OpenClaw for Product Managers - AI PM Community Session #85

Key Takeaways

For anybody who doesn't know what is OpenClaw, it's open source. It's self-hosted model agnostic. So, it uses I wouldn't say everything, but most models. It's multi- channelannel. So, I can connect via WhatsApp, via Slack, via Teams, via Discord, via all these other systems. The the real value here is it brings together all of your systems. It enables the system to plug in to all of these different tools you use, build custom APIs in certain cases, connect and figure out how to use the data together. It's like I said super powerful. >> Um so welcome [music] to the product management exercises community sessions. Um and thanks for joining us. Uh I know this is a last minute event but uh it is such a hot topic we wanted to run something um quickly on it. So before we get started I just wanted to um acknowledge product management exercises. If you haven't uh looked at our site please do. uh we help product managers prepare for their interviews and we also have a lot of uh AI courses for product managers. So uh both live inerson sessions on Zoom as well as uh asynchronous options. So depending on how you want to learn, we've got them teaching you how to become AI product managers, um teaching you how to build prototypes, uh teaching you how to build AI agents. So I'll put the link uh in Zoom and I think if you're on LinkedIn, you have it there as well. Um, without further ado, because we have so much to cover, um, I'm going to introduce Jason. Um, Jason is CEO of JHMG. He has built, um, his own, uh, AI assistant on White Claw, um, which is OpenClaw, sorry, which is such a hot topic right now. Uh, we're very excited to hear him speak, uh, show us how he's done that. It is a lot of material to cover. Um, so Jason is going to do his absolute best um to cover as much material as possible. So, uh, Jason, I'm going to pass it on to you. >> Let me jump over my presentation real fast and, uh, we will go from there. All right. Oops, wrong page. Let's start at the top. All right. Actually, before we do that, before we jump in and start screen sharing, let me tell you something real quick. All right. So, when I heard about OpenClaw, I was like, I don't know. It sounds It sounds dangerous. It sounds like that's Yeah, I don't know about giving AI access to everything. But then I went and I set it up and one of the problems that I had was that um let me think about word I one of the many problems I had is I run a small uh in addition to multiple other things I run a little executive coaching business and my sessions were all over the place and my my clients know that I use AI to to navigate their material one way or another. So, it was no secret. But what I had it do, the first thing I had it do was look through the some the nonsensitive documents that were sent to me, all the meeting transcripts, all of the items in my asauna board that I had assigned for myself there, and then tell me what did I miss, what were the things that I missed, and then just add it to my asauna board. And I realize really quickly that the power of AI is moving from information to execution. So that's what we're going to talk about today is information to execution. So with that said, my name is Jason Long. I think you can see my screen now. You got it? Cool. All right. My name is Jason Long. I'm a CEO. I'm a product owner. I'm a developer. I'm a UX designer. I'm a former private equity operator. I founded 10 businesses in the last 25 years. I've had two exits in the last 10 years. CEO at JHMG, CEO at Sporting Website, and I recently adopted OpenClaw as my 247 AI assistant. And I'm going to tell you today all about all of my struggles with OpenClaw and my successes. Uh, and sometimes I think there's a there you have to get to know the system and there's there have been struggles. So, I'm going to I'm going to show you what you're getting into and what uh what to avoid and how to how to fix it. All right. Oops. So, as I just said, AI up to this point has been about information. You drop something into claw or into chat GBT, you ask it for information, it gives you the information back. But it hasn't been so much as far as most users or let's say many users about the execution side. It's not like AI can go in up until now and see what's happening and then actually execute on things. But that's where we're coming that's what we're coming into right now. So before we get going today, we're going to do a technical presentation. I'm going to walk through this stuff with you. I'm going to try I think that the reason that my Zoom wasn't working is I could as I had to I'll get into a second. Um, so we're going to it's going to be a technical conversation today. We're going to see if Zuma is going to work while I SSH tunnel into the server. We'll see. Uh, but we need to talk about what you're getting into first. So, pros actual execution at the speed of AI. One of the things that I found about this is that it's so fascinating to see like an intelligence. I'm not saying it's it's an intelligence like you and I, but an intelligence using a computer that has access to just the raw data where it's talking back and forth to me and it's saying, "Oh, I tried this and it didn't work. I tried this in this direction, didn't work. Oh, I see a way over here. Oh, this is going to work. All right, great." and you start realizing how powerful it can really be because not only can it can it find the things that you can't find, but it can also build stuff that you can't build, at least not very quickly. I'm sure you guys all know development is moving incredibly fast right now. But it's it's really cool to see the that intelligence or whatever it is. I I don't want to necessarily say it's intelligence, but you know what I mean. Say see that execution happening. All right. privacy and local control. One of the things that I really like about OpenClaw is that I can run it on a local machine. And I realized after I got all this stuff set up that I should have actually done that for this meeting cuz I have it set up on one of my other local machines, but it was going to be um I already moved everything over to a um uh hardened uh remote environment, so I have access to it all the time. And it was going to be more more time than I had available to set it up for this. Somebody's messaging me, I think. Nope. Okay. Uh, no monthly SAS fees. So, there are no monthly SAS fees, but it's expensive. Uh, this is the thing that a lot of people don't talk about is that you blast through your credits unless you set it up not to. So, when we're setting this up or when I'm showing you this stuff today, one of the first things I realized after I spent like $200 in two days in credits trying to get everything, trying to figure everything out is that I had to set up different agents with different kinds of models. So, one of uh in addition to that, what I ended up doing is I moved my primary model to Gemini 3 instead of Opus because it's like a third of the cost. And then I immediately set up a fallback to Opus unless something went wrong with Gemini. And then I set up uh tons of other agents to do other things. And what I when I mean when I say when I say that, what I mean is I want one agent that's going to read my emails and get me a daily briefing. And I'm going to use like Haiku to do that cuz it's like super super cheap. Uh and then I'm going to use Sonnet maybe to do some intermediate stuff. And I'm going to use Opus or Gemini 3 or really Gemini 3 is going to sit over top of everything. and make sure it's moving. Or Opus is going to sit on top of everything, make sure it's moving. But then the others you're going to use for these smaller tasks and that cuts your cost way down, like way way way way down from like $200 in two days to like a dollar to something like that per day. Uh but that said, there's no uh there's no SAS monthly fees because you probably couldn't afford it if there were. Um you can do workflow chaining. So you can do things like first do this then do this then do this then do this and then show me the output. So things like project management become really like really this is a really powerful tool for project management. It has massive integration capabilities. Pretty much if the system you want to talk to has an API, you can just explain to to OpenCloud what you want and it will talk to the API for you and figure it all out for you. Is it that is so powerful like so so so so powerful. And it'll figure out all the little things like you you don't have to worry about like setting this stuff up. For example, in Zapier where you gota be like, "Oh, if this, then this, then this, then this, and what about that, and what about that?" You just say like, "Hey, I want this thing to talk to this thing in this way." Put together a plan on how you're going to do that. All right, great. That looks good. Click go. Done. And not only can it do that internally, can we just build the the uh the integration for you pretty quickly as well because it can see both sides really easily. Um, it's extremely customizable. But now, let's go over to the con side. It has security risks and I'm going to show you a little bit about how to check that in a second. Like there are ways to see where the security risks are. It has what's called uh open claw doctor or openclaw status and it will tell you all the security risks that you're running into. Remember it's an open open- source uh system. There's probably stuff that the system doesn't see, but we'll get into that. You've got a highly technical barrier. I'm going to talk about that in a minute. It's expensive. Setup and maintenance is a hassle and instability and reli reliability are problems. So, I was working on the thing this this morning when I was getting ready to put this together for you guys and suddenly WhatsApp just like stopped working my WhatsApp connector and I was just like, whoa, whoa, whoa, whoa, what happened? My my bot's name is Sam. I was like, Sam, you there? You there, man? No. All right. So, that's that. All right. So, all of that said, what I the way I'm using this right now, WhatsApp command center, I can get everything from everywhere just in WhatsApp or in uh Discord. Uh daily briefings, that's super easy, but it looks at everything. You know, you can set it up to look at just your stuff. You can set it up to look at your entire company's stuff. Um you can technically, if you wanted to risk this, you can read every single email in the entire company. um you and you can brief yourself on every single thing that's happening. Not necessarily recommended, but it's definitely doable. If it's a small business, two or three people, maybe it's worth it. Um uh client onboarding orchestrator. I'm still setting this one up, but there's a ton of stuff to do to bring on new clients. And you can just set up a skill set with like, I want you to do XYZ things whenever we have a new client, and it'll just boom, boom, boom, boom, knock it all out. Meeting prep assistant. Every time I'm meeting with somebody, in fact, ahead of this meeting, it sent me a message saying, "By the way, don't forget to cover this, this, this, this, this, this, and this, and you talked about these things, and you want to do this, and boom." And so, you got your meeting meeting prep. Um, we already talked a little bit about multi- aent specialists. Uh, one of my companies does website analysis and and site assessments, and it can just automate a lot of that. And a number of other things. There are other things I was kind of playing around with, uh, within the system. All right, for anybody who doesn't know, I just want to make sure I cover this. What is OpenClaw? It's open source. It's self-hosted model agnostic. So, it uses I wouldn't say everything, but most models. It's multi- channelannel. So, I can connect via WhatsApp, via Slack, via Teams, via Discord, via all these other systems. And I know this is a small thing here, but this I just want to give you an overview of how it works because we're going to get into more technical areas in just a minute. So, I'm just going to explain this piece. We're going to start getting technical on it. I I wanted to make it bigger. I don't know why. Maybe I can. There we go. That's bigger. All right. Cool. All right. Can you see it? Okay. You got it. All right. Excellent. So, what you got? You got your reasoning engine up at the top. You got your AI, your LLMs, they plug into the gateway. You got your channels that plugs into the gateway. You've got your tools, which is kind of which is inside the system that plugs into the gateway. And then you've got session persistence. So you've got the these different sessions that are happening and you can use different models in these same sessions. And then you've got sandbox execution. So you can um in well in this case I use the sandbox pieces to harden the system but I know I'm when I say I'm a developer I'm like a medium de I'm a mid-level developer. I'm not a senior developer. So let me put that out there first. I put that out there cuz there's some of the stuff in here where I have to really look it up to figure out what's going on. Uh persistent memory and then skills and knowledge. Now the the real value here unlike things like just regular claude is that really what this all does is it brings together all of your systems. It enables the system to plug in to all of these different tools you use build custom APIs in certain cases and connect and figure out how to use the data together. That's essentially what all of this is doing. But this is how it's all put together. It all comes together in this gateway. I'm not going to go about into the individual pieces in here. I'm sure there are people on this call that are better at this stuff than I am. It is really important for project managers, product owners. Um we're using it right now for Asauna, Google Calendar, Gmail, Google Drive, meeting notes, um web hook, we've got some web hooks, we've got a number of other things we it is not yet as connected to GitHub, a number of other areas. I I know a lot of people that are are connecting to notion, Confluence, Jira, all sorts of other things. So you can use it for you can if it if there's an API to write to a system or read from a system, it can use that. If there's an MCP connector, it can use that as well. So it's like I said, super powerful. All right. Now again, we're going to start getting into the technical pieces of this. How am I doing by the way on uh understanding everybody? Everybody keeping up with me so far? Everybody good? Cool. All right. All right. So, what do you need to know? You can set this thing up pretty easily on your local environment pretty fast. It's not so so bad. But I got a buddy of mine, a friend of mine, she um she called me yesterday and she was like, "Jason, my security team let me know that I'm leaking really important secrets all over the place." because she didn't know what she was doing and she set it up on a local environment and it was pretty insecure, but you can set it up pretty easily. Uh, and there's, you know, like it it's it's not uh it flows pretty easily there. However, if you want to set it up on a hardened environment, then you're going to need to it's it's going to take a little bit more skill and a lot more setup. So, keep that in mind. And we're going to talk about that in just a minute. uh skills development skill required a six out of 10 as a developer. If I'm doing it, you can pro and you have and I'm like a mediocre developer and you've got that skill, you can do it. But if you're just in the beginning, if I'm like, hey, open up PowerShell and you don't know what I'm talking about or open up the terminal and you don't know what I'm talking about, this might not be for you. Not to say you can't learn, but you're going to have to get real familiar with using the terminal, not uh a graphical interface. All right. Odds that something is going to go wrong when you're setting it up, 100%. And how long it takes to set up like an 11 out of 10. Like it is it took me I so I was working in the system that I have I had a couple of days ago and I broke it and I decided it was it was it be there was a security issue and I realized it wasn't worth it to keep on trying to fix it and I wiped the entire system and so resetting it back up. I started at like 7:30 2 days ago. I worked until midnight. Then I got up the next day and I worked the whole day pretty much to set the thing in. So like it takes a day minimum to set it up mostly because there were problems with uh with that hardened environment. All right, so now we're going to get into the real technical part of how do you set this up in a hardened environment? And I'm going to show you guys how I did it, what I did, and I'm going to give you a really quick walk through uh and give you a couple of notes on what you need to know. and what you um uh what works and what doesn't work. So, the first thing is and go over to Digital Ocean. I like Digital Ocean. Actually, we're just going to go instead of logging in here, I'm just going to go to regular Digital Ocean. And I think there it is right there. So, Digital Ocean has a um Open Claw installed, oneclick deploy. It's not one click. They just say one click. It's like a 100 clicks. Um but, you know, they like to they like to say they're oneclick deploy. And it does actually start at 12 bucks a month. So, I like that. It's a little bit slower. I actually tried setting it up on a server that was like a $30 per month server and is a bit a bit faster. But the one-click deploy on the the minimal level size at 12 bucks a month works the uh so all you're going to do if you can just follow these the instructions in here, but essentially you just click on the oneclick deploy. I think it's going to make me log in and all that. Yeah. And then click this create uh open craw open claw droplet. Once you have that, you'll log in and you'll come into something like this. Not quite like this. It'll it'll be a configuration process. It'll actually look like Got to take it down here so we don't have any security problems there. Uh it'll look something like this where you it pulls up and it starts taking you through how to see all the information you need to see. So you've got all of that and that should uh that'll get you started. Uh, and then a lot what you can do from there is you can run this right here, which is the uh, command which I'm having a hard time with. Actually got started because we talked about having problems. Uh, I have a gateway issue. And this is actually the thing I was going to tell you. If you're doing this in Digital Ocean, there is a gateway token off problem and I haven't been able to fix it yet. It still works, but stuff like the TUI just won't won't run appropriately. So, there are problems in there. But once you get in, you can run like open claw configure. So, all you do, by the way, you do have to know how to use SSH and you'll SSH into the server. All right, Ida, you told me I you said that going deep on this and and technical was okay. Am I going too technical? Is this okay? >> Yes, it's good. >> Okay, cool. Everybody, >> folks, if you have questions, please put them in the chat and I'll get them to >> throw throw them in. I I I can see the chat >> and it is being recorded so you can uh you can all watch it in your own time as well again. >> Yeah. So, uh you can go in, you can do all the configuration when you when it first gets started. It starts in the configuration. So, it sets up your um your gateway. It sets up what you know your um uh um what key you're using. So, you need to get an API key ready like an anthropic API key. Although you can use uh chat GPT anthropic and one others I one other I can't remember. Uh, I'm going to come out of this just so I don't reconfigure my server in the the wrong way. Uh, but it'll you you get started with this that sets up your configuration. If you figure out how to solve this uh token mismatch issue, please message me because I'm still trying to figure that out. And then I also recommend Oops. I also recommend uh doing uh open claw uh status just to make sure everything is running properly and everything looks good. Now there is also in addition to this there is a um there we go. So here you can see all the agents that I'm using. You can see that I'm connected on WhatsApp and Discord and you can see how everything Whoops. You can see how everything is put together there. You technically you can also just see my phone number. Uh I just took it off but um anyway yeah so you can see all of that information right in here. Uh there is also open cloud doctor which is really nice if you have to go in and debug stuff. This was one of the big lessons that I learned. Don't just talk to your bot about how to how to um uh debug itself because it will spin you in circles. Use the open claw doctor. It's way easier. Way way way easier. Cuz otherwise it's like, "Oh, change this. Oh, uh make this directory really insecure. Oh, like open all of your secrets so that they're really easy to see." And it will it will it will make your life you'll just spin you in circles for hours and hours and hours and hours. Instead, just use the open claw doctor and it will hold on one second. Let it run. Uh, and it'll show you what's going on within the system. There it goes. So, what's happening, where we're at on things, safety, integration, blah, blah, blah, all of that. If there's problems in here, you'll it'll note that there are problems in here. And then you can do open call out doctor fix and it'll actually fix some of them. Oddly enough, it's not showing the uh um gateway mismatch error anymore and I can't couldn't tell you why cuz it's definitely still there. So, that's a really good one to know. The the other thing to know when you're setting all of this stuff up is I connected read access to my systems. I'm not going to go into into the Google infrastructure because I I have a lot of uh secure stuff in there and I can't do that online in front of everybody. Um but uh I have um in fact I just realized I have to change my uh gateway token because you guys saw it in here as well. Uh anyway, I can't go in to um uh to show you my Google stuff, but the way that I set it up is first I tried doing service accounts and it wouldn't work. Like there were just too many there was something going on where it just wasn't working. I tried to do a service account, give it the prop proper scopes, tell it to to um just give it access to look at stuff. It worked for a little bit, then it just stopped working and I never could figure out why. and I spent hours and hours and hours on it. So instead, I moved over and I did OOTH instead. So I gave I gave it ooth access through Google Cloud just to read my emails and see some of the items in my drive. Um in fact, I uh and then to um uh what else? uh and then to look at my calendar and then I set up another account on a separate uh separate Google uh workspace like cuz I wanted it totally separated separate Google workspace for and gave it its own uh OOTH for its own account. So it has a Google account as well. So it can write to files, it can generate uh it can it send emails, it can invite you to calendar events, it can do all sorts of stuff. Uh, I wanted it I wanted it blocked off from all of my stuff just unless there's a problem. So, I just set it up in a separate system. I'm still investigating all of the um uh security uh security uh pieces of this. Uh, and so actually I did not set it up for my main business account cuz I still want to investigate that a little bit further. I have a secondary business account. That's what I set it up for. But it can see my calendar and I can forward stuff into the email. Still playing around with that. It can also see my Asana board which is really nice. Uh, okay. Where am I going now? Okay. I I think the reason that Zoom wasn't working for me when I first came in is because I would in order to access the hardened environment for the platform guey. I had to do an SSH tunnel into it. And for anybody who doesn't know what that means, I'm not smart enough to explain it to you. So, [laughter] you can look it up online. Uh, but essentially it's it sets up my my lo it sets up 127.00.0 whatever my local um uh um IP address to be that their local ho uh the machine's local machine. So, it looks like my browser is tunneling into their server. And I think the reason that I couldn't run Zoom and do the SSH that I had the SSH tunnel on, so for whatever reason, it wasn't letting me do Zoom. So, I can't show you all of the platform, I think. But I took screenshots in like the 10 seconds that I had before um before I I figured out what was probably going on. So, let me pull up some screenshots for you guys. All right. When you come into the main platform, it looks like this. You've got a chat system and this gives you a chat of everything that's happening. So, you can see like, oh, there's this problem over here. You have these things going over there. You know, the the different uh the different bots that are talking. Uh so, like your your WhatsApp session, your um your your Discord, your blah blah blah, whatever. I'm gonna actually got to remember I got to show you Discord in a minute how I have that all set up. Um, you can see everything here through the chat. So, that's really helpful. All right. Next one was in the channels section here on the lefth hand side here. You can set up all of your different uh channels. It's it's a bit of it's not easy. Like, it's not like uh when you set up Discord, you have to go on to Discord and set up a bot. It's actually pretty easy to do WhatsApp. WhatsApp was was quick. like it sets up a um device QR code in the system for you. You just take a picture of it and you're connected. So, I think there's like one other step in there, but it's it's not it's not like overwhelmingly difficult. Discord Discord was more of a challenge. There's like a dozen or two dozen different ways to communicate with the system. I use WhatsApp, I use Discord, we use Slack, which I didn't connect it to yet because I didn't want it give access to all the company stuff there. But those two work great. All right. Next up, I had did jobs. So, in this area, it sets up all of your jobs. So, my meeting prep scanner, my coaching call, uh, coaching call prep, daily morning brief, etc. The beautiful thing about this is you don't have to set this stuff up. All you do is you tell your bot, "Hey, set up a meeting prep scanner for me that's going to look over my meetings, review all of my last x number of things, and uh or say, review my last um download the meeting notes and transcriptions. Read those. Read the last x number of emails to and from this person. Take a look at what other people on uh other people in the company have said if you give it full um full access. and then uh put together a meeting prep for me. So, and send it to me 10 minutes before the meeting. It's really really nice. So, you just tell it that and it'll write its own jobs for you. And you can then you can you can either just ask it what' you write, what you put together, etc. Or you can go in here and you take a look at it, which is really, really, really nice. All right, next up. All right. Here. Uh, hold on. Let me see if I got the other one. This is the the agents page is where you set up your your tools, your skills, your channels, etc. I think I got screenshots for you guys here. I hope I did. Oh man, did I I ran out of time on this because I didn't realize I had it. Okay. So, when you come into here, this is where you can set up um the soul and it's actually called soul.md. Uh so, it's a soul markdown file and it's in I think the tools or the files area and where you can explain to to the bot who it is, what is the soul of this bot, what what are its values, what is it trying to do, how like how is it, how does it talk to you, what is it there's there's a number of files in there. So soul tools identity which is like my identity. So it tells them like who I am, what I'm all about, what businesses I'm running, how I'm doing it, what works for me, what you know I have bad ADHD, so like keep on top of XYZ things. Like all of that kind of information you drop into here. You've also got uh the skills in there. Skills are pre-built skills. A lot of these are like integrations. A really good example of this is going to be um uh OpenAI Whisper. So, for example, if you turn on OpenAI Whisper and then you drop your Open AI API key in there, I can now go into or I did this. I can go into my WhatsApp, which happens to be like right over here uh on my computer. Uh I can go into my WhatsApp and just give it a voice note and be like, "Hey, Sam, what's going on with blah blah blah blah whatever it is?" It'll it'll listen. It'll transcribe it and it'll send me an answer back. You can also give it things like uh a phone number. You can have it you can make it have phone calls. You can give it access to Nano Banana and make it have it make it do images for you. You can give it access to tons and tons and tons of really nice systems that will do all sorts of things for you uh and make your life really really really fast and easy. Uh it's uh you know obviously a lot of these things cost money, but if you're already paying for them and you just drop your API key in, it can smooth it can it can make things uh very smooth for you. All right, let me see if I I think that was it. Let me actually see if I can get in on this other screen real quick. Bet I can't killed the I killed the tunnel. So Oh, you know what? It's gonna Yeah, my I'm offline here. So, as soon as I go to another page, it's going to kill it. Uh, it's not like if I click on any one of these, it's probably not going to work. But maybe we'll see. So, here's the skills. So, the pre-built skills. Can you see it? Okay. You want me to make it bigger? Let me see. There you go. How's that? Cool. All right. So, password management, Apple notes, Apple reminders, bare notes, blog watcher. Let me think of some of the fun one. Cam snap. So, uh, so one of the things a lot of people are doing that I have not done yet is integrate with your smart devices and I, you know, via IoT, uh, integrations. And I think that's really cool. Actually, I would love to get it set up, uh, to connect to my, uh, the the, um, cameras and the, um, lights in my house. It's getting a little scary. It's kind of Skynet, isn't it? Yeah, it is. All right. Uh, let's see what else. Uh, coding agents. One of the things that I uh that I'm I'm I'm really excited about is getting these code a coding agent set up on a local machine. So setting up open claw uh open claw on a local machine setting up some of this stuff because then I can connect it and I can say okay now you can view my terminal now you can view uh the um the inspector and when I have problems you can just talk to claude code and you can tell it what the problem is because I'm sitting there working on stuff and claude code is like well what about this I'm like I don't I don't know man let me find that let me copy that. Let me get that. Let me get that in there for you. But I'm pretty sure if I set up some of these coding agents, it can actually watch that stuff on my local machine for me, figure out what's going on, and then drop it straight back into Cloud Code. I'll let you know if it works. I haven't set it up yet, but I would be so excited if that works because that would make um debugging stuff so much easier. Uh let's see. Gemini, let me find There's GitHub. It's not turned on right now, but you can go in and use that. Gogg, I got to talk about Gogg. I think this is Google on go. I don't know what they like it's something by from the guy that wrote open does is this is the system that enables you to talk to all the Google everything you had. There's a um the thing to know about it is to set up your GOG password not in your config file but as an environment variable. For anybody who was going to set it up as a config file is that was a bad idea. Set up as a config as a environment variable. All right. You can connect to go to places. Let me see what else. Local places. I'm seeing if there's anything else that's super fun in here. Nana banana. Nano banana. Notion of course. I know a lot of you guys are using notion and obsidian. Connect straight into those those tools. And I think that was most of the really fun stuff in here. There's tons of other stuff. I saw Spotify, Trello. There's the voice call. Uh, and then you probably will want to set up this uh WACI. Uh, I had it on. I turned it off. Uh, it's to look at the history uh for for WhatsApp. Okay, I'm going to try to click through and see if it's going to work for any of these others real quick. Oh, I got a list of the stuff that I can't show. Make sure I don't show something bad. It's got access to all my stuff. What does block mean? >> What does block >> Oh, blocked. It's not work. It's not turned on. It's not turned on. That's all. Yeah, good question. What do we talked about? We got recover chat channels usage. Usage is pretty straightforward. Cron jobs. Okay, here we go. So, in cron, oh yeah, we we talked about it, but here you can see all the cron job information. I'm so glad it's going to work. We can actually click through uh in the cron jobs. You can see all my different cron um all my different uh uh crons that are running. For anybody who doesn't know what a cron job is, it's something that happens at a particular time. So it's like every x amount of time do this thing or on this date do this thing. It's usually um every x amount of time do this thing. So meeting prep scanner, it's going to be like every blah amount of time give me the meeting prep scanner. Like do it every day. Do it before the meeting. Do it whatever. Uh, what else I got in here? Usage. So, here you can see just like how much usage you got in the system, which mine is a lot. It says zero, but it's not turned on right now. Um, and then let's see. Skills, agents. Let me see if I can go over to agents and if it's going to let me do this. Yeah, maybe it will. Files. Yeah, there you go. Okay, now we can see you got um the agents soul uh tools like all the tools it's working on identity. Pretty sure it's going to Yeah, it's not going to work which is probably good. Um you can see all of the different things that make up um you know like h how this thing thinks about everything. And so it can write these itself or you can you can write them in. I when I was resetting this up I had already set up spent a bunch of time on this. I just copy pasted, dropped it back in tools. Um, let's see. Is there anything good in here? Tools. This is more of a setup area. Skills. Now, that was skills. We already looked at channels and chronos. Okay, cool. So, that's that. And I do also definitely recommend before you get going on this reading the OpenClaw page. Just docs.openclaw.ai. It's worth a read because there's like a million ways to set this thing up. Like it's it's it is for developers and power users. Actually, I think it actually specifically says it's for developers and power users. There it is. It's for developers and power users. So, the setup on it is timeconuming. It's challenging if you don't know what you're doing and you can really screw it up. So, make sure that you read it so you understand what what you're doing. You don't make mistakes. All right, that was kind of a lot. What do you think? How'd it do? >> Good. We do have one question uh through LinkedIn and the question was, "What about setting it up through a virtual private server? Is that more secure?" >> Yeah, I don't know. Uh you know, actually it depends how the VPS is set up, you know, like there's uh there's a lot of questions on that. So, I would imagine I would imagine no, it would probably like you want something that that some pros have already hardened up for you, unless you're like a hardcore dev. So, if you're on a virtual private server that hasn't been hardened, you're probably opening yourself up not to just risks, but even more risks. So, the one of the one of the reasons that I I SSH in all the time, is because of the way that so many of the secrets are set up and how they're set up, etc., etc. Um, and and because of the way that the system is hardened, and so I have to SSH into the system rather than uh just logging into a VPS. I'm sure there's a way to do it to make it really secure because somebody's going to figure that out. I mean, it's not impossible. If you can do it on a regular server, you can do it on a VPS. But I think in the short term, you're going to want somebody who set up like a a professional has set up a hardened environment for you. >> Does that answer the question? >> I believe so. They're on LinkedIn, so uh if they can give a thumbs up, that'll be great. Um, everyone else who's on Zoom, uh, feel free if you have questions or you want, uh, Jason to show you anything else in particular, um, feel free to either throw it in the chat or, uh, or mention it and we'll give you a few minutes for your questions. >> Oh, just a quick question. Uh earlier you were showing the different MD files like tools.m MD but then I would wanted to see what is an example soul.md. I have an example for you. Uh give me just a second and I will get you some examples there. Uh, I have to separate it and I have to redact some information real quick and then I can get you what's in there. One sec. There we go. Cuz I have I made copies of these specifically if the system got wiped and I ran into these kinds of problems. uh the kinds of problems I I just got done fixing. And uh I just got to make sure I don't have any ease in here that I can't share. Uh so and I do. Uh so let me just give you an example. I'm almost done. All right, there we go. Let me uh spin up a new uh file real quick for you. And here we go. Let me know when you see my screen. Now, I have this in VS Code, but this is just a text document. So, make sure there's Yeah. So, I don't actually know how to make the make it bigger for you. Let me see if I can make this bigger. >> Anybody know how to make VS Code show bigger text? >> Do full screen. >> Do full screen. Yeah, I I mean like just the make the text larger though. >> Oh, plus something like this. >> Yeah, I just did actually. Oh, there we go. Yeah, I was trying it with the mouse. All right. Thank you. Okay. Uh so this is the agents workspace. Um home folder. Treat it that way. First run. So um when it first runs, it has this uh it it tells it to like uh follow it, figure out who you are, delete it, you won't need it again. And I think that um uh in my case, Bootstrap's already gone, but it basically is like introduce yourself. Uh get to know the person you're talking to. ask them who you want them to be. Uh there's a whole there's a whole like a whole protocol in Bootstrap that tells the bot how to interact with you the first time. Uh and then every session read the soul, read the user, uh read the uh memory, etc. Wake up. Here's some of the stuff in the memory. Write it down. No mental notes. This is really important. By the way, the system has a tendency to not write things down. So, you have to tell it, especially if you're on Sonnet uh and like Gemini uh 2.5 Flash, you got to tell it to write the stuff down. Opus will generally write it down for you. But if you don't write it down, it's going to screw it all up. Is definitely going to screw it up. Uh and it writes it into these documents here. Uh all right, safety. uh blah blah blah. Let's see what else we got here. Respond when heartbeat. Let me find some other fun stuff in here. Tools. So, this talks about like what the tools are. Uh how, you know, no markdown, you know, wrap multiple links and this. It's it's a bunch of stuff like that. But then also on some of my stuff, um yeah, like when to reach out, when to stay quiet, um tons of stuff. Here we go. Uh, the soul. This is what I was looking for. You're chat not um you're you're not a chatbot. You're becoming someone. I thought that was a a fun start to the soul. Um, poor truths, boundaries, vibe. The vibe is something a lot of this stuff is set up with the system, but you set up the vibe. And so, and you can come back and change this stuff whenever you want. I read through how they had all their stuff already set up, and I I liked it, so I didn't change it. Uh, but the uh the vibe I actually came in and I was like, I want like chief of staff, fish, efficient direct, no fluff, get things done. Um, I used to have snarky in there and then it was a little too snarky and I was like, come on, man. I'm doing the best I can here. And it was like it was like making fun of me because I couldn't debug. It's it's I couldn't debug it. And I was like, we're done with that. You're not going to be snarky anymore. Um, let's see what else. Uh, etc. So, oh yeah, this was actually really good. If I'm heading in a direction that doesn't add up or if he's unsure, ask the question, poke at it. Um, etc. Like, because at first it would just be like, "Yeah, man. Whatever you want. That sounds great." And I was like, "No, no, no, no, no. Tell me when I'm wrong. Tell me tell me if this doesn't make sense." And and I feel like it actually does a pretty good job of that. So, does that answer your question? >> Yeah, I I think it does. So, so basically you're describing the role the chatbot or or the agent should uh take on and how it should interact and behave for different situations. >> Does that sound about right? >> Exactly. >> That's exactly right. >> Okay. Thank you. We have another question for you, Jason, and it is um and again for those who also want to ask questions, throw it in the chat, mention it on Zoom, or if you're on LinkedIn, type it in LinkedIn. Um the question is, uh what's your estimate on the number of hours it takes to set it up um and or hit each milestone in the setup process? Oh my god, it takes forever to set this thing up. It depends how familiar you are familiar you are with all of the different systems that have to be set up. So, for example, if you're used to setting up clawed bots, you know, you can do it fast. If you're used to setting up um uh Google um uh projects in Google Cloud, even even if you are familiar with those, it still takes time because you still have to go in do the thing, get it set up, even if you know how to do it perfectly. I would say like if you had zero problems and you got it and and like you knew exactly what you were doing, which I'm like 90% of the way there now, it would still take me a minimum of six hours to set it up just because I have to go and do each thing and and like there and there's still it always runs into some kind of problem as you go through it. So, I'm going to say that a minimum six hours and a minimum of a hundred bucks of credits if you do it perfectly and fast. One other really important thing to to note is that you can talk to the bot, which is really nice because it knows its own system most of the time. Um, but the the more costefficient way to do it is if you're already paying for Claude or OpenAI or whatever. I like Claude, but you know, your choice. Talk to Claude. tell it to look up the documentation for the system and and tell it and and talk to the one you're paying monthly for rather than paying for tokens for because otherwise you're going to blast through all your tokens whereas you could just be using regular whatever regular LLM you're using the questions. >> Thank you. Yeah, we've got a few actually. Uh were you able to try out different LLM models? Uh which ones? How was your experience? Yeah. So I have tried a lot of different LM models not just in openclaw but also in just software. Um so that's a really broad really broad question. I'll answer it for openclaw first and then I'll go into a little bit deeper in other areas. Actually I'll cover a couple important items. Gemini is like uh twice to three times faster than claude or open AI. So you get responses a lot faster like a lot lot faster. It's also a fraction of the cost. It's like a third the cost. Uh uh Gemini Pro 3 Pro preview is also almost there or pretty close to getting there in my opinion uh against uh Opus. Maybe not 4.6 but 4.5. So I haven't seen the comparison on 4.5 4.6 because they just released it against Gemini but it's pretty good. Uh, so you know, like I I wouldn't say it's quite as good at reasoning, but is really good at knowing the Google systems. So if you're like me and you're integrating Google systems into this system, you can go and ask Gemini and it'll give you the right answer almost all the time. Uh, whereas Claude is a little more loose like sometimes it doesn't quite get it. Uh, I was still it's still 90% right, 95 97% right, but that 3% when it tells you the wrong thing, it's really annoying. Um, so uh so and then OpenAI was more aligned to uh to Claude as far as cost. I want to say it was Claude was the most then it's like Claude's up here then like OpenAI and then like Gemini's way down here. So that's that's that. But then reasoning and speed for OpenAI or for Gemini is kind of like, you know, like up here somewhere on the most recent models. So I think that Google is pulling ahead as far as that goes. Uh the the speed piece is also important. Uh don't don't underestimate that because like sitting there waiting for a while to get responses from uh from Opus can like eat up your day, you know, like it really can eat up your day and a quick response is is really nice. Um, all right. So, using the system now, uh, the primary model that I use right now is I I actually was just changing around. I want to say it's Gemini right now on top of everything as the overseer. I might actually switch it back to Opus. Either Opus or Gemini 3 as the as overseer. Uh, which does a pretty good job. Like I'm pretty happy with the way uh the way that's done. Um, then we've got uh I think the question was around like what is my experience with them? My experience is that Opus and Gemini 3 do a good job of answering questions. Opus does a better job than Gemini 3 does. It really thinks through the problems a lot. I fall back on Sonnet a lot because I want to keep my um my cost down and it it makes a big difference. And so it's way better on Sonnet or Haiku if you just have like super basic stuff. If you're like, "Look at my um look at my email and give me a brief." But here's the difference. This is a really important one. I actually was testing this today with Haiku. I was like, "All right, Haiku, give me an overview of my email." And it was like, "Here's all your stuff." And actually, I think I've got it pulled up and I'm going to show you the difference. It's an important difference. So, this is what Haiku put together for me. Now, where is it at? So, this was the problem. Recent notifications. It gave me a bunch of spam stuff. Moderation request, press ranger, Matt Diggity. Okay, not the Matt. Matt's a buddy of mine, but it's still kind of spammy. Uh, you know, we've got some other things. I was working on this accomplishments, today's schedule, which is actually right. Tomorrow's prep, etc. Um yeah, like it's um you know there's it was a nice little review, but the difference is that this is from Haiku and what um uh what we'll call it um what opus gives me is like your uh your emails are mostly just noise. You got these two important items. you got these three things that you really need to know about. You've already been on you've already been looking into XYZ other things. So, you probably just need to think about this and here's the the analysis of XYZ things that you should definitely know about. Like I've got this coaching session with Haime today. By the way, rockstar CEO Hime not uh Haime that dude is really good. Um he runs a company called Virtual Latinos. Definitely worth checking out. Um, so anyway, I got this coaching session with Haime today. Uh, uh, Opus would do like a full analysis for me and show me everything that's going on in there. >> Okay, we've got, uh, one time for one final question that we've got. Is there a limit to how many agents can be set up and how does that depend on the type of models being used? >> I don't know. I don't know if there's a limit. I I if there is a limit, I've never hit it. uh I would you know there is a rate limit so if you have many models or many sessions running you will eventually rate limit yourself I don't know if there's a builtin limit on the number of agents though but if you're using one API key and you're making too many calls you're going to get rate limited by whatever LLM you're you're uh sending stuff to and you even if you were to spread it out across all the LLMs. If you had an infinite number of bots, you would still get rate limited. >> Okay, thank you very much. Thank you so much, Jason. Thanks for joining us uh on this uh Friday afternoon. Uh thanks so much for the community for joining us. Um if you missed the beginning or want to rewatch a lot of that uh a lot of the session uh it will be on our website at product management exercises and our community sessions uh and it will go up on YouTube as well. Uh again a friendly reminder that we do have a lot of AI courses for PMs who want to learn to become an AI product manager, learn to prototype, learn to build AI agents and we have the async um courses as well. also async tran as opposed to live. Um, and that's on our website productmanex exercises.com. Thanks again. Uh, and have a lovely afternoon and you know where to find Jason. He's on LinkedIn if you want to connect with him uh after this to ask more questions. Thanks very much everyone. Take care. Have a