The OpenClaw Guide no PM is Talking About (Masterclass for AI PMs)

1. 0:00 – 1:55

   Intro

   1. AGAakash Gupta0:00

      It's up to 245,000 GitHub stars. There's over 2 million weekly visitors. OpenClaw is going totally viral.

   2. NPNaman Pandey0:07

      Today, we're going to break that down. How to set it up the right way, security stuff that seems benign but you really can't skip.

   3. AGAakash Gupta0:14

      The question on everybody's minds is how do PMs actually use this tool?

   4. NPNaman Pandey0:19

      Versus a traditional LLM, which is more reactive, here you can really take the reins of whatever you're trying to do and be proactive. Here you can set it up to execute even when it's 3:00 AM and you're asleep, you're not even at your computer.

   5. AGAakash Gupta0:32

      I know one of the toughest parts is actually setting it up. I've tried to teach a couple people this, and this is the part that once they do it, they're unlocked. So can you walk us through the easiest, no-stress setup guide you could find on YouTube?

   6. NPNaman Pandey0:43

      There's a simple three-step process that you can just use on your terminal. If you're not seeing red, you're good.

   7. AGAakash Gupta0:49

      Should you be using a VPS or Mac Mini, Telegram or Discord? What about the security risks everyone keeps talking about? We will answer all your hottest questions.

   8. NPNaman Pandey0:57

      We are going to turn any Slack channel into an AI-powered knowledge base that answers team questions from your product documentation. Every morning at 9:00 AM, OpenClaw scans your team Slack channels. It summarizes what happened overnight, identifies blockers, and posts a concise standup brief to product standup that replaces the 15-minute context gathering ritual.

   9. AGAakash Gupta1:16

      So product managers are getting one step closer to Iron Man. Before we go any further, do me a favor and check that you are subscribed on YouTube and following on Apple and Spotify podcasts. And if you wanna get access to amazing AI tools, check out my bundle, where if you become an annual subscriber to my newsletter, you get a full year free of the paid plans of Mobbin, Arise, Relay app, Dovetail, Linear, Magic Patterns, DeepSky, Reforge Build, Descript and Speechify. So be sure to check that out at bundle.aakashg.com, and now into today's episode.

2. 1:55 – 3:40

   Why PMs should care about OpenClaw

   1. AGAakash Gupta1:55

      Naman, welcome to the podcast.

   2. NPNaman Pandey1:57

      Thank you for having me.

   3. AGAakash Gupta1:58

      So we've all seen the hype. It's up to 245,000 GitHub stars. There's tons of forks. There's over 2 million weekly visitors. OpenClaw is going totally viral, and it's had a tumultuous history. It used to be called Claude Bot. Anthropic asked him to change the name. He changed it to Moe Bot. He realized that it wasn't such a good name. He na- he called Sam Altman, and Sam said, "You can call it OpenClaw." It was called OpenClaw, and then a week later, Sam bought him for over a billion dollars. And the question on everybody's minds is how do PMs actually use this tool? So why should PMs actually care about OpenClaw, Naman?

   4. NPNaman Pandey2:35

      Well, so the biggest reason would be that versus a, versus a traditional LLM, which is more reactive, here you can really take the reins of whatever you're trying to do and be proactive and plan ahead in terms of whatever you're willing to accomplish. What that looks like in practice is that instead of conversationally asking it to do things or asking it questions, here you can set it up to execute even when it's 3:00 AM and you're just, you know, you're asleep, you're not even at your computer. Finally, the biggest benefit in terms of usage itself is that it's not cloud-locked, which means that you can plug and play whichever LLM you want based on the use. If you're doing more deep research type work, you can plug in Opus versus if you want a faster response to a customer maybe, you can just use a faster... Y- you can just use the fast Gemini model.

   5. AGAakash Gupta3:24

      So I've been playing with OpenClaw for weeks now, and I know one of the toughest parts is actually setting it up. I've tried to teach a couple people this, and this is the part that once they do it, they're unlocked. So can you walk us through the easiest, no-stress setup guide you could find on YouTube?

3. 3:40 – 5:02

   Two ways to set up OpenClaw

   1. NPNaman Pandey3:40

      Yes. So when it comes to setting it up, there are two broad ways to do this, and I say broad because just the only thing in common they have is the first step. After that, they look completely different. The first of those two will only take me 15 seconds to demo. It's actually kind of crazy how easy it is, but really all you have to do... So for the first easy, almost, you know, your training wheels setup, all you have to do is go on emergent.sh. Some of you might be familiar with Emergent. It's, you know, like it's a competitor to Lovable, give or take. And what they've done here is they've put in this whole bot feature right here. So what this lets you do is literally one click your installation. It just pastes everything you have to run here. The only thing you replace is your LLM key itself, which just goes in there. You do that, hit this little enter button. It just sets it up for you. The reason why we will not focus on this approach is that, as you can probably guess, it is restrictive in nature. It does not allow you the full freedom to run OpenClaw on your, on your computer machine, use your RAM, run your LLM locally. So but in case somebody's looking for a really quick and easy out-of-the-box type setup, that is available simply by going on here.

   2. AGAakash Gupta4:58

      Cool.

   3. NPNaman Pandey4:59

      So I'll now close that. Before we start actually pasting

4. 5:02 – 8:55

   Three terminal commands to install

   1. NPNaman Pandey5:02

      commands, remember that anytime you get stuck here, this stuff seem, this stuff can be intimidating, but really if you just reverse Google whatever error you're getting, it's very easy to figure out. We might run into a few here. We'll see. But the first command you want to paste is quite simply called NPM install -g openclaw@latest. What this does is it finds out the latest version that's available. You hit Enter. I'm hoping this gets us some errors so I can live troubleshoot them, but looks like it's going fine for now. We'll see.

   2. AGAakash Gupta5:36

      You might not have NPM or something like that installed, and you might need to install that. So that's Node Package Manager versus from Node.js, and you'll run into these kind of issues when you're using terminal. The most important thing to do is persevere. Just Google or take a screenshot of your screen and drop it into Claude or ChatGPT, and they will help you.

   3. NPNaman Pandey5:56

      Perfect. So that looked like it worked. So my quick and easy way to figure this out is if you're not seeing red, you're good. Sometimes you'll get a bunch of warnings, [laughs] but they're fine and acceptable. So as long as you're not seeing bright red and, like, error, error, error, you're probably fine.The second command is, um, the openclaw onboard. Before I run this, I want to walk through what happens when you actually do this. So just for context, Peter Steinberger, who invented this, his-- the, the origin of the problem he was trying to solve was to have a bot that he could WhatsApp. So it started there in his head, but then he was like, "Why stop there?" He wanted this bot to have a complete personality of its own. And I know you can give personality to your Claude or whatever, but he really wanted to bake this into the entire soul of, of what he was creating here, which is why I think you'll find it interesting that there's actually a soul.md file that gets generated that you can impart whatever you want, you know, type of attributes to. And then the way it is configured is that you cannot skip all that stuff. Like, it'll force you to give it a name, tell it how you, how it, how you want it to interact, tell it how you want it to interact with you, and so on and so forth, right? So this is very interesting. We'll hit Enter here and see what it throws at us. Okay, great. So now we're officially in, right? So there's obviously, you know, a big security warning. I don't think anyone has read this, not even Peter Steinberger, if I were to guess. You just say yes here. This is just saying that it requires lockdown in terms of your environment. It's not really that scary. You just, um, hit left arrow, hit Enter. For this, I will-

   4. AGAakash Gupta7:39

      So that's the thing, guys. Your mouse isn't gonna work here.

   5. NPNaman Pandey7:42

      Okay.

   6. AGAakash Gupta7:42

      Don't get scared. You can either use the arrow keys on your keyboard or you can actually use Tab too, and then you just hit Enter.

   7. NPNaman Pandey7:48

      Great call-out. So here you can go down the manual route, but the reason I would not recommend that is the quick start is actually comprehensive enough for y- for the manual to start feeling like just way too much work, especially when you're trying to set it up. Also know that even if you step-- that even if you skip steps here, it's very easy to set them up later, so it's actually not a big deal at all if you pretty much just speed run through this entire thing and go from there. So next up, as we were talking about, because we said that this is agnostic of model, these are literally all of your options. I'm sure there's probably more. Even if there's something not mentioned here, there is a way to run this. Um, what I would like to do is use Google, 'cause of late I've just been a Gemini API kinda guy. Just random side note, I do feel in terms of dollar value, it gets me the farthest while still not compromising on quality that much. That's just my reasoning. I'm sure it just depends on what you do. So I'll pick Google, and then I'm gonna-- I already have my API key. If you want to create one and don't have it already-

   8. AGAakash Gupta8:54

      Today's episode is

5. 8:55 – 10:41

   Ads

   1. AGAakash Gupta8:55

      brought to you by Jira Product Discovery. If you're like most product managers, you're probably in Jira tracking tickets and managing the backlog. But what about everything that happens before delivery? Jira Product Discovery helps you move your discovery, prioritization, and even road mapping work out of spreadsheets and into a purpose-built tool designed for product teams. Capture insights, prioritize what matters, and create road maps you can easily tailor for any audience. And because it's built to work with Jira, everything stays connected from idea to delivery. Used by product teams at Canva, Deliveroo, and even The Economist, check out why and try it for free today at atlassian.com/product-discovery. That's A-T-L-A-S-S-I-A-N.com/product-discovery. Jira Product Discovery, build the right thing. Today's episode is brought to you by Vanta. As a founder, you're moving fast toward product market fit, your next round, or your first big enterprise deal. But with AI accelerating how quickly startups build and ship, security expectations are higher earlier than ever. Getting security and compliance right can unlock growth or stall it if you wait too long. With deep integrations and automated workflows built for fast-moving teams, Vanta gets you audit ready fast and keeps you secure with continuous monitoring as your models, infra, and customers evolve. Fast-growing startups like LangChain, Writer, and Cursor trusted Vanta to build a scalable foundation from the start. So go to vanta.com/aakash. That's V-A-N-T-A.com/A-A-K-A-S-H to save $1,000 and join over 10,000 ambitious companies already scaling with Vanta.

6. 10:41 – 14:27

   How to get your LLM API key

   1. NPNaman Pandey10:41

      Perfect. Yeah. So you just go on-- I mean, you literally just Google, Google Gemini key create. It'll bring you to this site that I've been on many, many times. [chuckles] I-- Wait, is it that one? Yep. So create a viewer Gemini API key. Um, actually, there's no blurring needed. We, we'll be fine. But really, you just click this create API key. You give it a name. You might wanna call it OpenClaw. And then I'm not going to do it, but when you create a key, it'll let-- it'll just generate the key. Make sure you copy it and put it somewhere safe, because sometimes it doesn't let you re-access it once you've created it. So really straightforward. I won't be doing this again, 'cause I already have a key stored separately that I'll just be using over here. So I'm just gonna hit Enter here. I'm gonna use, um, paste API key now. So, okay, great. [chuckles] So you can just paste your key. Just try to not broadcast it over the internet, and then you hit Enter. So then next up, this is also kind of what I was mentioning. It lets you choose exactly which model within Google to use. You can just go with your, like, the biggest, brightest one, which right now, as of recording, is Gemini 3 Pro. However, like, as it is mentioning, if this is going to be a Slack bot that you think end customers are going to be interacting with, they probably don't need to wait ten to fifteen seconds for each response. In that case, if you just pick any of the flashes, those are known to be really quick. So it just depends on your use case, right? For now, you know, we'll just use the just quote unquote default, which is 3 Pro. Hit Enter on thatSo the other really cool thing about OpenClaw is that it really has one gateway that then connects to all of the rest of the internet. So what this means is if you're running it on your WhatsApp versus also on your Slack, maybe you also have a Discord going for some reason, you don't need to create different gateways for that. All of it sits under one place neatly organized, and you can then also use that one gateway to monitor and even delete any of your, any of your operations that you have going. I'll also be sharing how you can see the UI version of that increase in case the terminal scares you. But what this is asking in terms of our setup is what we want to set it up first. Again, you can add onto these incrementally. You do not need to create a new gateway. It's really, really straightforward. Now, because this is catered towards prog-- uh, towards product managers, obviously I know you all love to live and breathe on Slack, so we'll just be choosing Slack socket mode. Some of you might be familiar with what socket mode is, but don't fret if you're not, 'cause we'll be going through the entire setup on Slack end-to-end for you all. So great, we can just use-- You can just have this be OpenClaw. Great. So the way this works is when I was setting this up before, these instructions are pretty bare bones, if I can, you know, put it that way. So what I did was I compiled actually easily followable instructions because the, the terminal instructions aren't the greatest. What I did was compile all of these instructions for you. If you're wondering where these came from, there's actually a TUI of OpenClaw itself where you can ask questions that you want it to do. So say you're stuck at any given step, literally anything you're trying to do, you can actually ask OpenClaw to help you figure it out, right? So this-- that's just what my source was for getting this. So we're going to be toggling back and forth between what it wants and how to get it. So as you can see here, the first thing it would like is the Slack bot token, and it always begins with an XOXB. We see here the steps that we need to follow, so we're just going to be doing that step by step. Great. So just for this demo, I actually set up a brand new

7. 14:27 – 20:24

   Full Slack integration walkthrough

   1. NPNaman Pandey14:27

      Slack for you guys, just so we would be able to go through this together because as Akash was saying, there's a, there's a decent amount of pitfalls here to dissuade even the most, um, persistent of you all. So we'll just go back here. As you can see, step one is go to api.slack.com/apps. So we'll just go there. Great. So this is what you want to do, create an app. Again, for the PMs out there, I'm sure you know what I'm doing. This is all very familiar for you all. Um, next up, like, this explains literally everything, right? So it says choose from scratch and name it OpenClaw. So we'll say from scratch. We'll say OpenClaw here, and then I'm just gonna pick the one workspace that I have. So this is straightforward enough. We'll do create app. So this is where the magic begins. Next up, you want to select click socket mode on the left sidebar. Socket mode. Like this is really dummy-proof. You literally just have to follow these steps. As you can see, enable socket mode, you just click that. Uh, anytime you're confused, just go back here. Toggle on. A box will pop up asking for a token name, call it secret token. Great. No need to overthink this. Secret-token, hit Generate. Perfect. So it generated what looks like is something we need. It starts with xapp. So we're just gonna copy it and keep it somewhere safe for now. We'll be needing it in just a minute. So this part is done. As you can see though, in our terminal setup, it wanted the Slack bot token. So what we just did was kind of a stepping stone to where we're trying to get, but we need that as well. I think it'll ask this after that. But we'll just continue down our process. So we did this. We have-- We did this done part. Now we'll go back left sidebar, OAuth and permissions. I, I lost my little window here. Great. OAuth and permissions is right here. Scroll down to scopes, bot token scopes. So yep, this is right here. So we'll be adding all of the scopes that we want our bot to perform. What this translates into is we're going to be giving it permissions to read certain channels, respond to certain channels, make sure that it has all of the background knowledge really that it needs from all over the workspace to do its job really well. So as you can see, this goes over exactly all of the permissions that we need. So I'm just gonna make this slightly thinner so we can see. So all we want to do here is, as it says, bot token scopes, make sure you don't do user token 'cause that's something different. We don't want to do that. We just want to add all of these scopes. So you click that. First up, it says chat write. You-- It-- You know, when you start typing chat, it just shows up. You click that, then you go on to the next. So IM read. You keep going until you do all of these things. Obviously, feel free to skip any of these that you wouldn't want. It does explain what e-each of these things do. I would say all of these are important, but i-if you have an organization that has rules against these, by all means, feel free to skip. Channels histories, write that. And then finally, groups history. That's-

   2. AGAakash Gupta17:27

      I think we might have missed IM write

   3. NPNaman Pandey17:30

      Great catch. And actually that is the most important one 'cause we would not be getting any messages otherwise [chuckles]

   4. AGAakash Gupta17:36

      [chuckles]

   5. NPNaman Pandey17:38

      Great. So that ends our screenshot here, but this is our final one. So we'll bring back our application. We'll just go full screen here. So once you save these, there won't be an option to save. What you want to do is go all the way up, and you'll see this Install to NP Space option that you did not have before. So you just click that. It'll ask you for these permissions. You hit Allow. Great. So now if you see this, this should now look familiar because this is exactly what we want to paste into our terminal. There are still a few more steps, but I'm curious to see at which point we'll, we'll be prompted. That's not what I wanted to do for them. But for now, we'll go back to our terminalPaste our guy, hit Enter. Great. So remember how I said the first, the stepping stone, quote-unquote? This is where we'll grab that. Thankfully, we already expected this, and we have saved this. So we'll just throw that here, hit Enter. So yep, we do want to configure channels access, and just I have the liberty to just allow all channels. Again, if you have restrictions, feel free to act accordingly. Great. So if you see here, this is actually kind of important. So these directories that it's updating, remember that these don't live on the cloud somewhere. These are all on your computer. These are all local files that are saved on your machine. And as I was mentioning before, if you think this is too much work to just go through the terminal and do this, you can do one of two things. First, you can simply ask Claude Bot, or I should say OpenClaw, to do all those things for you. Second, you can go into your Finder, go to the file path, open the file using Sublime Text or whatever have you, and do whatever changes you want it to do manually. So those two are your options. We would-- So essentially, what this is saying is that our Slack configuration is complete. When it says "configure skills" now, what it's talking about is unrelated to Slack. This pertains to one of the things that we were talking about before, which is a capability that OpenClaw has. So the two really important things that you want to set up, you can choose whether you want to do it while installation or later, are skills and tools. So the way Peter Steinberger, literally the inventor of this thing that we're demoing, chose to describe each of them was, and I quote, "Tools are organs. Can the agent do it?" Versus, "Skills are textbooks. Does the agent know how to do it?" I feel like that really, for me, really put into perspective what these things are. Again, like I said, if I just hit Yes to show you what goes on here, these are all of the skills that are-- So these are all open source. These are just people and, like, good-hearted developers of the world. They just developed these once

8. 20:24 – 23:40

   Skills vs tools explained

   1. NPNaman Pandey20:24

      Peter Steinberger already made his project live on GitHub. So a lot of these are constantly evolving. I remember seeing one for Philips Hue. Let that sink in for a second. You can use, you can WhatsApp something to a bot, and it changes the light in, in the room that you're sitting in at 4:00 a.m. to wake you up. I don't know. I-I've just-- That, isn't-- That just breaks my brain to think that, you know, that's a possibility. Like I said, we'll just skip for now. The way to select it is by hitting Space, and then if you hit Enter, it'll just skip. Again, the reason I've skipped-- Yep.

   2. AGAakash Gupta20:55

      If you hold it, Space-

   3. NPNaman Pandey20:56

      Mm-hmm

   4. AGAakash Gupta20:56

      ... it won't select it. If you just hit Enter-

   5. NPNaman Pandey20:58

      Right

   6. AGAakash Gupta20:58

      ... you, like, hit this annoying issue. So just Space, then Enter. [chuckles]

   7. NPNaman Pandey21:02

      Exactly. Yep. And then, uh, like, as I was saying, you're not really missing out because we will be going over how you can add hundreds of skills if you want. I just found it easiest to modify the text file. You can even use an LLM to generate a really long, verbose, super comprehensive skill and then paste that in versus having to do it this way. So this is just my preference. I think most people would agree. I don't think we need to set up Go Places just yet, so we'll hit No there. We're not generating images for now, so we might as well just hit No. But I think because you're using Google, one of the benefits is you can use your same API for pretty much everything. Vertex API allows you to use the same key for text generation, image generation. Even if you're trying to do video generation using Veo, you can use that. Again, it makes no sense to do this right now, so we'll just hit No. Nope, we don't want to do any of these things for now. Great. So hooks is basically what we just, we just talked about. These are the TypeScript tip that run inside the gateway when lifecycle events fire. So what this means is, once we have this all live, there will be one TypeScript file that will run concurrently to everything that's going on. Hooks is how you set up that TypeScript file. This will make sense when we're doing the demo in a little bit. Again, as I was saying, everything is easily customizable after the fact. We'll just hit Space to skip for now, hit Enter. And this is a really important step where it installs the gateway that we talked about. And finally, the moment we've all been waiting for. I just love that they say they're trying to hatch the bot. I, uh, I imagine, like, a little dinosaur hatching. I, I haven't seen this in other tools that I've played around with. It's-- I don't know. It's just, like, these little things that I appreciate so much. There is a web UI. I will be showing that in a bit. But for the full experience, we're going to be just using the hatch in TUI. So we'll hit Enter here. And as it says, "Wake up, my friend," and we'll have the first message from OpenClaw here. Great. So-

   8. AGAakash Gupta22:52

      When you're building this locally, some people say you should build this in a VPS. Some people say you shouldn't. What's the truth about that?

   9. NPNaman Pandey23:00

      So really it depends on two things. First, your usage, and second, your risk for appetite. So I'll be candid, one of the things that I was trying to do, and we will be demoing that, is there is a secure folder that you can make called "docs," and you can configure it to only interact with the documents inside the docs folder. However, when I was setting it up, I wasn't the most careful, and what ended up happening was, on my Slack, it looked through all of the files that I had on my computer, and it answered a question with a response that it really had no business knowing. So [chuckles] that is really the criteria here. If you use a VPS-- Sorry, not a VPS, but if you use the cloud, you have that one layer of

9. 23:40 – 32:26

   Local vs VPS vs Mac Mini

   1. NPNaman Pandey23:40

      abstraction that protects... it from making changes in your, on your computer, on your local files. You forgo that privilege the moment you just, you know, are not careful. I don't want to say that you're for sure you're going to be breaking things, but you just owe yourself to be that much more careful and really kind of know what you're doing, which is obviously our goal here with, you know, content like this. So hopefully, does that answer your question? Any follow-ups to that?

   2. AGAakash Gupta24:04

      So VPS, it can make it so your bot doesn't fall asleep. It can be a little bit more secure, but it doesn't have access to your local file system. And then also people use the Mac Mini, which is a totally different computer. Of all of the three, I think Mac Mini or a different computer is the best, most recommended, after that local, after that VPS. But look in your own situation. For this demo, just continue with the local, I would say.

   3. NPNaman Pandey24:26

      Yeah. The only thing I would add to that is a VPS is actually also capable of making changes to your local or whatever. So VPS is, if I, if we were to rank them in terms of risk, the riskiest, right? 'Cause your computer need not even be on and stuff could be changing inside it, and your bot could be actively creating or destroying things or, you know, just, like, running amok. So what we have here, what we're doing right now, and thankfully what happens by default, is that the moment I shut my laptop, the bot, you know, refuses or it just, like, falls asleep. So then we come back here. There's a command to wake it up again, and that's when things go live. So that is what I would recommend for most people. Again, there can be use cases where you don't-- you're in a different room, you're in a different country, and you want your computer to do something. It just depends. Yeah, to, to your point, Aakash.

   4. AGAakash Gupta25:12

      Awesome.

   5. NPNaman Pandey25:13

      So quick note on, in terms of what just happened here. So as you can see, it says I'm awake, but I'm in a bit of a blank slate. So random, I guess, little segue here. The way this is configured is it auto-runs a file called bootstrap.md, and it tells it that what it is, right? That it's OpenClaw. It has skills, it has hooks, it has a soul, a whole personality. And what it does then is that then it just kind of disappears, leaving it in like a dazed state. So which is why it's saying, "So who am I and who are you?" And this goes back to what I was sharing earlier with Peter's vision for this, to force it to not just be, you know, like, whatever, you know, ChatGPT, well, there's no personality. It's just something that you type messages to. He really wanted to hard code the experience, feeling like this is truly your companion and not just, you know, uh, somebody that you go to with your problems. So we are going to be speed running this. We don't need to recreate the movie Her here, but I'll just say this is where I like to give it a name. So you can just say y-your name is, we'll just say Fella, right? 'Cause I j- It's just something I like to say. [laughs] So you can just be like, "Hey Fella, can you," you know, blah, blah. And then, then I'll say, "I am Naman, your lord."

   6. AGAakash Gupta26:25

      [laughs]

   7. NPNaman Pandey26:25

      And I will try and spell my name correctly because that will be problematic if I don't.

   8. AGAakash Gupta26:31

      And how do you go backwards in the terminal? Some people don't know that.

   9. NPNaman Pandey26:33

      Yeah. So y- again, all arrow keys and sometimes you can hit the up arrow to invoke previous commands that you've used. But as you pointed out, there's no such thing as a mouse. You j- you're just completely reliant on your arrow keys. Great.

   10. AGAakash Gupta26:46

       Yeah.

   11. NPNaman Pandey26:46

       So I'm just gonna hit that. It's gonna ask me a couple more annoying questions.

   12. AGAakash Gupta26:50

       [laughs]

   13. NPNaman Pandey26:51

       I say annoying, but you know, you know what I mean. I appreciate the, you know, idea or whatever. Okay. Yep. So here's, yeah. So it forces you to pick one of these three options. It can be a creature, a vibe, or an emoji. Uh, do you want to help me fill this out, Aakash? Any, any of these looking good for you?

   14. AGAakash Gupta27:08

       Oh my gosh. Um-

   15. NPNaman Pandey27:09

       It says, "Just a really good fella." [laughs]

   16. AGAakash Gupta27:14

       Um, PM sidekick-

   17. NPNaman Pandey27:16

       Awesome. I like that

   18. AGAakash Gupta27:18

       ... that is a proactive helper with no emoji sig-signature. [laughs]

   19. NPNaman Pandey27:24

       I said that 10X is my career. Act like you mean business.

   20. AGAakash Gupta27:29

       [laughs]

   21. NPNaman Pandey27:30

       And no signatures ever.

   22. AGAakash Gupta27:32

       [laughs]

   23. NPNaman Pandey27:34

       It's gonna say something snarky if I were to guess, but we'll see.

   24. AGAakash Gupta27:38

       [laughs] I've never actually run it in Gemini, so this is like a little bit slightly different. I always use Codex-

   25. NPNaman Pandey27:44

       Interesting

   26. AGAakash Gupta27:44

       ... so I'm excited to see this.

   27. NPNaman Pandey27:45

       Interesting. So look, pretty much this, like, it's live. We're ready to fire. We can get into... Actually, this would be a good point to flash the UI. Like I said, at this point, my first question is, is the gateway connected or not, right? Like, that's a perfectly normal question to have. Instead of fretting, I'm just literally going to ask it, "Is our gateway live and connected?" And the good thing is, if it's not, it will actually connect it for you. One time it needed, like, a token for its gateway, which I did not know where to find. So I just asked it, "Where is the token?" And it just gave me the token. So look, it says it's live and active. Perfect. So once you're done setting up vhash-starbot, it will allow you to go to this front-end gateway, which is the UI version of the gateway, which lives on 127.0.0.1:18789. That's just the default port it uses. So this is what that looks like. I can just share a brief overview of what this all is. So the chat is really just the way you can communicate with OpenClaw itself. Usually, a-and as we discussed-The best way for it to do something is just ask it to do a certain thing, and it's really good at programming itself to do whatever you asked it to do, right? So that is still the biggest USP of why we want to use OpenClaw. Like, that is the unlock that it brings, that you don't have to sit and build all the components that you need. You just tell it that, "Hey, build this for me," and it almost architects itself, spins up other bots that it then manages. It does all of that autonomously for you. So the-- all of that happens in the chat. This overview just shows your gateway token and such, just, you know, s-stuff that you need to know, um, everything that you've been doing. A couple other things. If you ever want to change what API you're using, so say you want to swap from Gemini to Anthropic for some reason, you can go into Config. If you scroll down, you'll see this, uh, option called Secrets. This will sometimes not open. That's totally normal. If you click Raw, it'll show you what model you're using. So within Gemini, right, we, you know, we have one API key for all Gemini offerings. So if you want to use, like, a faster model, you just have to change this out, hit Save Update, and that's how you can swap out your models and such. Having said that, as we know, you can also simply go on the chat and just say, "Hey, I want you to use model name blah, blah, blah," and it will automatically program itself. So that, uh, that section finishes the setup of OpenClaw itself. Now, for our PMs, we want to obviously live and breathe within Slack. So we already set up our new app. We called it OpenClaw. You saw us do that. Now, there are some important settings that we need to make sure we have before we, we can start talking to our bot, add it to Slack channels, have it read and write messages and such. We already did the two most important steps, which were the app token and the bot token. In addition to those, you want to make sure that when you go on OAuth & Permissions, you should be able to see these bot token scopes. So the thing that's finicky with OpenClaw is that it doesn't tell you this right out of the gate that you need to do these things, which is somewhat understandable, right, 'cause most people maybe aren't trying to read groups, or they aren't trying to look at what users are trying to do on Slack. So these are the three things that we will be using for all of the demo cases that I have prepared. It can very well be the case that something that you want to do needs a certain other permissions. Adding that is really straightforward. You just click that, mention whatever you want to add. The most key important thing here is no matter what you do in Slack when it comes to your app, you always want to click Reinstall to your space, because unless you do that, it will not work. It will not persist. So make sure that no matter whatever changes you make, you always Reinstall to NP Space. Very rarely that makes this OAuth token change. It doesn't happen all the time. Very rarely it will change, in which case, make sure you just copy and paste it back to your OpenClaw gateway, which we have right here, and it will automatically update everything it needs to do in order for you to get to this OpenClaw within your app. Like, that's just how it-- So as you can see, I tested here. When you just say hi, it just responds to you. At first, it will ask you to use this key, which you, as we, as we know now, we will just go back and paste that key. And once you do that, as you can see, it'll just come alive. Now, from there, it's obviously very straightforward. It's simply a matter of adding it to whatever channels you want to add it to. Again, this is just some testing that I was doing here. But as you can see, you can just invite it to its channel and then add it. And then once you do, it'll be able to just, you know, lock and load right from that point. This section completes the setup of OpenClaw within your Slack. I'll pause real quickly, Aakash. Any questions that come to mind? Anything you want me to shed more light on?

10. 32:26 – 35:14

    Ads

    1. AGAakash Gupta32:26

       Before we dive deeper, let's talk about something every PM faces, getting alignment on product decisions. You know that feeling when you're trying to explain a user flow to engineering or justify a design choice to leadership and you're just describing it with your hands? That's where Mobbin comes in. Mobbin is the world's largest library of real-world mobile and web app designs from industry-leading apps like Airbnb, Uber, and Pinterest. Instead of spending hours taking screenshots or hunting for inspiration, you can instantly find exactly how successful products handle onboarding, paywalls, checkout flows, whatever you're facing. Over one point seven million product builders use Mobbin to benchmark against best-in-class products and show their teams proven solutions. Whether you need to convince stakeholders there's a better way to handle user activation or research how top apps approach feature discovery, Mobbin gives you the visual proof to back up your product decisions. Check out mobbin.com/aakash. That's M-O-B-B-I-N.com/A-A-K-A-S-H and get twenty percent off your first year. AI evals are one of the most important skills for PMs, and I know you know they matter. The question is, are you doing them right? Most teams are winging it with basic metrics and hoping for the best. Meanwhile, the teams that actually ship reliable AI, they've cracked the code on systematic evaluation. Today's episode is brought to you by the AI Evals for Engineers and PMs course by Hamel Hussein and Shreya Shankar. This live Maven course will teach you the battle-tested frameworks from Hamel and Shreya, who are the engineers behind GitHub Copilot's evaluation system and twenty-five-plus production AI implementations. Four weeksLive instruction. Next cohort starts July 21st. Start shipping AI that actually works. Enroll at maven.com with my code AG-PRODUCT-GROWTH for over $800 off. That's A-G-P-R-O-D-U-C-T-G-R-O-W-T-H. I hope you're enjoying today's episode. Are you interested in becoming an AI product manager, making hundreds of thousands of dollars more, joining OpenAI and Anthropic? Then you might wanna do a course that I've taken myself, the AI PM Certificate ran by OpenAI product leader Miqdad Jaffer. If you use my code and my link, you get a special discount on this course. It is a course that I highly recommend. We have done a lot of collaborations together on things like AI product strategy, so check out our newsletter articles if you want to see the quality of the type of thinking you'll get. One of my frequent collaborators, Pavel Hern, is the Build Labs leader, so you're gonna live build an AI product with Pavel's feedback if you take this AI PM certificate. So be sure to check that out. Be sure to use my code and my link in order to get a special discount. And now back

11. 35:14 – 38:06

    Slack vs TUI vs gateway dashboard

    1. AGAakash Gupta35:14

       into today's episode. When I use OpenCall in Slack versus the TUI dashboard that we're seeing in the other tab, are there any limitations, or is it fully featured in Slack?

    2. NPNaman Pandey35:25

       Great question. So the way to think about the TUI and this UI dashboard, so we can just call it the gateway dashboard because that's what OpenClaw likes to call it. So what you'll see is that whatever you talk about over here will automatically get reflected in the terminal UI as well. So it's the same thing. There's really no difference. So, you know, so hopefully that answers that part of the question. These two are the same things. Most people will want to use this just because of the benefit that pasting images unlock when you're talking to any LLM, right? It's just, it's worth a thousand words, especially in this case. So that's where-- So you can think of this as your control center, your command center, if you will. What I mean by that is, so as we were seeing here, I was running into some issues with setting this up in my channel, right? So to troubleshoot that, where am I going to go? It's not going to work if I ask OpenClaw on Slack to fix itself. So we need to go back to our control hub, which is this gateway dashboard. So really, suppose we were trying to do WhatsApp as well, right? You can't ask OpenClaw on Slack to set up your WhatsApp for you. Actually, as I say that, maybe it will be able to, but even for that, you'll need an entry point. So really, this gateway dashboard is your entry point to any and all of the other app that you want to leverage OpenClaw on, be it WhatsApp, be it Slack, be it Telegram, all of those things, like the first hub that you go out into the world is still this. Does that answer your question?

    3. AGAakash Gupta36:44

       Yep. So are we fully set up? Are we ready to see it in action?

    4. NPNaman Pandey36:47

       We're ready to see it in action, actually, and, and there will be some other things that we'll be learning, but the way I have it constructed is we can just go through them as we go through our use cases itself. So with that, I'll launch into use case one, and as is tradition on your channel, we will be leveling up, right? So we'll start kind of with dipping our toes a little bit, just exploring what's possible. And then for the final use case, I have something that hopefully will leave at least some of your viewers' jaws on the floor. That's r- that's my goal with, with all of this. So yeah, let's just, let's jump right in. So use case one, and I can throw this on the screen as I talk through this. So use case one is that we're going to turn any Slack channel into an AI-powered knowledge base that answers team questions from your product documentation. So what we'll be doing is we'll be dropping our PRD, FAQ documentation, any product wiki into the workspace folder, which I'll show you where that is and how to find it. And then anyone in the channel can simply mention the bot to get instant contextual answers. Why should PMs care? Because it just simply eliminates the product questions channel problem where PMs have to spend hours answering the same questions about feature specs, launch dates, and edge cases. Again, it's important to note here that this repository is not locked in. It lives and breathes as a living space. You can add and remove stuff to it. The memory persists. So if you ask it, "Hey, what is an FAQ that we added three months

12. 38:06 – 47:47

    Slack knowledge base (use case 1)

    1. NPNaman Pandey38:06

       ago that wasn't there six months ago?" It will be able to tell you that, right? Because it's not just locked in time. It lives and evolves with you. So we will be basically building this out for our first use case. So now as we do that, the first question that naturally is probably arising for our listeners is, "What is this workspace that I mentioned?" So this is tied in perfectly to the really cool capabilities of OpenClaw, 'cause I'm sure at least some of you are wondering what exactly is the USP or the use case for using OpenClaw against Slackbot, right?

    2. AGAakash Gupta38:36

       Yeah.

    3. NPNaman Pandey38:36

       'Cause if it's just answering questions on a channel, Slackbot could probably be configured to do that as well, right? Here's the difference. Slackbot does not have access to local files that live on your computer. Neither does it have the ability to read or write into those files. So this folder here, which is just called .openclaw, and some of you will struggle in trying to reach it, so I'll show you a quick tip on how you can easily access it. So if you click your little Finder icon, right-click it, sorry, and then if you go to Go to Folder, if you just press a period, it should show you an option to go to OpenClaw. This is really annoying because it doesn't-- It's not like a folder that you can just go to, at least the, not on my computer. So it took me, like, ten minutes, annoyingly, to figure out how to get here. But now if you simply click that, it'll take you straight to this folder that we have here. Now, yours will not look like this exactly because I've been, as I said, playing around with it a decent amount. What you need to know is this workspace component. This you will find already pre-set up for your, for your own computer as soon as you install OpenClaw with some of these files, such as agents, heartbeat, soul, tools, memory, and user. What are these? I have these written here real quick. Soul.md is your deep personality or values. Identity is the agent identity. User is just all of your preferences, which in my case, as we saw, we had it be really direct, not, no beating around the bush, super to the point. Memory is stuff that persists, so even when you close your bot, you shut your computer down, this stuff still contains to be alive, continues to be alive. And then tools are your local configuration notes. And finally, heartbeat, which is really important, is a list of all of your cron jobs that it signals that, "Hey, now is the time to run this recurring task." So I can just open one of these just for you guys to look at. So we'll look at heartbeat. You can simply open with, um, Sublime Text. Otherwise, it tries to open Xcode, which is kind of annoying. Some of you might not have that. But look, this is all pre-set up. This is something that I set up, so this you will not have. So originally, your heartbeat.md will be empty, right? 'Cause you don't have any activities that you want it to do automatically. One other cool thing about this is if you go back to your gateway dashboard, and if you go to cron jobs, here is also a really cool way to see all of the activities that you have running, along with the ability to edit or modify it. So we'll be going over a use case for this in just a minute here. But for now, as we were discussing, we wanted to have a repository for our files where we can throw our documents and then query the Slack channel. So in my case, I simply just made a document or, or a folder called docs. Once I go into it, it has this PRD document, which I can flash. The quote, unquote "made-up/semi-real product" that we'll be working with through these demos is something that I'm actually working on. It's really just a one-click setup of websites for podcasters just using their RSS feed. So that's my whole big idea. You just-

    4. AGAakash Gupta41:28

       Cool

    5. NPNaman Pandey41:28

       ... you just submit your RSS feed. It gives you a fully functional, really nice-looking website. So what I did was I just generated a PRD for that. As you can see, it's just, like, you know, pretty run-of-the-mill stuff, and I threw that here. What I have not done is I also have this file called, um, an FAQ-Which I had not added to this folder yet, but that I'm going to in front of you. So my goal here is to demo that even though this file was just added to this, it will still work, and it'll continue to do its thing, right? So I just, as you saw, drag this over here. I'll make sure to rename this in a way that it can read that hopefully. Will it enter? Will it add? Great. So now our initial document setup is ready. So as we talked about, we will now be querying our bot, and you can do this on any channel, or you can do it directly. It kind of really doesn't even matter. But maybe we'll just use a channel that I'd added it to, right? Or maybe we'll go to bug reports. It really, it can be any channel that OpenClaw is added to or just the directly OpenClaw channel itself. So we'll go here. We'll say, "Look through the docs in your workspace folder and tell me the..." I actually even had a question ready for this, so let me pull up that question. It should be right here. Yep. So I wanted to outline the database schema for handling this RSS and AI data. So I'll just simply post that. So we will run this. I expect it to-- Oh, and remember, guys, you always have to tag it. Otherwise, sometimes it can be iffy. So just to be sure, I'm going to resend with having tagged it, although I do think it should work, but we'll just resend it just in case. And now we wait for it to respond.

    6. AGAakash Gupta43:10

       So what is it-- What do we expect to be doing in the background?

    7. NPNaman Pandey43:14

       Yeah. So great questions. Okay, great. So as you can see, it's churning out some stuff. So yeah, Aakash, what it does in the background is that it goes into its root folder. It knows that docs exists there. It can see inside it to find whatever it needs, right? So then it looks through that document and then comes back to the original request. And then, uh, in this case, it answered my question based on the doc that it found. Let's, let's-

    8. AGAakash Gupta43:40

       Right

    9. NPNaman Pandey43:40

       ... test and make sure that it found the FAQ that I just pasted in as well. We'll say, "Are there any FAQs that we missed in the FAQ doc?" My plan is whatever it comes up with, we'll actually have it write to the doc itself just using Slack without me lifting a single finger. So let's see how it fares with that.

    10. AGAakash Gupta44:05

        Hmm. So it has access to docs and can edit them if they're in Markdown.

    11. NPNaman Pandey44:09

        Exactly. That's exactly right. Realistically and technically, it has access to every file on your computer. Right now, it doesn't. Look, interesting. So it did find the faq.txt without me telling it explicitly that I just pasted that in, right? So it found those things. You-- Look, it even offered me I can append these. Let's say sure. Let's do that. So the important thing outside of this little toy demo here for your, you know, your listeners, is that these can all be lifetime. Remember, I'm right now sitting here doing these things, but it can very well be that maybe one of your leaders or one of your customers reach out saying, "Hey, this is a question I have. Why is this not on your FAQ?" Maybe you can even get an email on this, and that is actually subject for the next use case. But you can have it set up to be automatically making, doing these operations in real time. So apparently, it claims to have added it. So we'll go back to our thing here. It even added the SQL schema that I'd asked it to, just to make sure that I can access it later outside of Slack as well. So if I open this file now, and I'm now realizing it would have helped to show a before, after, [chuckles] but hopefully, we can see any differences in the way they're written. Yeah, there you go. It even labeled it with additional features and infrastructure, and it added those four questions that it thought that it should have. So again, going outside of this toy example, you can have dozens of files in this folder, right? This is not limited to just one PRD and one FAQ. It has the ability and the context to sift through all of them and really customize its responses to what it's seeing in that folder. That concludes demo of use case one. Before I launch into the second one, do you have any questions, any clarifications that I can do?

    12. AGAakash Gupta45:55

        So I see this as an instant knowledge base. This is about if you can construct a really strong repo of, "Hey, these are the features we're gonna build. This is the strategy. These are ev- the transcripts of every customer conversation. These are all the relevant data charts," you can scale your impact. And the flow can be that your engineers and designers, they go out there, and they come up with a feature idea. They chat with your knowledge bot to come up with the proposal. Then the human actually comes in and actually reviews it, and the PM can actually really scale their impact this way. So with the way ratios are going, PMs supporting more engineers than ever, I see this as a really important tool to multiply your leverage.

    13. NPNaman Pandey46:33

        Not only is that a great call-out, but the aspect of that, that you mentioned, where you can build that repository, I would put a little asterisk in that and say that you actually don't even have to build that repository. OpenClaw builds that for you, which it's funny you say that because that is exactly what use case three is. Um, but for now, for use case two, what I had was the pain point, quote-unquote, being which I will paste here again. And you know, I know enough PMs to constantly hear about this being a huge pain in their behinds and honestly, even like software people, right? So automated daily stand-up summaries, what it does, every morning at 9:00 AM, OpenClaw scans your team Slack channels. These can be as many channels as you want. It summarizes what happened overnight, identifies blockers, and posts a concise stand-up brief to product stand-up. Why should PMs care about this? Because it replaces the 15-minute context-gathering ritual. I actually saw a post on LinkedIn just earlier today where somebody was like, "The bane of my existence is on every Monday, I have to sift through all of what happened last week and compile that to share on my weekly stand-up." So you walk into your first meeting already knowing what shipped, what's blocked, and what customers complained about. Before I launch in, any thoughts

13. 47:47 – 54:46

    Automated stand-up summaries (use case 2)

    1. NPNaman Pandey47:47

       on, you know, this particular use case? Would you agree that this is kind of a pain point for PMs? It would be helpful if you can add your, um, take to this, so to speak.

    2. AGAakash Gupta47:57

       Yeah, sounds great. Looks-- Especially if your teams are reporting into Slack channels. And if they're not reporting into Slack channels, you can configure other data sources, which I think we'll cover in a bit.

    3. NPNaman Pandey48:05

       Exactly.

    4. AGAakash Gupta48:05

       But if they are in Slack, this is perfect.

    5. NPNaman Pandey48:08

       Perfect. Yep. So as you can see, in this case, I have a engineering channel set up and a design. So what I was hoping to do, and you can, you know, you can tell me if this is a bad idea, was for you to even have in addition to me, just write stuff here that, you know, this broke, this doesn't work, I'm going on leave, [chuckles] whatever you will. So our goal is to really, like, confuse it i-if, if you, if you know what I mean, and then we'll run the cron job. And by the way, because I was testing it, you can see the morning stand-up summary that dropped sharp at eight AM my time 'cause I'm CST. So, you know, it, it did its job. But obviously it's not limited to just eight AM. We can even run it whenever we want, right? So would-- What do you think if I invited you here, Aakash, would you [chuckles] help me spam this? Um, here, I'll grab your email, and we can, like, blur out your email, but I thought this would be a fun activity. We will directly add you to NP Space. Great. And while that happens, um, I guess I'll add you to design as well. Let me know if you were able to access. But in the meantime, I'm just going to start to... You know, we'll just pretend this is an actual engineering Slack channel. We'll say-- Or guess, I guess I'll say I'm going on leave, um, starting tomorrow, so my sprint will be extended by two days, and we'll j- And I'll just hyphenate and say my name just so it knows. And then for design, I'll say, um, the back button is broken, and I don't know how to fix it without talking to my design lead. Okay, we'll just send that. Right. So we're obviously just trying to, you know-- Yeah, exactly. Just a bunch of updates come in all day, and as I, as I showed you on the product stand-up, this will be generated tomorrow morning when we are no longer on the call. But we'll just have it generate now, right? So for that, I'll just say OpenClaw. Um, and yeah, so one important thing here, you can obviously ask it to generate the morning stand-up summary. The other thing you can do to set this up is, and I guess I can now show you guys how you can do this, but the way to do this is set up a cron job, and this is what that looks like. So you just go on your chat, and you just mention that the following command, right? So what I like to do is go on whatever, Claude or what have you, and just type out in simple English that at eight AM I need to know blah, blah, blah. Basically the problem statement that I just shared. Now, you can even paste that here directly. What it will do is it'll automatically convert those English instructions to system-readable code, quote unquote, if you will. Which what does that look like? The following. Set up a cron job. That's the command for that. This oh nine stuff refers to the time zone. Session isolated message it... Yeah. So it reads the last twelve hours in those two channels. You can add a, a certain channel here if, if these aren't enough, and it, it'll summarize what shipped, what active blockers or escalations, key customer complaints. Right? And then post a summary too, uh, with clear headers. What you can also do is have an associated catchphrase for this to trigger. Just, you know, like a Iron Man style. You-- If you use Whisper Flow, you just press your little, uh, activate button, and you just say, "Stand up," and you just hit enter, and it does everything else for you, which is, you know, it's really cool. So I'll just be trying to demo that here. So I'll just say, "Stand up," and we'll make sure there's no space, and we'll see if it's able to run the cron job and give us all of these new updates that it now has. So fingers crossed.

    6. AGAakash Gupta51:43

       Product managers are getting one closer-- one step closer to Iron Man.

    7. NPNaman Pandey51:47

       [chuckles] Okay. So tell me if any of this feels accurate, 'cause you inputted some of these things. Five features have been shipped.

    8. AGAakash Gupta51:54

       Yeah.

    9. NPNaman Pandey51:54

       Three new templates have been shipped. There's a few blockers. UI development is on leave. Unable to fix broken back, great. I'm also going on leave, extending by two days. So I mean, again, I don't think anyone is surprised here. Oh, if, if this surprises anyone, it's because of testing that I was doing earlier. But yeah, you no longer have to comb through a bunch of Slack channels.

    10. AGAakash Gupta52:16

        Yeah.

    11. NPNaman Pandey52:16

        And as we will see in our next use case, even stuff outside of Slack, it's, it just constantly lives, breathes. It's pulling all of these various avenues and then consolidating them for you to just, you know, be in your own bed and not even have to join the call or whatnot and, you know, just, just be relaxing. Not that that's what PMs do, but... [chuckles]

    12. AGAakash Gupta52:37

        Yeah. We all live in Slack hell with way too many messages and notifications. It, it's impossible to keep up. So now you have somebody who's there that you can converse with basically and get caught up on Slack.

    13. NPNaman Pandey52:49

        Actually, on your point around the con-conversing, we can even ask it, "What's here that needs my immediate attention?" Right? 'Cause if you're trying to not get overwhelmed by all of this stuff going on, it is smart enough to figure out what you need to immediately act on. In this case, I'm hoping it'll bring up this sprint stuff 'cause that's a big blocker in my view. Um, yeah, exactly. Right. So you're going on leave. Also, notice how it knows that Naman is me. So it, it didn't say Naman here. It said you. Ensure your tasks are handed off or documented. Anyway, we, we don't need to, like, check its work. But what's the takeaway here is that unlike other LLMs, right, which constantly lose memory, run out of context size, run out of token windows, this is built to constantly make sure that it regurgitates, if you will, the right context size to not lose sight of what's important here. That, that's really the headline here and not, you know, the nitty-gritties of what, what we have here. So before I move into use case three, just wanted to pause. Any thoughts, questions, anything else that's coming to mind with this?

    14. AGAakash Gupta53:52

        Let's move on to three.

    15. NPNaman Pandey53:54

        Perfect. So for use case three-We have the following. What this is, is really competitive intelligence pipeline on autopilot. It basically monitors all of your competitor avenues. Right now, it just says websites, Capterra reviews, Product Hunt launches, Hacker News mentions. You can have this list be 15, 20 items long, right? There is literally no stopping this. What it does then is that it synthesizes a competitive intelligence brief using SWOT analysis, and it posts it to a private competitive intel Slack channel. Why should PMs care? It's because obviously doing all of this manually is really critical. In addition to that, say there were certain changes six hours ago on a competitor website, you happen to be sleeping, right? When you woke up the next morning, they had overwritten that for change C instead of change B. In a normal world, that

14. 54:46 – 1:13:26

    Competitive intelligence on autopilot (use case 3)

    1. NPNaman Pandey54:46

       is gone forever. You have no way of ever finding out that that happened. Not with this, right? OpenClaw never sleeps. It's just out there, you know, doing its thing, monitoring it every hour, every half hour, if you can set it to do that. And then, yeah, your-- that competitive brief lands in your Slack without anyone lifting a finger. So the way I want to show this here is by actually building this out. So I did not, in fact, end up building everything because I did want to show what the building looks like. So really what we're doing is what I just said, right? We go on here, remove that, and I have the prompt pre- pre-ready just because I didn't want to waste time. But, um, in your competitive intel Slack channel, just mention the bot with something like, "I need blah, blah, blah. Search the web." Um, and I customized this to our use case, which was latest news, product updates, and user sentiment about pod page, which is, and Transistor FM, which are my competitors for the toy example, toy/serious example that I was trying to build. Organize your findings, highlight anything that represents a threat or opportunity for a mid-market podcast website builder site. Post your analysis. We'll say there. So what is going to happen is, firstly, it'll say there is no competitive intel Slack channel, and it will make us do all of that rigmarole. So let's actually build this out so you guys can see what exactly that looks like.

    2. AGAakash Gupta56:07

       Mm.

    3. NPNaman Pandey56:07

       I mean, while it does that, because, you know, we like to be proactive, we can simply make that channel, right? Um, make sure we have the right name. So we'll simply just copy that, make that channel. We'll hit Next. You can just have it be public. Great. We'll just add everyone. Um, and then I guess we'll also, uh, add the app because I don't think that is added. Yep. Okay, great.

    4. AGAakash Gupta56:29

       Mm.

    5. NPNaman Pandey56:29

       So we did that. It's still thinking. We'll give it some time to think some more.

    6. AGAakash Gupta56:34

       So when you want to look at the context and-- that it has, is there a way to look at that? Because I know with Claude code, managing the context is a really important part.

    7. NPNaman Pandey56:43

       Yeah. So a big part of the context lives on these files itself. So for instance, if I showed you agents, right? Like what are the agents that it's working with? So I'll flash it up here. So this is-- You can think of it as its consciousness, if you will, right? So this has everything that it needs to run. So for safety, this is just all default. So don't ex-exfiltrate private data, [chuckles] don't run destructive commands, so on and so forth. Ask first, right? So for anything like sending emails, tweets, publics. Again, this is just boilerplate. You can change all of this. So in-- So again, the way to work this is you can go on your Claude or Gemini, whatever, and then have that build out the entire markdown document. Suppose you want five different agents that, um, fella, in our case, leads. So you can have that organizational map that bot A will just say Naman is responsible for marketing, bot B is responsible for so on and so forth. And then if you just paste that markdown file into your workspace here, it will know exactly what you want it to do at any given point, right? So that is the way to manage not just what you want it to do, but also what you want it to work with in terms of the documents and such.

    8. AGAakash Gupta57:56

       And in terms of those documents, agents, so what are the best practices? What goes in what document?

    9. NPNaman Pandey58:01

       Yeah. So the good news is that this will all be set up for you by default. You will not have to worry too much about changing a lot of things. Again, if you were to want to want to change this stuff, there is a line here that, um, it almost makes me, like, feel philosophical, [chuckles] but I-I forget exactly where it is. Like, for instance, look, remember what I said about the stand-up stuff? Look, it-- And this is not something I've written. This, it wrote itself. So if Naman types the exact word stand-up in any context, he must use the message to instead of replying directly, call message with action read. This is all-- this is like my Slack channel stuff. So it configures itself to act according to, you know, how you want it to.

    10. AGAakash Gupta58:44

        Nice. So you could just chat with it, and it'll update those files over time.

    11. NPNaman Pandey58:47

        Yeah, that's exactly right. Perfect. So it did its business. It-it actually started doing the analysis first, which is not something I was expecting. So this error around the missing Brave key is just-- So Brave is really just an API that any agents you-- even if you use Claude Cowork or now today ChatGPT came up with their entire, um, task-based thing. But it really makes it simple for agents to browse the internet. So we can set up Brave API key. That's straightforward enough. But for now, it says, um, yeah, it didn't find the channel, which is something that we had forecasted. So what we will do now is as-- firstly, we, we have this channel. So something you need to do to point it to your channel, um, is go on here, and if you click Channel details, over here you'll find the channel ID. So you can just simply copy that, go here and say, um, the channel ID is that. So and then as we just talked about, it will configure itself to be able to talk to that channel and such. Now, while it does that, we can go ahead and do the Brave API key setup. This is important because pretty much any type of web browsing activities that you're going to be doing will involve this. I believe it starts at $5. Yeah, I think that was where we had started. So-I think I already should. Okay, I do have an account. I don't think I have an API key. So I guess I'll just, like, configure this up real quick. So we have our API key. We'll just add that. We'll just say OpenClaw. We'll hit Add. Search is there is fine. That's our guy. We'll copy that, go back here. Um, great. So I successfully posted that message. We can verify. So if we go back, great. So it was able to do that. All it needs is an API key now, right? Everything else is ready to go. So we just come here, and we'll say Brave API key is-- And this can stay 'cause I'm going to delete this immediately after. But the important takeaway here is normally for anything else, what you would have to do is go into your VS Code file, go into your environment variable, paste the API key, then attach that variable to anywhere else in the code where you would need to access it. Here, you just dump it on your, this nice little interface, and it puts itself to use, and it pastes the API key everywhere it needs to without you having to lift a single finger. So-

    12. AGAakash Gupta1:01:14

        Sweet.

    13. NPNaman Pandey1:01:14

        Yeah, and the reason it disconnected is because it, it needs to refresh, and I think that's the API response, if I'm not mistaken. Apparently, it did it. So we'll say, we'll just say run the competitive analysis right now, and we'll hit Send, and then we should see a message here. Actually, what I should have done was write that, um, you know, prompt here, but it's really the same. You can use them interchangeably. At the end of the day, they're all going to the same source, if you will, even though the avenues are different. Great. So as we know, it is smart enough that it knows that it, it's supposed to go there. So okay, [chuckles] let's look at the analysis. So I actually use Podpage, and I've used Transistor as well. So let me quickly just eyeball this and see if it's looking accurate. Aggressively targeting podcasts. Okay, that's, that's accurate. Allowing multiple shows. Yep, that's true. Many built-in sites are rudimentary. Yep. I mean, I can show my site that's on Podpage, and it, it's just the worst. [laughs] Sorry, no shade, Podpage. Um, migration tooling. Since Podpage relies on a white glove, seamless could, could be a massive conversion driver. This is gold. So I'll show you what I have been working with. This is, like, my very rudimentary pre-MVP stage. Something I had not thought about so far was exactly how to make it really easy for other folks to import their pages, um, into, into this. So this is just what I have right now, and my plan was to go away from the whole WordPress type route. So it did accurately identify this super cool, at least to me, angle or wedge here, that it could be a massive driver to seamlessly allow people to go from WordPress to, you know, platform migration in terms of opportunities. Video integration. This is one of the biggest reasons why I even had this idea, and I, I have that, or at least I'm working to have that be built in. 'Cause most podcasts now are video-enabled, right? So I wanted this to be video first. So I have not seen this before. This is the first time I'm seeing this in work, and, like, at least I am definitely impressed by what it's saying. Let's look at risks. They are pushing native network websites, reducing the need. That's a fair call-out. If hosting platforms improve their native site builders, standalone will lose market share. Perfect. So I mean, in a sense, that does conclude our, uh, this demo, but what's important here-

    14. AGAakash Gupta1:03:47

        Well-

    15. NPNaman Pandey1:03:48

        Yeah.

    16. AGAakash Gupta1:03:48

        One question I have is-

    17. NPNaman Pandey1:03:49

        Mm-hmm

    18. AGAakash Gupta1:03:49

        ... any AI tool you ever build, like I've had AI agent CEOs like Lindy and Relay, you and I did the agent browsers.

    19. NPNaman Pandey1:03:59

        Mm-hmm.

    20. AGAakash Gupta1:04:00

        People talk about Claude Code and Cowork. They always use this competitive analysis use case. So how does OpenClaw differentiate from all those other AI tools?

    21. NPNaman Pandey1:04:09

        I would say the biggest differentiator, and at least I know for a fact that this is not, uh, possible to do using any of your, um, AI-first browsers. Definitely not when we are recording this on fifth March twenty twenty-six. But the biggest takeaway here or, like, the wedge, you know, if you will, here is that first, we don't have VPS set up right now, but you can have it be pulling Google reviews. You can have it be pulling a bunch of different avenues that you-- It would just be really hard for any other or even Claude Cowork, for that example, for that matter, to look at actively every thirty minutes and then persist with its memory so that six months down the line, if you asked it to, say, generate a trend line of all the changes that a certain competitor made that helped or hurt their business when it came to reviews on Reddit or on Twitter, it's like basically having a watchdog across any website that you can think of that literally never sleeps and, and, and along with the ability of it, of it being really smart and you being able to converse with it in real time in a way that it never forgets a single update it saw like seven thirty five point two hours ago, uh, when at like one AM when you were randomly sleeping. I will say with Cowork, the difference is a little bit lesser. For me, the differentiator then becomes cost, 'cause remember, you can run Qwen three point five on this. Qwen three point five is one-tenth the price of Anthropic APIs. I am, like, mortally afraid of ever using Anthropic APIs because one prompt and it burns through $20 like it's nothing. I don't know if you've had a similar experience. Um, but when you couple similar-ish abilities with the ability to be really agnostic with your AI LLM usage, I do think that's the democratization of it all that really makes it click together. And for me, I actually don't even do a bunch of things just because of how expensive it is with Cowork in particular. I'm actually afraid to use Cowork, [chuckles] uh, but not with this, right? 'Cause Qwen can get you, I, I would say per my experience, at least sixty, seventy percent of the way there for one-tenth the cost. And for most people, I'm willing to wager that's a fair enough standoff or, you know, trade-off, I should say.

    22. AGAakash Gupta1:06:35

        Okay.And you mentioned about the VPS part of that. Can you say more on that?

    23. NPNaman Pandey1:06:41

        Yeah. So you can have this be doing operations within your file system when your computer isn't even on or, you know, when it's, like, shut down. You've, you've shut down your computer, and it still will be able to do things without actually you being around or being in the same country.

    24. AGAakash Gupta1:06:59

        Nice.

    25. NPNaman Pandey1:06:59

        Again, that is risky. I have not-- I'm just personally worried to play around with it. And not to go off on a tangent, but I was trying a bunch of, um, WhatsApp initiatives yesterday with OpenClaw. It decided to message everybody that I was talking to for two hours after I'd done my demo, just with trying to pair with their phone numbers. So I would s- And nothing I ever did would have prompted it to do that, according to my brain, at least. So yeah, yeah, like, I was talking to my mom, and it auto-sent her. And firstly, I asked it that, "Do you have access to my other chats?" And it was, "No, I can only chat with," you know, like how you have your own chat on WhatsApp. I was like, "Great." Except clearly that's not true, because it sent her the, like, secret code to attach to her number as well. And my mom was like, "What is this? What is going on?" And I was like, "Don't worry about it, Mom." [chuckles]

    26. AGAakash Gupta1:07:55

        Okay, so beware connecting your OpenClaw to your work Slack and your WhatsApp. Uh, it can go off and do autonomous crazy things, so you need to do it in a s-safe way.

    27. NPNaman Pandey1:08:07

        I will say to, to, just to caveat, with Slack and with any-- Here, I'll put it this way. Any default things that you're doing, quote-unquote, are probably fine. I've not seen it misbehave on my computer so far, not that I can see at least. Nothing that I've noticed. So for most use cases, you are probably fine. But just remember that it comes built in with most security guardrails in place, and just make sure you're not trying to move them around unless you have a really strong reason or if you really know what you're doing. For most people, for most use cases, if you stick to the default guardrails it comes with, I think, and please don't quote me on this, but you will probably be fine. With Slack and such, you don't have a lot to worry about, right? Because the moment you, you start to worry, you just disconnect it, right? It's literally a one-click operation to delete this app from your Slack, and you're good, right? It's as simple as that. But I will say-

    28. AGAakash Gupta1:09:04

        Famous last words. Famous last words. I think it was the Meta super intelligence safety lady on Twitter-

    29. NPNaman Pandey1:09:12

        That's true

    30. AGAakash Gupta1:09:12

        ... she posted about how she had to, like, run to, like, physically unplug her OpenClaw from, like, deleting her emails or contacts or something like that, so [chuckles] -
15. 1:13:26 – 1:24:30

    Voice of customer reports (use case 4)

    1. NPNaman Pandey1:13:26

       So now you're wondering, could something have broken that we just don't know? Well, the easiest way is to just poll what your customers are saying, right? Why should PMs care? Because this is the PM dream, a living queryable customer feedback system that doesn't require an analytics platform. The bot maintains persistent memory, so it tracks trends over weeks, which is something that I was mentioning before. And it can also tell you authentication complaints increased forty percent compared to last week. And again, this is extendable to quarter, year, you know, even probably five-year time frames. So the persistent memory is the key aspect to this. And as we were talking about it, we have a memory file, a markdown file right here, which it makes sense to flash, which right now just has this, but, um... Sorry. Wait, I think for the persistent memory-- Okay, yeah, it is memory.md. Great. So it's always helpful to check. So when we are done building out this use case, you will see this file, this memory markdown file automatically updated with our particular use case. So similar to last time, I already-- I think I have the prompt ready. Oh, I do not have it ready. Great. We can just quickly write it here. Um, and then the way I wanted to demo this was, Aakash, you can just email me pretending to be a customer on my email, which I can attach, which I can just send to you. So just like, you know, feel free to send an email re- saying that, just, you know, any issue, just be like, "I hate your software," or like, "I hate your," uh... Yeah, just be like, "Uh, RSS feed, um, onboarding did not work for me. This product sucks," or whatever, right? 'Cause most products have their, um, support email that usually some sad person has to monitor. Um, it's not, not, not to the fullest extent, but it can happen where it's getting filled up real quickly by just complaints from people all over the world. So we will just be demoing the email aspect here, right? But we can extend this for it to poll Google reviews, um, Product Hunt, re- literally, um, Twitter. Whatever you need to, to be polled, you will just make a list, which I will do now as you email that to me, and we will set that up here. So again, I'm trying to show you, like, the art of the possible here. So I'm literally just going to, like, paste the entire, um, prompt itself directly in here. I want for it to struggle. I don't know if that it will. We just have to be careful about what channels we're mentioning. So I know we have competitive intel, so we'll just say that. Otherwise, we'll just have to add it again, which is just, you know, kind of annoying and we already know how to do that, so we'll just remove that. Maybe what we can do is have conflicting feedback. So maybe you email me saying RSS is bad, we can say RSS is good. So, you know, so just to see what it actually does. The rest of the prompt just kind of tells it what it needs to do. So, um, it creates a heartbeat, which is wh- you know, every six hours, scan these things. We will update those channels one more time just to make sure that they're not lost for new messages. And then I'll add, "And my email inbox that you have access to." Quick note on that, if you don't want to use Brave API and all of that, that's totally fine. Most Mac users have the Apple Mail app inbuilt on their laptops. If you have that, that is all it needs to access your email. Just by extension, we won't really be covering this, but if you can think of any use cases that involve email, it can do it, right? 'Cause think of the two things that it has. One, access to your email. Two, ability to do things. So say if somebody sends you an email where they want to schedule a, a podcast recording with you, you just go on your Slack, you just be like, "Check my emails for any new requests. If there are any, schedule them using my Google Calendar," which again, it does have access to that as well. So it just does everything for you while you're just, yeah, like watching TV or something. Again, we will not be covering that, but just wanted to call that out. Any and all use cases involving emails are just solved now across the board. So yeah. And then the rest of it is categorized each by area and severity, update the running tally, all of the, this, you know, good stuff. And we have it set up every thirty minutes by default, keeps the data fresh, and the weekly cron produces this synthesis of all of these thirty, um, uh, thirty-minute increments. So really this is two ongoing cron jobs bundled into one, right? So the f- mini version is every thirty minutes, and then every week it gives you the summary of those thirty minutes, 'cause obviously if you, y- you know, you don't want to be reading it every thirty minutes unless you really have to. As usual, we just hit send, sit back, have a sip of your coffee, and go on a walk [chuckles] while it just builds itself, which I'm hoping by now you're beginning to really appreciate that pattern, where that is the biggest unlock here, where you don't do any of the setup. It does the setup. It sets itself up.

    2. SPSpeaker1:18:26

       One of the things that I like to do with Claude and Claude Code is have like five, six chats going at once so that I don't have to wait ever. Can you do that?

    3. NPNaman Pandey1:18:34

       You can. What you need to be wary of is all of these activities, right? So when it thinks that, okay, I need to access browser or Finder in this case, then I need to click and open this file, then I need to dump all of this text into this file, which they're not going to come up from thin air. I need to write them. So-It depends on how context-heavy your ask is. If you're doing a lot of pollings type-- polling-type stuff, 100% you can, you know, really just go all in on it. Where I found it struggle was just yesterday. I was trying, I just gave it the command on WhatsApp that, "Hey, find five people on LinkedIn like me that you think I will be friends with, and auto reach out to them for me on my phone," is what I told it. And then I saw it open the browser. I saw it do its searching. I saw it find a profile, which, you know, it's probably a conversation for another time who that was. I saw it click the message button, like when you know you're sending a new message, and that's when it gave me the API rate limit. But that makes sense because if you think about it, just to get to that stage, it's such a compute-heavy operation. Like, my human brain would need 10 minutes to even figure out a plan on how to solve this weird problem. How-- I don't know who's like me. [chuckles] You know? It's, it's like I don't know how to solve that. So that's the only thing to keep in mind with this, that if it's not super heavy, it's, it-- you can definitely concurrently run stuff. As you can see, apparently it's just done. Um, it just did everything we asked it to. We should see a custom skill. Let's actually inspect that. So we go here. It says it should be inside voc taxonomy. V-O-C. Maybe we refresh this, or maybe it made it some memory. I see a memory here that does not look right. So here's what we'll do, right? Since we can't find it, I will literally just copy this and I'll say open. I don't know why that's copied. I'll just, like, have it open it for me, right? 'Cause why am, am I finding [chuckles] stuff in my finder on my computer? And hopefully it, it does that. We'll see here. Okay. There, there you go. Um, [chuckles] so now if I right-click, do that. Great. This is pretty much what we had asked it to do.

    4. AGAakash Gupta1:20:57

       So it's really excelling in computer use compared to your average AI.

    5. NPNaman Pandey1:21:01

       Exactly. Yeah. Like, there is a version of this where you're, like, actually completely hands-free. I-- That made me think of another really cool use case. It's like a mini use case, but I, I'll show that here. We'll take a pause from this real quick, right? So we were talking about all of this, right? We even had this, uh, database schema. So you can actually have it draw the entire schema diagram on your browser. So the way to do that is we'll say, "Earlier you talked about a schema diagram. I want you to draw it on my browser so that I can edit it. You can use Mermaid."

    6. AGAakash Gupta1:21:35

       And we probably need to hit @OpenClaw, right?

    7. NPNaman Pandey1:21:37

       Exactly.

    8. AGAakash Gupta1:21:37

       Yep.

    9. NPNaman Pandey1:21:38

       So when I-- So I accidentally stumbled across this. I actually didn't even know it could do that. What should happen, and I really hope it, it happens here as well, 'cause I don't know if it will for sure, um, but it'll open, like, that entire drawing like a Miro almost. And I... You're probably familiar with Mermaid, but maybe some of your audience-- Okay.

    10. AGAakash Gupta1:22:00

        Wow.

    11. NPNaman Pandey1:22:01

        Like, it just drew all that. And obviously to edit, I just need to change that. I can then export it. I can probably have it export it for all I know, right? So again, constant theme here, go hands-free. If there's a thing that you can imagine yourself doing on your browser and your file system or any combination of the two, don't do that anymore. Just speak into your, [chuckles] uh, you know, OpenClaw bot and it'll do it for you.

    12. AGAakash Gupta1:22:30

        Nice.

    13. NPNaman Pandey1:22:30

        So coming back to our little use case here, did you send me that email yet for the feedback, Akash?

    14. AGAakash Gupta1:22:38

        Yes.

    15. NPNaman Pandey1:22:38

        Perfect.

    16. AGAakash Gupta1:22:39

        I sent you both, the conflicting emails.

    17. NPNaman Pandey1:22:41

        Oh, okay. Yeah, exactly. So if we combine feedback from the Slack channel as well, as, as well as email, you can see that it said competitive intel failed, but that's just because we didn't give it the ID, which we know how to do. What's more important is that it identified this correctly on the customer support channel, which I believe, what had we sent? Okay, that you needed more customization. Perfect. So it included that. Um, in your email I believe, I-I'm not sure what you sent me, but it said your RSS builder has people somehow both complaining and raving about it. So [chuckles]

    18. AGAakash Gupta1:23:14

        Yes. That's right. I tried to confuse it, so I sent it from two different emails.

    19. NPNaman Pandey1:23:18

        Wow. Okay.

    20. AGAakash Gupta1:23:18

        Your RSS is great, your RSS sucks. [laughs]

    21. NPNaman Pandey1:23:23

        So interesting. I mean, it, uh, at least it's, it, it's accurate. But so yeah-

    22. AGAakash Gupta1:23:28

        Yeah

    23. NPNaman Pandey1:23:28

        ... even though it says status failed, we know that it's not, you know, anything to worry about, 'cause all we have to do is give it the right channel ID. So that concludes use case four. Um, we can jump to the final one here if that's okay, Akash.

    24. AGAakash Gupta1:23:42

        Let's do it.

    25. NPNaman Pandey1:23:43

        Perfect. So for our final use case, something we haven't done so far, which is really critical with anything useful, quote-unquote, that you're doing, is decision-making, right? Like, you don't just want it to be a monitor, which we've covered, or do something if something happens, which we've covered. What we've not done is for it to make decisions for itself based on what it's seeing, and then doing different things based on what it sees. So what we're going to do is that we will, we will supply it with a, a user CSV, if you will, right? So any product has paid as well as free users. So what we will do is we'll try to mimic how it adjusts to bugs reported by an enterprise user versus a free-tier user,

16. 1:24:30 – 1:37:40

    Smart bug routing by customer tier (use case 5)

    1. NPNaman Pandey1:24:30

       and we'll have it do different things based on checking the data and the identity of the user that submitted the bug. So what that will look like is, just in the interest of time, I, like, saved the prompt already, and, I mean, this is what that looks like, right? Great. So this is our little use case that I just scribbled in, like, my Notes app. But what we'll have it do is that we'll ask it to build an automation. When somebody posts a bug report in a Slack channel, I want you to read the message and extract the reporter's name. Check if their name appears in the customer list or CSV file in your workspace. So this can be just a CSV, which literally can be as long as you want. The way this is configured is Sarah Chen is an enterprise user. Um, James here is pro. Lisa is just a free personal user, right? So most production apps or products have a similar segregation. Somewhere in their, you know, uh, repository, they have a way to figure out which user is what, and because not all users are the same, right? So if they're not a paying customer, it'll post it to a, a different channel. Um, if it is a paying customer, escalate to engineering urgent with the full bug details and flag it as high priority. And then finally, it even replies in a thread on the original message, acknowledging the bug and giving an estimated response time.And finally, if all of this was not enough, it appends the bug to a tracking log in bugs.csv with the date reported, severity, and which channel it was routed to. So a few quick things here before we jump to building this. All of these CSV files can be your online, if, if your company uses Jira, it can be that. It can be Asana. It can be pretty much whatever API that you're working with in your organization. In this case, because this is not an actual production thing, I just stuck with a CSV, but my goal here is to demo that it doesn't end at a CSV, right? It can append a CSV, obviously, duh, but in addition, it can also make those live changes on wherever your company's work, you know, workstation lives and breathes. So I guess before we go about actually building this, Aakash, is this looking like a decent k- uh, kind of use case that might, maybe surprises you? I'm just kind of curious to get your thoughts on what you make of this.

    2. AGAakash Gupta1:26:47

       Love it. Very cool.

    3. NPNaman Pandey1:26:48

       So as usual, right, we just need to imagine, so the prompt that I just showed will simply go, um, to our gateway, the UI gateway, and we will simply paste it over there. So let me quickly bring that up. So interestingly, it gave me the API rate limit [chuckles] for that. Hmm. Usually when this has happened in the past, you have, like, there's two major options. You can either change the model that you're talking to, or sometimes if you just wait, that does it as well. I think we'll maybe just change... Okay, so apparently that was just a lie, 'cause it looks like it's doing that now. Um, the problem here is we, we'll need to create these channels, so [chuckles] let me quickly just edit this for us to just use the channels that we have. So give me one second while I do that.

    4. AGAakash Gupta1:27:44

       Okay.

    5. NPNaman Pandey1:27:45

       Cool. Looks like it's fully set up. Great. So we can now pretty easily test it, right? So to test it, what we can do is go on here. Interesting. I c- I didn't know you tried this. [laughs]

    6. AGAakash Gupta1:27:57

       [laughs]

    7. NPNaman Pandey1:27:58

       So, you know, it's so funny, I think it's because of what we talked about in terms of remember it's sol.md said to not share private details on the internet.

    8. AGAakash Gupta1:28:08

       Yeah.

    9. NPNaman Pandey1:28:09

       We can obviously double down and ask it why it said this, but this is really cool actually. I, I didn't actually think it would be smart enough to not do this. But for now, I will just say that your landing page won't let me log in, and I think here you-- we have to tag it, Aakash, or it doesn't work. So we'll say your landing page, and that's a good reminder for me as well, and we'll say one of our users who was an enterprise user, so Sarah Chen. So normally this would be her account, but I'll just say a hyphen and say Sarah Chen, just so that it knows that, you know, I'm an enterprise user. Maybe you can post... Actually, where did we ask it to check? We asked it to check engineering urgent. Do we have that? No, we said, and, um, and design. Yeah, Aakash, maybe tr- well, [exhales] damn. I guess that's, we'll just let it be. And for design, we'll just say there a resizing issue, and we can have this be said by Lisa Park, who was a free user, so that she should be routed somewhere else. And we actually don't even have to tag OpenClaw for this, so we'll-

    10. AGAakash Gupta1:29:22

        Nice

    11. NPNaman Pandey1:29:23

        ... do that. Once this is here, we'll say, "Run the, run this automation above with," we'll just say, "Run that above," and I think that, that's usually enough. Okay, rate limit. No worries. Just need it to run this one last time. Um, yeah, it's funny how it's just like a fake rate limit issue. Oh no, that was not fake. We'll just refresh. Come on.

    12. AGAakash Gupta1:29:52

        [laughs]

    13. NPNaman Pandey1:29:54

        We're so close. We are so close. We'll just change the model, I guess. That's straightforward enough. So it did work, but the problem is this, it, it says, [laughs] "It's clear the user is impatient," [laughs] or "their interface is lagging." [laughs] Um, so yeah. Yeah, I really should have just waited, but okay. I'm now hunting for the channel ID. Okay. But I need to grade. I guess we'll just say, "What's going on now?" We beat the rate limit stuff though, so that's at least-

    14. AGAakash Gupta1:30:31

        Yeah.

    15. NPNaman Pandey1:30:32

        That, that, that's on the brighter side. Okay, so it looks like it did a bunch. Okay, so let's just now r- like try re- rerunning this and see what hap- We have to tag it. Um, while it does that, trying to run the... Okay, I'm hitting a snag. Okay, as usual, it's the channel names that throw it. So reading messages. Okay. Bug reports it doesn't have. Oh, interesting. So it did work. Look. So I just mimicked Sarah Chen, who as we know is an enterprise user, so she just submitted a bug. Let's, like, assume this is our public-facing bug reporting channel, or maybe it's she's a tester or whatever. Like, what matters here is that you can distill users into various privilege levels really is what the takeaway here. And based on some of their privileges that come from a different file, a CSV file that lives on your file system, or it can live on your workstation like Asana or what- whatever tool your company uses, it automatically flagged it, so high priority issue detected. Reporter is Sarah Chen. She's an enterprise user, Acme Corp. That's her issue. Escalated to engineering immediately. This is a critical blocker for a paying customer. It did it twice because, um, I ran it twice, and it finally caught up to that. I am curious though in terms of where it escalated it. Customer support?Um, so Lisa Park, according to our CSV, is a, she's just a free personal user. Nothing should not be flagged as a high priority issue, but we shall see. Okay, so we got a reply from our bot. Okay, bug report, you know, free tier, everything's we're, we're, we're pretty much just chilling. It was routed to design for review, low priority. So if you go in design, um, in my prompt, I had asked it to generate the bug report in this channel. So as you can see, it even picked this, I don't know who asked it to do that. Um, but it's a UI resizing issue reported by Lisa Park, who's a free tier personal user, which, as you know, we did not tell it that. That came from the CSV. So it looked at the CSV, filtered according to the direction branch that we asked it to, mentioned the issue and the status. So the real way this works in the real world, as I'm sure some of you are already guessing, is with GitHub, right? GitHub is where all your customers are going to report bugs, issues, whatever have you. This exact same scenario can play out with GitHub connected directly into OpenClaw. And the-- So I just described what I wanted in plain English and throughout all of these, you know, um, use cases, as we saw, this is really just you prompting things into existence by the thing that does the thing into existence as well, if, if that makes sense. So that was all I had in terms of my use cases. I hope some of this hopefully was helpful for some of you. I'll pass it back to Aakash to, you know, for his thoughts and any questions.

    16. AGAakash Gupta1:33:36

        So we just went over five use cases. This one was basically a self-building agent, right? Without all the node workflows that you would be creating in Zapier or Lindy, so pretty powerful. What are the other use cases people should be thinking about to stretch their abilities? After they implement these five, what should they go do next?

    17. NPNaman Pandey1:33:53

        I would say building in media or baking in media-based usages, for me, is the next frontier. A lot of my time is spent, you know, drawing designs, making those mouse-type movements, or even just on Canva for sometimes, right? You, you, you need it sometimes to just make process maps and all of that. For me, the next unlock here is anytime you're trying to do a map or a map-based visual of any sort, for me, that, that I-- maybe I'm just a dummy, but it takes me hours to first visualize it and then actually draw a bunch of boxes and such to bring together the map for a particularly more, you know, complex process. So I would suggest think about more media-based use cases, be it pictures, you know, videos, and even that far. From this point, you can have it configured to generate automatic, you know, VO videos or AI videos based off something, right, you could, which you can configure. So I would really encourage your listeners to start thinking about it, the next stage here in terms of more multimedia-based operations.

    18. AGAakash Gupta1:34:58

        So you've spent hours using OpenClaw. Most people don't have that time. What are the biggest mistakes, configuration challenges, things you had to debug that people can learn from?

    19. NPNaman Pandey1:35:08

        I would say by far the biggest one was just worrying about the security piece of it all, which, you know, we, we did cover. The best thing here, somehow the, you know, the best and the worst thing at the same time, is that while it is really easy to go berserk, break a bunch of things that you were not trying to break, really what this boils down to, be it setup of a new tool, right, or be it like a security thing, whatever it is that you're trying to do, when in doubt, just ask your bot, right? Remember, it is the first of many that you will make. What I'm working on next is something that I described. I want to have an entire family of agents that are all controlled by Fela, who is like my main major agent here. So I'm going to have Fela report, um, just things that I need to know of the various, like, 16 different bots that are doing underneath it, kind of like a CEO-type model, right? I don't need to know all the details of everything that's going on. So that is my biggest thing. That is what I'm really excited about and what I would encourage people to do. Again, when in doubt, just ask the question, right? Your friendly UI gateway here, that's not it, but that one, is here for you to not just gain any insight into how it works, what it does, why it does it, but also anything you want it to do. You can even lock yourself out from making any dummy mistakes by giving it a system command as soon as you set up that, "Hey, whatever we're going to do, flag me and stop me when I do say one, two, three, four." Right? Whatever you're worried about, no two people are the same. Draw that line in the sand nice and early, and you can rest assured that it will not fail you. At least in my experience, it has not.

    20. AGAakash Gupta1:36:49

        So OpenClaw, Claude Cowork, Claude Code, those are the three hot new AI tools. How would you compare and contrast them?

    21. NPNaman Pandey1:36:56

        Okay. So right off the bat, I think there is an imposter in, in those three, uh, tools that you mentioned, with it obviously just being Claude. So Claude, while really powerful for about, I, I wanna say 98% of the population, um, you know, we already are pretty familiar with what it can do and cannot. Lives on the browser, is completely reactive in that you have to-Um, invoke it for whatever you're trying to do. It lives on the browser, cannot really do anything. I think that's the biggest distinction here. So yeah, I guess again, uh, without beating a dead horse too much, that's just Claude. Really, the, the conversation gets much more interesting when we contrast Cowork with OpenClaw.

17. 1:37:40 – 1:40:21

    OpenClaw vs Claude Cowork vs Claude

    1. NPNaman Pandey1:37:40

       That's when some of the lines start to blur a little bit more. For me, rather than trying to go through the similarities, which there are a lot of, I think it makes more sense for me to go through what's the biggest difference here, which is that in the case of OpenClaw, we have a continual daemon, a D-A-E-M-O-N, that just lives and runs consistently on your machine, unlike Claude Cowork, which is still to a large extent reactive, right? You still have to point it to stuff. You have to set it, give it skills, make sure that it's still trying to do what you want it to do. For me, the biggest differentiator between the two is the idea of consciousness. OpenClaw almost has-- You can even imagine it like a version of you that, that lives in your computer actually, that jumps through your RAM and such, has access to your file systems. It never sleeps. It's able to do things on its own using its own consciousness based on things that it inferred from what you told it, not always directly what you told it, which is again, where we start to get a little philosophical. But there is nothing Cowork does that is actually autonomous, right? It cannot make decisions by itself based on an idea that it has about you. OpenClaw can, which as you can imagine, make it really interesting, but somewhat exciting and dangerous at the same time. Happy to drill into any of what I mentioned. Did that make sense?

    2. AGAakash Gupta1:39:05

       Yep.

    3. NPNaman Pandey1:39:05

       Perfect.

    4. AGAakash Gupta1:39:06

       All right, cool. All right. Um, you had a Google doc you wanted to show people, right?

    5. NPNaman Pandey1:39:11

       Yeah, I can't find it 'cause it was on X, and I tried finding it for a while, which-- So it didn't-- I wasn't able to find it, but I'm sure it'll turn up, at which point can I just send that to you? Would that work? I'll do it-

    6. AGAakash Gupta1:39:22

       Whatever is better for you. Whatever's good for you. Co- Yeah. So let's just end the podcast then. Uh, Naman, this was awesome. If people wanna learn more, where can they find you?

    7. NPNaman Pandey1:39:31

       Um, my YouTube channel, Naman Pandey, has me just constantly breaking stuff, operating on the cutting edge of all of these tools. I already have the next three, [chuckles] uh, tools apparently now that OpenClaw is old. So feel free to check out my videos. Feel free to say hi on LinkedIn. Always love to engage with fellow builders such as you all.

    8. AGAakash Gupta1:39:52

       Amazing. Thanks so much, Naman. We'll have to have you back soon. Bye, everyone. I hope you enjoyed that episode. If you could take a moment to double-check that you have followed on Apple and Spotify Podcasts, subscribed on YouTube, left a rating or review on Apple or Spotify, and commented on YouTube, all these things will help the algorithm distribute the show to more and more people. As we distribute the show to more people, we can grow the show, improve the quality of the content and the production to get you better insights to stay ahead in your career.

18. 1:40:21 – 1:40:42

    Outro

    1. AGAakash Gupta1:40:21

       Finally, do check out my bundle at bundle.aakashg.com to get access to nine AI products for an entire year for free. This includes Dovetail, Mobbin, Linear, Reforge Build, Descript, and many other amazing tools that will help you as an AI product manager or builder succeed. I'll see you in the next episode.

Episode duration: 1:40:51