OpenClaw 2.0 just dropped. It's officially over...

0:00OpenClaw is back from the dead. After not shipping a single update in seven weeks, which is about three decades in AI time, they are back with OpenClaw 2.0. This is their biggest update ever. But does it make them better than Hermes Agent and Grockbot? Should you make the switch? I think the answer is going to surprise you. Let's lock in and get into it. Okay, let's just get this out of the way to begin with. About four months ago, I switched from OpenClaw to Hermes agent. The reason why I did that is because OpenClaw just became wildly unreliable for me. Every single time I updated OpenClaw, it would break and

0:39then I'd have to open up another agent, Claude Code or Codeex or something else and say, "Hey, fix OpenClaw for me." It was the most annoying process of all time. And I said, "Screw it." Even though at the time I did like OpenClaw more than Hermes, the unreliability just pissed me off so much I had to switch. So I went full-time Hermes agent. I have been using Open Claw since and I have testics. I want to know, you know, is it keeping up? Is it getting better? So I have been doing updates, but again, literally 70% plus of the updates, it just breaks when I do it, which is

1:14completely completely unacceptable. I've never experienced that with Hermes Agent. never experienced that pretty much with any other AI tool I've ever used. And here we go. I go to my OpenClaw Henry. I say, "Oopenclaw 2.0 just came out. Can you upgrade yourself?" I give the link to open. And it's not like this is some weird process like this is how I upgrade Hermes. This is how I upgrade literally every other open-source tool I have. I just give a link and say upgrade yourself. In the age of AI, this has worked with 100% success rate outside of OpenClaw. It goes, it takes the link, it starts updating, and then it just stops talking

1:50to me. I say, "What the hell is going on here?" Doesn't respond. "Are you there?" Never responds. Completely, completely broken. As unacceptable as it possibly gets, it's non-stop. It's the same story over and over and over and over again with OpenClaw. And it is so so frustrating. It's like just test your updates, people. Test your updates. And it's not like I'm editing. And in a second, I will get in I actually ended up using it and updating it. So, I will get into the review in a second, go through all the functionality. I just had the vent here. Can you let me vent a little bit, but it's like this is your

2:22biggest update ever. Do you not test this? And it's not like I was like doing crazy configurations in OpenClaw and changing things, right? Cuz that's a risk with open- source software is if you customize it a crazy amount and then you try to update, you can have merge conflicts. I'm not customizing. I'm barely using Open Claw. I'm not customizing anything and it's still breaking for me. Like come on guys, test the up test the app. This is your biggest one ever. This is your comeback story. And it breaks. So here we are. I installed OpenClaw 2.0. Fresh install on my Omachi laptop. This is brand new. No

3:00memories, no nothing. It's really unfortunate because I have like eight months of memories and skills and everything on my uh Mac Studio, which is where Henry lives, my original Open Claw, but I just do not feel like fixing Open Claw again. I've done it 30 times. I don't want to do it anymore. So, we're going with a fresh install. With that out of the way, and by for the record, for all the people that reply to every,

3:23"Oh, you just you just glaze every tool. Glaze every tool." Well, now that you can't say that anymore. So, with that out of the way, we have it installed. Let's first go over what the updates are and what's new with OpenClaw 2.0. All right, so here are the big updates in OpenClaw 2.0. Number one, multiplayer AI. And by the way, I'll get into a demo of all this in a second. I just want to make sure before we get in demo, you understand what we're cooking here with

3:46with OpenClaw 2.0. Multiplayer AI agent. You can now put your agent in group chats, in your Discord, in Slack. So you have multiple people talking to your agent, kind of like Buzz, what they were doing there. Number two, clawed subscriptions are back. I I don't know if Peter Steinberger has been negotiating with Anthropic, but uh luckily Anthropic did a pro-consumer decision, which very rare, but the cloud

4:09subscription in OpenClaw is back. Automatic skill improvement, basically bringing up to par with Hermes agent. This is like kind of what Hermes agent made their name on when they first launched. OpenClaw has finally caught up with that better sub agent UX swarm. So, this is an interesting one. When you give complex tasks, which again show you in a second, it actually spawns multiple agents, spawns multiple chats, and you have an interesting way of tracking all of that. Pretty good. And I'll compare how this works compared to like Rockbot and Hermes in a second, but it's pretty solid for conversations. You can be in the middle of a conversation and then

4:47halfway through kind of rewind, say, "Okay, let's fork it here and try different things in different ways." Widgets better UX in the chat. The chats with your OpenClaw can do many things now, include create its own widgets in the chat. It actually is a pretty good UX update for OpenClaw. We'll go through that in a second. And mass credentials, so you can type your passwords, your OpenClaw, and it will never give that password to the models or anything like that, which makes it easier to log into sites. These are the biggest updates for

5:15OpenClaw 2.0. Let's get into the demo. Then I'll tell you how this compares to Hermes and Grobbot and if you should upgrade, if you should switch everything over back to OpenClaw 2.0. So the first thing we talked about there was the multiplayer AI agent. I'm going to be honest with you, I'm not going to demo it for you. I personally just don't want to give other people access to my open claw. If you read through the blog post, they do give new permissioning functional like you can say, "Hey, anyone in the group chat, they can't like change passwords or things like that." That's fine. But one, I just

5:49don't understand the use of allowing other people to use my personal agent. And two, I'm just not going to go through that security risk. I understand there's permissioning. I understand there's different rules you can set so people can't manipulate your agent. I just don't feel like giving other people access. So, kind of the big headline feature, Open Claw, for me personally, uh, just not useful. And I'm sure for many other people, like they just they're not on teams. If you're like in an enterprise or you have a big team, I'm sure that could be useful in some sort of way. I just do not do that. I

6:20don't. My only team are like my Grockbot. So, I don't have any use for that. The second thing we had in that list was the Clawed subscription being back. I don't have much use for that either. I think GPT56 Soul is just as good as Opus. that was useful like 6 months ago when Claude was like 10,000 times better than chat GBT when it came to openclaw and other AI agents. But to OpenAI's credit, GBT 56ol is just as good uh as any of the Claude models when it comes to using agents. In fact, I think Grock is actually just as good as

6:54Claude Opus for using agents as well. So, scratch that off. Don't care about that. But let's get into some of the cool things here. What you're looking at here is the new OpenClaw web app interface. And actually most of the features they added in this OpenClaw 2.0 update is only available in the web app. You can't use like Telegram and see a lot of this, you know, sub agent orchestration or like the widgets and cool things like that in Telegram. So if you want to leverage most of this functionality, you're going to need to use the web app. And that's fine if you sit your computer. A good amount of my

7:28work though is on mobile. You know, I haven't used the mobile app for OpenClaw in a while, but this functionality is not in there, right? They need to do a lot of updates with that mobile app. So, kind of a downside. You can only use this on desktop. But let's talk about the web app. I actually like the web app. I like the updates to the web app for a few different reasons here. Number

7:49one, it pulls in your chats from codec. So if you actually hook this up to claude code or codeex, it will pull in my projects and chats from codeex. So I can work all in one place. That's really cool. Like these are other projects I was working on on my Omachi laptop and they're now here inside my OpenClaw web app. That's awesome. So now I can use my OpenClaw agents and kind of my codeex project side by side which is sick. Now let's get into the demo of the sub agents here. So, I actually gave this task to my OpenClaw agent 27 minutes ago because I wanted to come prepared this

8:26YouTube video with a nice interactive demo I can show you. Um, well, it's just been stuck here sitting here for 27 minutes not doing much. But some interesting things it did do before it seems to have crashed is number one, it actually automatically created goals. So, something really interesting here is it seems like they've integrated a lot of the chat GBT desktop app functionality into OpenClaw, which this might be one of the benefits of Peter Steinberger being hired by OpenAI is they just take a lot of the assets from OpenAI and put it into OpenClaw. As you

9:04can see here, I gave it a long prompt. It automatically created a goal. This is literally the exact same uh user interface of goals and codec. So that's cool. It created separate subtasks for that prompt I gave. Looks exactly like the way they handle goals in codecs as well. That's cool. I love it. And this is interesting. It actually created a project on the left hand side in my sessions and then made a separate chat for each of the sub aents that are working on that project. So I didn't create this pocket pilot launch. I kind of gave it a prompt that's like, "Hey, I'm building this fake app, Pocket

9:39Pilot. Do a bunch of cool things for it." It created a new project on the left side automatically and gave a chat for each of the sub agents uh to do work in here and then actually orchestrated those sub aents and gave them specific uh prompts telling exactly so like there's like this like marketing sub agent it made say hey you're a marketing analyst and then over here you can see this is a pricing analyst that I didn't give these prompts this was like the orchestrator agent that did it so this is really sweet as well this sub agent functionality is really really nice I'd say this is probably a step up over

10:13Hermes when it comes to kind of sub agent UX and UI. Hermes agent is solid when it comes to this stuff, but I think this is a level up being able to track each sub agent individually. That's pretty cool. And also kind of the goals functionality here with the task progress. That's really clean as well. I, you know, I've been saying this for a long time. I think the Chad GBT desktop app is excellent. I think it's the best in the business when it comes to AI app functionality. Seems they just kind of copied and pasted a lot of that functionality over, which I give two

10:42thumbs up. That's great. The next big feature I actually like is forked conversations. This is a really cool one. This allows you to have a conversation with your OpenClaw agent and then fork conversations at different points to try different strategies. What I mean by that is this. So in this chat, I'm saying, "Hey, can you build me a

10:59quick 3D firsterson shooter using 3JS?" But say I wanted to test this out and I wanted to use just straight up HTML or do it in Unity or do it in Unreal Engine. I can rightclick on the message. I can say fork from here and then it starts a new session at that exact point with all the context from that conversation before it. So, if I want to say, "Can you build me a a quick firsterson shooter using just vanilla HTML in one file?" I can hit enter and it will resume that chat from that exact point with all the context, everything we done before it. This is great if

11:36you're building something, if you're researching something, you want to try different strategies for the goal you're pursuing. And now you can see on the left hand side here, you can see that original chat I had and now it has a forked version under it. So, if you're building an app and you want to try building out using different technologies or build out different features and and test out different hypotheses, you can do that by forking the conversation multiple ways. This is a great unique feature. I really like it. Just checking back on this uh sub agent demo I wanted to do with the widget. It's 42 minutes and it hasn't

12:08done anything. It's completely frozen in time on the exact same step. It's very frustrating. It's very frustrating. This is the experience with OpenClaw at the moment. The reliability is it's a it single Achilles heel. It's the feedback I've gotten from 99.9% of people that have switched away from it is the reliability, the reliability, the reliability. It can't keep breaking like this. I can't have to keep going to Claude Code every 5 minutes saying, "Hey, fix OpenClaw." It's just not the user experience people want. The next big update they included inside OpenClaw 2.0 is widgets. It can now create its own custom user interfaces inside of your chat. So if you want to create

12:49dashboards, if you want to create interactive things, you can ask your agent to build widgets and actually show in line with the chat. This I think is pretty cool from a practicality perspective. I don't know what I would do with this. I spend a bunch of time going back and forth in the AI trying to figure out use cases for the widget in a chat where everything keeps moving up, keeps moving up, keeps moving up. I'm not sure what the point of having a interactive widget would exactly be. I guess if you want to quickly test apps inside of your chat, it could be pretty neat, but if I'm testing apps, I don't

13:24want to test it inside of a little chat on my screen. I kind of want to test in like a full browser experience. But there's no other agent doing this and it visually looks pretty cool. So that's neat. Checking back in with the original sub agent demo I was trying to do. We're now 48 minutes in. It's done absolutely nothing. It spun up those other chats and did research, but nothing's coming out of this at all. It's just completely frozen in time. It is so frustrating. It is so frustrating that this is the consistent experience with open claw. I am a day one open claw guy. I'm an

13:57OpenClaw guy through and through. One of my big tipping points in my career was I was like the first one to create content on OpenClaw back in January and it went super viral. It was my first millionview video. It gave me a 100,000 subscribers overnight. It helped me accomplish a lot of my dreams in my career I never thought was possible. I've been a day one open claw guy, but this is just kind of inexcusable. this unreliability. You can't release a product that just constantly breaks. This is my life. AI tools. I'm fine spending hours a day working on them, tinkering with them, things breaking, fixing, but the normie

14:35that works all day, comes home, puts their kids to bed, then has 20 minutes to use these AI tools, they don't want to update the tool, then it breaks, and spend the entire 20 minutes fixing it. And that's consistently been my experience with OpenClaw, having to fix it over and over and over again. I have literally never once updated Hermes and it broke. Hermes has updated flawlessly every single time. And I have like customized my Hermes agent way more. So

15:04there should be more merge errors. There's never been a merge error there. Grockbot, zero issues, zero breaking, a completely flawless agent experience. So let's talk about OpenClaw 2.0 you know, advantages, what it does better than others. Then we'll get into OpenClaw versus Hermes versus Grockbot, which you should be using. But number one, and

15:23there's a couple here, personality. Since day one, OpenClaw just always had the best personality. I don't know what they're doing over there. I've always just had the most fun talking to OpenClaw, and that stands true today. It just has the best personality of all the agents. The codeex integration here is excellent. You get all your codeex projects automatically put into your workspace. So you can have openclaw sessions and codec sessions side by

15:47side. That's great. I really enjoy that. I do like their sub agents UX the best out of all of the agents. I do like that you can see the separate threads spin up. You can see the chat history between all of them. I think that's really neat. Again, there's been rough edges. I've showed it to you. One of them just straight up crashed halfway through. But I do like the direction they're going with the sub aent UX and those I see are the big three advantages of OpenClaw versus the other agents. But which agent should you be using? Should you be using

16:15Hermes? Should you be using Grockbot? Let's get into that. All right. So, which should you be using between Hermes, Grockbot, and Open Claw? Hermes you should be using if you are a power user. Hermes is open source, completely configurable, completely customizable. You can use it any way you can think of. You can change the code, change the way it works, change the way it thinks. If you're a power AI agent user, Hermes is great. If you're a tinkerer, again, because it's open source, you can change any of the code. Local AI, if you want to use local AI models, I guess if you just want to use any model outside of

16:47Grock, Hermes is going to be your choice there. Like I have a Hermes powered by Quen 3.8 that's running on my 5090 computer. It's excellent. Unlimited usage, super fast. I can't do that in Grobbot. It's great and it's a reliable experience. It has never broken for me a single time. I've never updated it and just stopped working. I've never had a chat crash. It just works. It doesn't

17:09break. I like when things don't break. Grockbot is absolutely fantastic tool. Zero config. If you want to start using AI agent, you want to use out of the box, get value immediately, Grockbot is the way to go. There's no configuration necessary. It just starts working and doing things. The cloud usage, I think, is revolutionary. Each agent has its own

17:28built-in virtual machine in the cloud. It is fantastic if you have slower computers. If you just don't want your agent doing things on your computer, Grockbot is the way to go because it can do all its work in its cloud computers. It's just the easiest to use. It's the most pleasant to use. It's the most fun to use. The UX is amazing. I just love using it. It's fun using it. And I think that's a kind of like a very underrated feature of software today is like do you just have fun using it? Grockbot I just have a ton of fun managing all my agents. Open Claw I honestly at this

18:04moment in time can't think of a reason why it would be your daily driver, why it would be your number one other than like you're just kind of locked in and you have it totally customized and you have tons of memories and tools and features and functionality that like is custom to you. If you're locked in, you're locked in. And like listen, it's still a great tool. We're kind of nitpicking some things here, but it is just like so frustrating at the end of the day. It just keeps breaking. It is so annoying and I just wish they would figure it out. But really, at this

18:33moment in time in August 2026, we're about to hit September here. Uh I am leaning towards Grockbot as my daily driver. I use Grockbot for most of my knowledge work. And then I use Hermes as more of a kind of technical overseer of everything I do. If I need to set up demos, Grockbot's great for doing things in the cloud. Hermes is great for doing things locally on your computer. Hermes, I think, is the best for just doing

18:58things locally on your computer. Grockbot by default thinks in the cloud and does things on its cloud computer. But if you want to do things locally, I go to Hermes. For example, I'm recording this on my Mac Studio, but all my OpenClaw stuff is running on my Omachi computer. I just went to Hermes and said, "Hey, I need to demo this on my Mac. Can you just figure out how I can use the Open Claw that's on my laptop and show it on the computer?" And it just figured it out. It's great. So, local stuff, I'm using Hermes. Cloud

19:23stuff, I'm using Rockbot. Open Claw. I'll keep testing because I want to be able to give you good opinions on this stuff, educated opinions, but right now, I don't really see any reason to use it. It's just too frustrating of an experience. I've met Peter Steinberger twice now. I respect the guy. He's been very nice to me. I respect what he has done greatly, but clearly there's just been kind of a lack of focus on the open claw side ever since he got hired by

19:48OpenAI. I don't know what's going on. They didn't ship for 7 weeks, which is just like absolutely nuts in the AI world to not ship anything for seven weeks and then 7 weeks later, uh, the result is something I had to spend half an hour just like fixing, which is so annoying. But that's it. That's which AI agent you should be using. If you learned anything at all, leave a like

20:07down below. Let me know in the replies. Uh, which agent are you daily driving right now? You daily driving Hermes, Grockbot? What are you using? You you upgrade OpenClaw 2.0? Curious to hear. Subscribe and turn on notifications cuz I will keep you up tod date with what's going on in the AI world. If you watch this channel, you'll be an AI master. If you want to take it to the next level, link for the Vibe Coding Academy down below. It's my personal private community. You can hang out with me, ask me questions. I do live calls multiple times a week. I'm sure you'll love it. I

20:37will see you in the next