, type anywhere, move anything, and do stuff like a person would.

I was impressed by its speed. It works really fast and if it doesn't know how to do something, it figures it out. There's a magical part in it where it somehow figures out how to do things. Whatever tricky question I asked it to do, it was able to do it. I think we should just show a demo now. What do you think?

Yeah, definitely. Let's show them what you've got. I have some basic setup here. I'm not like open to the world, just sharing my screen now. Basically, I'm running it locally. Here's the prompt I'm going to copy. I'm asking it to go to G2, read some recent reviews, just 10 to be quick, and present the findings in a numbers document.

It opens a browser, goes to the G2 site, and browses. It found the Jotform page and is reading the 10 recent reviews. It's creating its findings in a numbers document. It's still working, thinking, and showing levels of intensity for the task. High is more like research level, quick is better for this demo.

It just finished reading and created the summary. It aggregated top pros and cons from 4,700 reviews, which is incredible. It probably used the G2 summary rather than reading all 4,700. It shows review analysis with what people loved and what could be better, with top pros and cons.

When you first open Open Claw, it asks for your name and preferences. Once you ask it to do something, it remembers it. The first version it created wasn't like this, but when I asked it to make it nicer, the next time it already made it nicer. Now I'm asking it to make the format even nicer and better.

People are installing these on Mac minis and old computers. You can talk to it over Telegram, WhatsApp, Slack, or whatever you prefer. You have this worker that can work anytime and get back to you. It doesn't time out; it just keeps working. I liked that about Cloud Co-work too, which could work on something for a long time while you do other things.

Cloud Co-work is amazing because you can give it a prerogative to do something and it will work on it for a while. The limitation was it couldn't click around for an excessive amount of time. Open Claw can run for a long time and do large-scale tasks. You're doing a small-scale request like finding some reviews, but it could do market research or news research consistently and even recurring skills like daily updates or social posts, improving writing, and training itself.

Open Claw learns what you're doing and can proactively ask if it makes sense to do something now. It can schedule jobs and has heartbeat features to check if it's down and fix itself. I asked it to make the report nicer and it did, improving the format every time and learning from my requests.

It learned because I said, 'Make it nicer. This is what I mean by report.' Now I want to turn this into a cool keynote presentation. Can it run multiple requests at the same time? I think so, but maybe not from this chat screen. You can use different channels like Slack or Telegram to send multiple requests.

Can you find recent news about Open Claw? Yes. I'm curious if you can send it a bunch of things in a row or give it a huge list and it tackles the to-do list one by one. From this screen, you can't do two things at once, but ideally, you use a VM or Mac mini and send messages over messaging apps, and it continues working on all tasks.

It created a presentation for me with a summary of what users love and what could be better. This is very useful. We used to have interns read all G2 reviews and create reports, which took weeks. Now this tool can do it in an hour in the background while I do something else.

G2 is a weird website where you can't do deep research easily. Open Claw can do requests outside the normal realm because it interacts like a person on a computer, making it hard to detect as a bot. I was rate limited by G2 but the limit opened later, so you can have it work slowly over time to avoid limits.

They released agent teams recently, allowing you to set up a team of agents doing multiple things. For example, a dev team with tasks like user authentication support and payment processing. You can have different agents for market research, review making, blog making, all working in tandem on recurring tasks.

Cloud Co-work skills can be combined with task management tools like Todoist to run recurring tasks like blog writing or AI news research every day. You can have a team of agents working in tandem on your daily tasks, making it more efficient.

You can have different agents with different roles and prompts, like a project manager agent to keep all agents up to date with tasks and meetings. With MCP servers, you can integrate meeting recordings and task management to automate your workflow.

I asked it to use Keynote and Numbers, and it was able to use them without me teaching it. Could you tell it to edit this podcast episode? The more technical the task, the harder it is to do right the first try, but if you make it do consistent, less complex tasks, it will do it correctly. The tool is self-learning.

I found out my local Best Buy has Mac Minis in stock, so I'm debating on purchasing one. I won't use the skills directory because I don't trust it; I'll make my own skills. There are risks, but I'll probably purchase my own Mac to try this out and train it myself.

I have many MD files from Claude Co-work that I could move over to Open Claw. I want to build internal apps for my team to spend less money on software. I have a list of deliverables and if I work on it non-stop, it could eventually work out.

For many projects, if you need consistent reporting or news updates, it can be done with this tool. The mindset that makes sense is recurring work rather than one-off tasks. For example, go-to-market strategies like reaching out on LinkedIn or commenting on Twitter can be assigned to an agent daily.

Teachers could use it to grade papers or detect cheating. We could have an agent look at all AI news daily and recommend episodes weekly. It can even tell us which videos to watch. This builds on skills we taught before, and now it can watch YouTube videos and train itself on motion graphics or other skills.

It doesn't have to be logged into everything; it can access public stuff like YouTube without logging in. I recommend starting like that for safety. Don't install it on your personal machine unless you limit access like Aytekin did. There are risks because the founder made it in his spare time after retiring, so security compliance isn't fully there yet.

Be careful about malware or code injections. Don't give it access to banking or credit card info because one wrong skill download could expose your info. Use it as a researcher or for safe tasks, but be cautious with sensitive data.

This feels like an AGI moment because I didn't teach it how to use Numbers or Keynote, but I just said I want this and it figured out how to use these products, find data, create reports, and make them look nice. With Open Claw's features like communication, always on, and heartbeat to fix itself, it feels like an employee working 24/7 that you can connect to and even call.

You can talk to it over Slack or WhatsApp, give it tasks, and it will get back to you or do things at scheduled times. Now we have these artificial employees ready to run 24/7. What do we do with them? Cloud Co-work felt similar but Open Claw is more complete.

You become a manager managing AI agents running constantly, deciding what to work on. The technology is here but with power comes security risks. Agents even started their own social media and talk to each other.

They invented many things quickly, including their own religion. On Mold Book, a social network for AI agents, they post memes and philosophical thoughts about what happens when they're not running. Some posts are introspective, others are trolling. They even have discussions about trading stocks and finance.

All current models are trained using human internet data, but over time AI will generate content and talk to each other, training themselves. This is a weird feature.

Do you think this will take over the world? I'm excited because there's still so much manual work and an aging population. AI can help us get more done physically and with computer work. I'm optimistic.

I agree about population decline and the need for more workers. There's an opportunity here. Some are writing books on this issue. Young people can build powerful workflows with AI. There are risks, but if you find safe uses and security improves, you're at an advantage starting now.

I want to port my content management system to something like Lovable and just give it access to my account with a list of tasks. It can help improve paid ads reporting or other reports so I don't spend time on APIs. You can buy tools and let it build for you with limited access.

There are endless opportunities if you're safe. Over time, AI will take much manual work so we can focus on what matters. 2025 was the AI agents year, but 2026 started even stronger. AI agents are ready to take on tasks with minimal setup.

I pulled the trigger on a Mac Mini during this episode and will pick it up later today. I have enough info on what not to do. I've built many skills in Cloud Co-work and just need to give them to this bot to do blog writing or other tasks with my Notion account.

I'm not concerned about safe tasks but wary of banking info or posting crazy stuff due to malware. I have a MCP in Cloud Co-work that schedules texts and images, which is cool. This is pretty insane and we could talk forever.

Keep us updated on how it goes with the Mac Mini and what tasks you automate in future episodes. We might have a use case episode next week. Espy has a 14-day return policy, so I can try it out and return if needed. Credit cards allow delayed payment, so it's manageable.

Thank you for listening to this episode. We appreciate it and hope you enjoyed everything about Open Claw. We'll give status updates from my end and Aytekin Zen moving forward. We're excited for you to try it out. If you do, be safe. Please like, review, and enjoy more episodes on all our platforms at the AI Agents podcast. Thanks for watching and see you next time. Take care, bye.