you2idea@video:~$ watch i13XK-uUOLQ [52:03]
// transcript — 617 segments
0:02 How can you make money from OpenClaw? Like, how can you spin up these OpenClaw
0:07 instances, these sub agents, these digital employees that can go out and
0:11 make you money while you sleep? Is it even possible? Well, in today's episode,
0:15 I brought on Nick, and he shows a tactical tutorial for how to spin up
0:21 multiple Open Claw machines in a virtual [music] instance, how you can be
0:25 automating tasks on Upwork [music] and these boring business automations, and
0:30 how you can actually make money from OpenClaw. If this doesn't get your
0:34 creative juices [music] flowing for the future of SAS, how people are going to
0:38 make money and how to actually use Open Claw [music] from not just a cute little
0:45 use cases, but actually money-m uh opportunities, then I don't know
0:48 [music] what will. I had such a good chat with Nick. It got my creative
0:53 juices flowing. I think it will yours, too. And this is, I think, one of Nick's
0:57 first podcast. So give him a like and comment to juice him up because he
1:10 [music] >> I couldn't be more excited to have Nick on the pod. He's one of my go-to people
1:17 when I have questions about OpenClaw. Nick, by the end of this episode, what
1:23 are people going to learn? Yeah, people are going to learn that OpenClaw is more
1:28 than just a personal assistant. You can actually deploy this into businesses.
1:31 You could drive actual business outcomes, generate revenue off of
1:37 OpenClaw as an opportunity. Um, and yeah, we're seeing it on X like people
1:42 who are deploying OpenClaw for um kind of executives or um individuals who are
1:47 super busy. uh they're making thousands of dollars, you know, setting OpenClaw
1:51 up, getting it up and running for these people and managing it for them. So, I
1:54 think there's a huge opportunity here and yeah, just excited to jump in. Cool.
2:00 And before we get going, I need you to make a commitment to me and to the
2:04 person listening or watching, which is I need you not to hold back any sauce. I
2:08 don't want to know just about the opportunity. I want to know how are
2:11 people doing it tactically. And by the end of this episode, what I want is for
2:16 people to take away like I want people to know how they can actually make a
2:20 dollar from this. And is that a commitment, Nick, that you are willing
2:24 to make to to us? >> Absolutely. Absolutely. I I'm not going
2:30 to hold back anything. I I'm I've in fact I think OpenClaw is is a is a tool
2:35 that allows us to be able to do the things that we have always been able to
2:40 make money from like automation with AI but do it even better and I'm going to
2:43 show you how to get it all set up so you can do that and the wedge to get going.
2:46 So >> let's do it. >> Yeah. So I guess jumping right in as far
2:52 as like getting set up with OpenClaw. You can see here you know this is Orgo.
2:57 This is our startup. You don't have to use Orgo to get started with OpenCloud.
3:01 Just full disclaimer, this is what I'm using. And what I'm going to do is I
3:05 have a project here. Uh you can see I have a couple a couple projects and I
3:09 have Greg. I set you up a project. I hope I hope you uh enjoy your, you know,
3:15 five computers. Um, and so you can imagine, Greg, you, let's say you're a
3:20 business owner and you have a busy life, you know, um, you got all you have the
3:24 podcast going on. You have all these businesses you're running, the agency,
3:28 the idea browser, all this stuff, and and you need you need help automating
3:32 some stuff. So, what I'm going to do is I'm going to come in. I'm Scrappy Nick.
3:36 I'm going to come in and I'm going to help automate some things in your in
3:41 your business, in your life, um, you know, as a busy executive. and I'm going
3:44 to get you set up with OpenClaw. So, when you open up a computer here, you
3:49 can see I have it open. This is the the Clawbot um computer I made for you. If I
3:56 actually just type in OpenClaw TUI, this will open up OpenClaw in the terminal.
4:00 And you can see I I actually already started like, hey, I'm Greg Eisenberg.
4:05 And it's all ready to get set up. And so, actually, what I could do is um I
4:10 could invite you to this project. And then you would be able to do this as
4:12 well, like in your terminal, you'd be able to spin this up and now you're
4:16 talking to Open Claw. So, um, super easy to get set up. Once again, you don't
4:19 have to use Orgo. You could use whatever you want. You could use, I know, uh,
4:24 Manis just dropped their, uh, their own version of like one-click deployment
4:29 OpenClaw. Um, also, Kimmy launched their version as well. X is down right now, so
4:32 we can't actually pull it up on Twitter or anything, but, uh, Kimmy launched
4:37 their version. And so, uh, there's all these options as far as getting started.
4:41 You could use a Mac Mini, whatever. So, the key here, Greg, with OpenClaw and
4:45 actually creating money from it is to have the wedge to know what is the
4:51 specific use case in a person's business that we're going to automate like first
4:56 because for when when you see open call on Twitter, it's very much a personal
5:02 assistant. It's exciting. It's fun. But all the demos that go viral, including
5:06 me, I get it. I I'm I'm guilty of this, too. All the all the demos that go viral
5:11 are a little bit, you know, kind of toyish. They're a little flashy, but the
5:17 the real power is in finding the thing that actually, you know, drives business
5:22 outcomes, saves time for a business, finding that and and building the
5:26 automation around that. So, I have something running here. This is my open
5:32 claw doing looking up products for uh a business that I uh deployed for this is
5:39 a a promotional distributorship. And what this is doing is it's looking
5:44 up products and actually downloading all the product information and then like
5:48 parsing all that information. There's all these reports it needs to download
5:55 and then uploading that into a Zoho CRM. um so it can essentially you know create
5:59 a central source of truth for this client. So this is a perfect example
6:06 here of like actually creating a agent you that openclaw deploys to be able to
6:12 automate something end to end. Um so just just to recap are we all good so
6:16 far? >> Yeah so a few things I I just want to talk about. So one thing is when you
6:23 showed you know that orgo screen you did you had like five machines running. So
6:28 what's interesting is you know I've got my Mac Mini going you know I've got one
6:34 instance so using something like this is cool um because you can have multiple
6:39 instances going right and you can see them all in one screen in one view. So
6:43 that's really cool that that was sort of an aha moment for me.
6:48 >> Yeah absolutely. Yeah, it's like everyone, you know, as far as where you
6:52 deploy your your main open claw, you can see here I like I starred the main one.
6:57 >> Uh, and so that's like you could have that wherever. But what people don't
7:02 realize is openclaw can spawn sub agents >> and I think this is going to be huge for
7:08 people who start like right right now we're at the phase of having one open
7:11 claw. It's going to happen quickly. you're already seeing on it on Twitter
7:14 memes about like, oh, what if you have a, you know, 10 Mac minis or all the Mac
7:19 Studios are being bought out. So, this is happening fast where you're going to
7:22 want 10 open claws, you know, 100 open claws. Right now, you could have one
7:26 open claw and just have it spawn up to, I think, eight sub aents and each sub
7:31 aent could have its own computer. And you could do this like this is why this
7:35 is like kind of where orgo shines is you could spin up multiple computers for,
7:39 you know, each individual sub agent of your open claw. And in this case here, I
7:43 had it looking on Upwork for actual things that we could automate with
7:46 OpenClaw. This is like a little hack here as far as like, well, I want to
7:50 make money with OpenClaw. Well, uh, oh, I don't know any any business that I
7:53 could reach out to that could automate stuff for. A great place to start is
7:58 Upwork because there's jobs on Upwork that are literally posted. They're
8:01 they're asking you. They're like, "I want to pay 500, a,000, 1,500, 3,000,
8:07 $20,000 for this AI workflow, and you can I spawned sub, it went viral on
8:12 Twitter. I spawned sub agents to go find all these jobs and then build out little
8:16 demos for each of them, and and then we picked the best one, and okay, let's
8:19 let's apply for that proposal with that." So, that's a little tidbit there
8:24 on on Upwork and all of that. So, which is which is kind of hilarious
8:29 because I mean Upwork, you know, is designed for human beings
8:31 to complete work, right? It's not designed for machines, let alone
8:38 multiple machines to complete work. Um, but I mean, as long as the the quality
8:43 of work is good, um, you know, customers going to be happy, right?
8:47 >> Yeah. And I think like it's good to, you know, it's good to treat it as a as a
8:51 starting point. like you know if you can save time preparing a proposal for a job
8:56 on Upwork it's like um what is that worth you know and if you could do just
9:01 a 100x volume uh what is that worth so there's a couple things on that of like
9:06 the parallelization of work with openclaw so there's like could you have
9:11 10 open claws working on a given task and it breaks up that tasks into 10
9:16 subtasks and so each open claw does one of those subtasks that's one way of
9:20 having parallelization but Another way is to have 10 open claws working on the
9:26 same task uh just 10 different instances of it. And that was kind of like what I
9:29 was doing here. It's you know four different instances of the one open claw
9:34 that you know or four different open claws doing the same thing of looking up
9:38 different jobs on Upwork that they could apply to. So that's kind of an
9:42 interesting topic there. Um but yeah, so like as far as OpenCloud
9:46 goes, it's it's it's uh a huge opportunity. I just want to like throw
9:51 this in here. Andre and Horowitz talks about computer use agents. I view open
9:55 claw as a computer use agent. You know, you're giving an agent a computer and
9:59 it's able to do it's able to use that computer. It's like it's a computer use
10:02 agent. But that's half the story. The other half of the story is for it to be
10:07 able to like click around actually operate a graphical user interface on
10:12 like le legacy softwares and systems. And so um you can imagine like you know
10:18 in here this this automation I built this is navigating a legacy platform for
10:22 this client that doesn't have any clean APIs and it's able to click in download
10:27 reports and actually you know be the universal API uh to be able to solve
10:31 problems that you couldn't previously solve without computer use agents. So I
10:34 think there's a huge opportunity here and and Horwitz they talk about it and
10:39 they say we believe that the properly uh to properly verticalize computer use
10:43 agents and assist companies in adopting it will be a major area of exploration
10:47 for startups and [snorts] this is like this to me this screams openclaw you
10:51 know can you create a vertical use case for openclaw for a business and actually
10:57 assist that company in adopting it I think that's that's the huge opportunity
11:01 here um So, going back to this like workspace we have set up for you. Um, you know, with
11:08 with with all the people setting up OpenClaw, you could set it up as easy
11:14 as, you know, I invite Greg to this workspace. I create him a new computer.
11:18 We could just do it now. >> OpenClaw 2. I select how much RAM. Let's do let's do
11:29 eight gigs. Launch that. [snorts] Open that up. And then all we need now
11:36 to get you set up is to get the um let me get the the curl command for open
11:42 claw. I copy that from their website and then as this computer loads
11:46 we'll be able to then literally just paste it into the terminal
11:52 and once I see the interface pop up boom. Okay. So now I'm going to hit
11:55 enter and now we're off to the races installing open claw. So like it's as
12:01 easy as that. And I think there's this is this in and of itself as like a
12:05 workspace where you can invite people and get them set up with openclaw or
12:11 clawed code. I think this is like a huge a huge opportunity as well of like just
12:15 there are executives right now that are like reaching out to me like law firms,
12:20 insurance companies, they're like can you can I just like pay you to teach me
12:24 how to use this stuff? So like that that's its own whole thing as well as
12:27 far as like if you're savvy enough to even know how to install open claw and
12:31 get it set up in the first place. Um [snorts] I just think there's a huge
12:34 opportunity around like just helping executives businesses adopt it. Um so
12:40 yeah you can see it's as easy as this to get set up. And then as far as like what specific
12:48 things can we can we automate with OpenClaw there's a couple it takes a little bit
12:55 of a designs thinking approach. So when you go into a business let's say you
12:58 find a project on Upwork and you want to you want to automate that with open
13:02 call. Let's say you go into a business and you're talking to the executive the
13:06 decision maker and it's clear that they have things that need to be automated.
13:10 Well, as far as like design thinking goes, like you need to have a a clear
13:15 way of like first mapping like all the different possibilities. You can see
13:17 here, this is something I did in the past of like there's all these different
13:21 things that you could automate and you want to map them off two very simple
13:26 metrics. What is the value that we can create by automating this thing and
13:30 what's the relative effort, cost and time? And so we ultimately want to start
13:34 with, okay, we want to automate things with Open Claw that are high value and
13:38 low effort, cost, and time. And that's like your your lowhanging fruit. Um, and
13:45 so you start there. And so like for for this client, this was like that this was
13:49 that this was like, okay, we're looking up products on this website. We're
13:52 downloading them. We're parsing all the information. Uh, that's the lowhanging
13:57 fruit. So start with the design thinking approach of like okay simplest fastest
14:04 to deploy and then you need to map out like the systems design around how like how
14:10 is this thing going to be automated right so for this for this client she's
14:15 like okay I send an email to a to a client of hers right she sends an email
14:19 to a client and she has a presentation link with all these products okay and
14:24 and all those products she needs to look all of them up and get all the
14:27 information on them and then upload them into Zoho. So then your next step after
14:33 identifying the opportunity is to literally like map this out. I use
14:37 Figma, you can use whatever, but map out the the actual workflow process of like
14:41 okay step one, step two, step three. What is this automation going to look
14:45 like tiptoail so that we can do the whole thing? Um because with open cloud
14:50 computer use now is now you can do that you can do you could do things from tip
14:54 to tail. It's not like um you know it used to be where you'd have to like go
14:59 into a website and click a button and then it'd be able to do some 50% of the
15:02 whole thing but then you'd have to copy that and paste that somewhere else and
15:05 do it on your own. Like we can do it tiptoail. Um, so quick recap,
15:13 install open claw into a computer, identify the next, identify the the
15:18 lowhanging fruit opportunities, the highest value opportunities, and then
15:22 begin to map out what that even looks like to begin with. Couldn't you, you
15:26 know, sort of, this is meta, but couldn't you use Open Claw to actually
15:31 do some of the priorit prioritization on the auto automations and actually, I mean, you as
15:37 a human being did the Figma, but couldn't you actually just use the open
15:40 claw or cloud code or something like that to to help you with that? So, for
15:47 example, like if you go back to the the Figma >> Mhm. Like
15:53 you could walk into a business and basically say, "Hey, I want to figure
15:59 out uh what we could automate here and you do customer interviews with
16:03 different people on the team. You record those customer interviews. You get the
16:08 transcripts. You upload the transcripts and then you say, "Hey, based on that,
16:12 then you're like, you can actually say, you know, you give this as a reference
16:16 image." basically say like, "Hey, I want to figure out which which automation
16:20 opportunities have the highest amount of value, lowest amount of effort, cost,
16:25 and time. Give me the top three and then create, you know, Figma." And I think
16:32 there's like a Figma MCP even uh that you can use and you can say like, "Hey,
16:35 like can you map this thing out based on these customer transcripts?" Does that
16:38 make does that make sense or am I >> Oh, absolutely. Yeah.
16:42 >> No, that's that's that's the way to do it. Like I whenever I do any kind of
16:47 call with a client or like potential customer, oh my gosh, Gemini for uh for
16:54 Google Meet is amazing. you have it take take all the notes and then actually
16:58 that's that's how I even got because I don't know about I don't know about you
17:00 Greg but sometimes when you're in these calls you kind of for you you know this
17:04 this this industry you know you're in a new industry you're helping this
17:07 customer you don't understand their domain expertise the lingo and so you
17:12 got to go back and and like okay like what what was it that they said and so
17:16 have the granola or Gemini notes or whatever and then literally ask it to
17:20 like okay what's the stepby-step workflow and then map it out it just
17:25 helps me to map it out visually. Uh but you could literally ask it, yeah, like
17:29 you said, um based on this transcript, what is the the automation workflow look
17:33 like? You know, step by step. >> And if you don't want to use, you know,
17:38 Figma, you can also even say uh you know, do output in mermaid
17:44 uh code and then you can use the mermaid code and insert that into an Excala draw
17:50 or a TL draw or something like that. So uh a little pro tip there.
17:53 >> Nice. Nice. Yeah. Yeah. And the Yeah. And the Figma MCP is pretty cool, too.
17:57 Yeah. Definitely check that out. So, >> yeah. So, then once you figure out what
18:03 the workflow is, right? Um this is where like you have to actually
18:09 um be able to know okay how much how much can I really ask like how much can
18:15 I just say like to openclaw right now hey like build this like hey build this
18:20 thing and you just describe the workflow versus genuinely using something like
18:26 cloud code to build out like what that workflow would look like with you know
18:31 Python APIs a genuine automation pipeline and process that your open claw can actually just
18:38 trigger upon um whenever it's like contextually relevant. So for instance,
18:43 this this whole pipeline here of like going to these websites, looking at this
18:46 product information, downloading the information, parsing it, uploading it to
18:52 Zoho, the trigger of all of that is uh you know the open claw being CCD in an
18:58 email and it seeing that email it has a link that is relevant for this type of
19:02 workflow to be triggered. So that that is like the thing that's like the
19:05 listening event that open clock can do with like a cron job that it sets up to
19:10 just like okay listen for this trigger and then once that trigger starts it can
19:16 then activate the whole Python script workflow automation everything that you
19:19 would need downstream of that. So you're not relying too much on open claw's like
19:25 um abilities in and of itself. you're more of so creating specialized AI
19:29 workers under each uh underneath the open claw that it can call individually
19:34 if that makes sense. >> You know, you talked earlier about sub
19:37 agents. I think a lot of people are confused about what is a sub agent
19:42 versus a task and stuff like that. Can you just clearly explain that?
19:51 >> Yeah. Um so sub sub aents are um I I I there's a couple ways to view them,
19:55 right? So like in the context of cuz there and the reason I say there's a
19:57 couple ways to view them is because there's a couple ways of using them. So
20:02 in the context of like open claw you can ask it to spin up five research sub
20:06 aents that all go and research some given task and and like and you know
20:10 like I said earlier you could you could have it parallelize that task across you
20:15 know splitting it up across each sub agent or having each sub aent actually
20:19 go and do the same task you know across five different instances. But the next
20:24 thing around sub aents is that you can actually like you said um maybe think of
20:29 them in terms of skills. So if you're familiar with anthropic skills um you
20:34 can have like these specialized instructions and rules along with actual
20:39 code that you can provide to your agent for it to be able to go and do a given
20:46 task. And this is really nice um because it, you know, it gives you a very more
20:51 per uh a very more powerful general purpose agent that can do many of your
20:56 specific nuance tasks um across various domains. But the thing is is like I want
21:03 my general agent to be freed up and uh to more so just be the orchestrator. And
21:09 what if the the general agent this one right the the one I have start here can
21:15 just call a sub agent like worker number four here to do a given skill that you
21:21 have created. So if your skill is that it goes on Twitter and finds the most
21:26 viral ideas and it bookmarks them. Um rather than having your main agent do
21:29 that and now you can't talk to your main agent for the next 20 minutes because
21:33 it's working on that. Can it call that skill into a sub agent and have the sub
21:38 agent do that? That I think is where things get really interesting and and in
21:42 terms of the context of like deploying open clause for businesses. I would
21:46 think of everything that you have in terms of an AI automation opportunity
21:52 around workflows, skills, tasks, etc. I would actually just create that as its
21:58 own like specific sub agent with its own skill that that your open call could
22:03 then call if that does that make sense? >> It does. It does.
22:08 >> Nice. Um it's it's you know I think the basic idea is like you know in layman's
22:14 terms it's as soon as you have your um your open claw instance you know do
22:22 something they're busy you know it's like they got to they've got a a mug of
22:28 hot coffee and so you and your your job is you want to leverage this as much as
22:33 possible. you don't want, you know, your agent to hold a a hot coffee. So, if if
22:39 you ask it to do to move this desk into this area, you know, it it says, "No,
22:43 I'm holding a a cup of hot coffee. I can't do that." So, what sub agents do
22:49 is it basically creates leverage for your open claw and it basically says
22:53 like, "Okay, I'm you're going to create a set of sub agents who are going to be
22:59 good at XYZ thing." And that way it frees up your your main agent to to you
23:07 know as you say orchestrate to basically be the manager of the sub agents. Uh and
23:13 you know what that could mean is uh like looking at um quality of work. It can
23:19 mean uh checking for certain things and stuff like that. >> Exactly. Exactly. I think that's going
23:25 to be huge. You know, when when you start working with these businesses and
23:29 and customers who want things to be automated once you show them what's possible,
23:34 their eyes light up. They get all these ideas themselves. These are high agency
23:38 people, you know, they they come up with creative ideas that they want to start
23:41 implementing and and then what you realize is there's just a a huge a huge
23:45 list of things that can be automated and they're excited about that. And and so
23:51 actually like the the ability to okay first solve a vertical specific workflow
23:56 for for a customer and then that opening up their mind and then them being like
24:00 oh I wonder if I could could I text this thing and it does this. This is kind of
24:05 where like the whole open claw moment um is really powerful. It's like it's it's
24:10 the it's the it's the assistant like capability. It's the you know I have it
24:14 I have it here. It's like what a lot of people might get confused about. Why is
24:18 it that open call is so special? It's it's the ability that that it has its
24:23 own computer. It's running 24/7. You can text it and you can schedule tasks and
24:29 really if if we just removed OpenClaw from this card here and you just called
24:34 this a really good employee, it it would just make sense be like, "Oh, works 247,
24:39 can code, can schedule tasks, I can text it, and they have their own computer."
24:45 Um, so I think that's like kind of why Open Claw is exciting for a lot of
24:48 people. And if it's not, you know, if some people think it's it's overhyped,
24:51 you have to kind of look at the whole picture, I think, and and then you're
24:54 able to really gauge it. >> So I I'm certainly bought in on this on
25:00 this idea that, you know, it potentially it could be be a really good employee. I
25:04 think there's also cases where people aren't setting up their open claw in the
25:07 right way where it ends up being a bad employee. Um, and I think, you know,
25:15 that's sort of like the the issue with that is, you know, sometimes you have a
25:19 bad employee because the manager, the coach essentially is is not doing a good
25:23 job at giving the right context at the right time. So I think you know do you
25:29 have any tips and tricks around um besides spinning up sub aents like how
25:33 how could people listening to this if they want to go after this opportunity
25:39 of um essentially verticalize you know open clause uh and automating some of
25:47 these flows. How could people actually you know take their open claw from a bad
25:52 employee to a good employee? Yeah, I think it comes down to let's maybe we
25:57 should walk through. So I know you have I was actually looking at this. So idea
26:02 browser. So for everyone if you don't know Greg has this amazing product idea
26:07 browser and I love this trend to this idea today Tik Tok trend tool that
26:14 catches viral waves before they peak. So I was actually looking at this this
26:16 morning. I was like, "Wow, this is actually something that you can maybe
26:21 turn into a skill for an open claw to create a specialized skill around this."
26:27 So, let's just do a live. Let's see. I I genuinely don't know how far can we get.
26:31 Can we build this out now? Let's see if I copy all of that. And now I'm here in
26:36 the open claw that that we set up for you. And you can see it just got set up.
26:39 It's it I just told it, hey, I'm Greg Eisenberg. So, we're getting started.
26:44 That's it. And let's just say I want to build a um a a specialized skill to be
26:53 able to do the following. And I'm just going to paste that entire um idea
27:00 browser idea. And I'm going to ask it as far as like creating these automations,
27:04 creating workflows, or doing anything with OpenClaw, my number one tip is
27:09 always ask it to ask you questions. So, what do you need from me
27:16 um to be able to build this out? Um let's create a plan. And so, a lot of
27:22 people need to remember OpenClaw is like a almost a little bit of a wrapper
27:28 around like clawed code in a way. Um so, let's see. Okay, cool. This is a big
27:33 vision, Greg. I like it. Let's break down um a realistic build as an openclaw
27:38 skill. So now it's saying okay I need data access I need the scope I need the niche focus
27:48 um that'll shape what we build a lean skill so the first thing that we can
27:53 maybe do is here in like orgo we have this playground mode and you can ask the
28:01 playground agent here to do things in this computer in this this computer
28:05 environment. Um, so one thing I might test first is like can we get can we get
28:11 this agent to even just spin up Tik Tok and just scroll Tik Tok and identify
28:16 what is on the for you page of Tik Tok. So let's like maybe start there. Does
28:19 that sound good? >> Yeah, absolutely. >> Let's do that. Open Firefox.
28:26 Go to Tik Tok. Scroll. I'm gonna say scroll Tik Tok looking [snorts] for what the most
28:37 common videos are on the for you page. Give me a summary. So, let's see how
28:43 it's able to do this. So, this is using our our playground mode
28:49 and It's going to go to Tik Tok. And for those listening, I'm just gonna talk
29:01 through a little bit about what this agent in our our playground is doing.
29:05 It's it's visually interacting with the screen. It's, you know, clicking into
29:08 Firefox. It's opening up the browser. Now, it's typing in tik tok.com.
29:17 And let's see. This is always such a magical experience just watching a computer navigate the
29:28 web like a human being. >> It's amazing. You know, this is Yeah,
29:34 there it goes. Uh it's on it's on the it's on the homepage. And now it's going to it's going to
29:40 actually scroll TikTok. It's going to probably take a screenshot of this. Get
29:44 the context of based off that screenshot of what the video is about. You could
29:48 even see it. It has hashtags movie hashtag for you page. So, it's going to
29:51 be able to infer a lot of things. Boom. It scrolls. It scrolls. It gets a popup.
29:58 It's going to close out the popup. Um, but on your point, Greg, Dario Amade,
30:03 the CEO of Anthropic, he just had a podcast with Doresh, and it came out a
30:09 couple days ago. And you know what's really interesting what he said in that
30:14 in that um in that podcast? He said >> he said this idea of his around this
30:20 data center full of brilliant um you know scientists and uh Nobel prize
30:25 winners essentially his concept of what AGI will be like. He says, "The
30:30 constraint to getting there is computer use agents. The ability to have an AI
30:35 that can operate a computer like you and I can, but better. You know, interact
30:39 with the visual interface, also be able to do things under the hood, kind of
30:44 like cloud code." Um, this is the constraint he said. And I mean, it makes perfect sense. If if it
30:50 can do anything that you and I can do on a computer, that seems like it can go
30:56 pretty far. Um, openclaw is like a chat GBT moment, I think, for for this kind
31:03 of idea of computer use. And and and as far as like building these computer use
31:07 agents out, you can see this is this is clearly working. So, we know this is
31:12 possible. Um, you can build a lot of people might get confused with Orgo when
31:15 they come to our site. They see computers for agents. They're like, what
31:18 does that mean? Well, I think they get it now. It's like okay you want your
31:21 cloudbot to have its own computer but also we provide and I'll show this we
31:28 provide the doc in our docs like we're actually provide the programmatic APIs
31:32 so that you can create custom computer use agents that um do a given task very
31:38 well. So you can bring any model. You can get Kimmy 2.5 is like the super
31:42 cheap um you know Chinese model. It's very good at computer use and you can
31:46 give it the ability to click, drag, scroll, type in the keyboard, spin up a
31:51 computer. Um and you could create specialized really fast, performant,
31:57 lowcost computer use agents um using our docs. And I just say that to say like as
32:01 far as you know this this thing that we're doing here creating a skill around
32:06 scrolling Tik Tok. Um I think that's how we can we can actually give it a shot is
32:11 let's just we see this is working now. Let me just grab the ordocs and let's
32:16 start building this skill out. If you didn't see here I'm at the Ordo docs. I
32:23 click this thing called llms full.txt. This is like all the instructions for
32:26 the LLM to be able to build on top of Orgo. And what I'm going to do is I'm going to
32:33 wait for I have to probably I'm going to wait for this to finish this its task
32:39 and then I'll be able to tell it. Um but [snorts] while this goes, Greg, you have
32:43 any thoughts? Um, well, just, you know, sort of a thing I was thinking about is
32:47 it sounds like when whenever you're trying to, you know, do a new automation, you start
32:54 by thinking about what is a lightweight skill that I should create. Is that
32:57 correct? >> Right. Exactly. What's the MVP? >> Yeah. So, you start with like a
33:06 lightweight skill. Um, you test it and then from there you're probably like,
33:10 "Okay, here's what went wrong. here is what could be better that sort of thing
33:16 >> right exactly just fine-tuning you know debug like I think the design thinking
33:22 process around all of this is super important of like okay if I want to
33:29 build a car maybe the first thing I do isn't to build the frame of the car or
33:36 you know to build um you know the whole body of the car that's not the first the
33:39 first thing I should do if I want to build a car. Well, why do I want to
33:42 build a car? Well, I want to build a car to go from point A to point B. Oh, okay.
33:46 So, really, maybe I should start by building a skateboard. How can we accomplish the task to get
33:55 the get the dream out the dream outcome? And um and sometimes that means like
34:00 starting with something that's completely different than the end state.
34:06 Um so, this is done here. I think it interrupted itself. probably a context
34:11 thing. But now we can actually what we can do. This is all live. So we're
34:16 you're seeing this in real time. You [snorts] can literally there's a couple
34:19 ways to go about this. You can actually install cloud code into this computer
34:25 and have it build out uh the automation in here or we can just ask this agent to
34:32 do it for us. So let's see. Um I want to build a computer use agent
34:38 that does this exact thing. uh but more programmatically using the Orgo API docs
34:48 and I paste that here. What do we need to get started and we send that off? All
34:53 right, here we go. So, we need an Orgo API key. Here's the architecture of what
34:57 we'll build. Okay, this looks good. Um orgo [clears throat] key, anthropic
35:06 key. It has all the code here. Cool. Um, you're looking at a Tik Tok
35:11 video. Extract the username, video description, the category, the
35:16 appropriate like count. Boom, boom, boom. Okay, cool. So, [snorts] I have
35:22 all these things already. And don't worry, I'm going to delete these keys.
35:25 So, I'm not worried about leaking or anything. I'm going to copy my Orgo API
35:29 key. I'm going to paste it in there. And I'm going to say Orgo API key. I also have my enthropic
35:37 key. Let me grab that. I'm going to paste that here. Can we build this out?
35:45 And so now we're going to have our our playground mode build out this computer
35:48 use agent to be able to go do this thing that we just tested out. We know it
35:52 works. We know we can do it. Let's turn it into something that could be more
35:55 programmatic, kind of like a skill. And let's give it to Claudebot so it always
35:58 has access to it. >> The dream. It's literally any idea you have, you
36:12 >> And that's sort of the arbitrage opportunity, right? Especially like in
36:17 our world, of course, like we're so used to this now. Uh even though it's only
36:23 been like two months. Uh but you know the opportunity is the vast majority of
36:29 people on this planet and businesses would love to have better automation and
36:34 would love to, you know, have computer use agents working for them aka really good
36:42 employees working for them. Um but they don't know how. And so I think which is
36:46 which is cool that you're like you're showing us like some of the best
36:50 practices on on how to do it. >> Exactly. didn't. >> And I think this is also like I I can
36:55 imagine, you know, the the whole audience of this podcast, we're all
37:00 pretty techsavvy. We know how to do things like vibe code and and and you
37:03 know, play around with cloud code and be able to do these things and we take it
37:08 for granted in terms of what that value is worth. >> Um,
37:13 >> as far as like OpenClaw and the opportunity around that, I mean,
37:17 OpenClaw started going viral on Twitter around two to three weeks ago. only now
37:21 is it starting am I'm starting to see it's starting to go viral on Tik Tok a
37:28 little more mainstream. So um yeah I I just mean you know people are
37:34 catching on and a lot of people still need help with getting you know up and
37:37 running on this type of stuff and what you might think is oh I mean I have a
37:41 basic understanding of open cloud and cloudbot and cloud code uh and you might
37:47 under underwrite that. a lot of people find that valuable. So, being able to
37:51 help businesses adopt it or just people in general, I think it's super good
37:54 opportunity. Um, >> yeah, I think my only advice for people
37:59 would be to f focus like don't be everything to everyone. Like
38:07 don't don't help any every business. help real estate agents, for example, or
38:12 like pick a vertical that maybe you have some unfair advantage for some
38:16 particular reason. And that unfair advantage doesn't necessarily mean you
38:19 have 20 years of experience. It might mean just that you know you want to
38:23 build something for uh real estate agents cuz your mom was a real uh a real
38:28 estate agent. So you you know you know the customer, right? Um so I think
38:36 that's that's the way to do to do this. Absolutely. Yeah, it's it's um whatever
38:41 you know that's your advantage. And I mean I guess as far as like some maybe
38:46 this doesn't apply to everyone. Obviously if you're in this industry
38:49 then go for it. But like you know probably things to avoid things that are
38:53 have a lot of like red tape like healthcare, finance, you know I
38:59 recommend maybe start in something. >> Yeah. like you said um you know you
39:03 could do even manufacturing or or uh there's a lot of distributorships out
39:08 there who you know they distribute uh merchandise um just like the the demo I
39:12 was showing earlier of that computer use a agent working um so yeah I I think as
39:18 far as you know doing doing what you know and starting there and then the
39:22 market will tell you you know the market will pull you into specific verticals
39:27 you'll start seeing ways that as you build these spec this is a Great point.
39:31 As you build out these specific workflows around these verticals of like
39:35 let's say you do, you know, let's say you do the manufacturing thing and you
39:40 do manufacturing for doors. Um over time you're going to have a workflow for
39:43 almost like every kind of thing you could imagine in that industry. And if
39:49 you have the agents all built out, can you imagine you have a workspace and you
39:55 invite uh you know some some new company into this workspace that uh for for
39:59 automations, for manufacturing, for doors, like luxury doors and you invite
40:03 them and they have all these AI employees in the workspace in Orgo set
40:07 up ready to go. You just see them all here and they're all ready to go. It's
40:12 like you feel like you just hired not a person but a team. Um, I think that's
40:18 that's that's a very near future. In fact, I don't think there's anything
40:21 stopping us from having that right now. It's all about just, you know, who's
40:25 going to go out there and and put in the work to actually do that. And if you do,
40:29 I think it's I mean, it's it's pretty clear. >> Yeah. I mean, and that's that's why I
40:35 truly believe that, you know, agents are the new SAS, like you know what I mean?
40:40 So, yes, I agree with the like the vision you painted. I think that, you
40:45 know, in the past, you know, we created software that we would sell to these
40:50 businesses and then they would have people actually, you know, press the buttons, touch the
40:57 knobs to to make it useful. Now, you don't you're not going to create
41:00 software and invite them to the software. You're going to create agents
41:05 and you're going to invite them to the agents and then the agents are going to
41:10 do work that creates value for these companies. So that's the mindset shift
41:14 and you know it's it's only recently actually that people have been you know
41:22 I think over the last like two weeks I would say 2 3 weeks that people have
41:25 been like on X talking about this um h how agents are the new SAS but I I do think that um
41:36 like over you're going to see over the next two to three months like some
41:40 really big winners um and you're going to start to see it work. So, I'm excited
41:43 for people listening because I think that um this is the a type of audience
41:51 that will act on some of this stuff and it'll be interesting to see what what
41:54 happens. >> Yeah. Yeah. I I I think it was it was it Sam Alman that just said, you know,
42:01 every company is turning into an API company. >> Yeah. And that's interesting because
42:07 interfaces are in a sense dying in that way of like you know you won't interact
42:11 like the ultimate interface for whatever reason seems to be chat and text message
42:19 and it's happened twice now of like okay the chat GPT moment was a chat box and
42:23 now it's the openclaw moment which is like a text message or telegram so it's
42:29 like okay chat has happened twice um and that just means okay so we just want to
42:33 wait for our agent to be able to use all the tools that we use and we don't
42:37 really care about how it does it. It just needs to be able to do it and and
42:41 it runs in in the background um and does it under the hood. Um
42:47 so here, okay, I started uh I built you can see that this agent in the
42:51 playground built out the Tik Tok agent.py inside the computer. So now I asked
42:58 OpenClaw, hey um I built a Tik Tok agent.py in your desktop. Can you take a
43:03 look? And it's re it says, "Okay, I've read it. Here's what I see. Uh there's a
43:09 skeleton for, you know, using Orgo um plus uh Anthropics API."
43:14 Go ahead, Greg. >> No, you keep going. >> And and it was and it was um you know,
43:21 all all the actual Tik Tok logic, uh trend detection, all this stuff. Um it's
43:25 like, okay, maybe we should build that out. >> And it's like, oh, you got the API keys
43:31 hardcoded, etc., etc. So, let's just say Let's use this script build on top of it
43:40 or whatever you need to do and let's spin up a orgo VM inside of
43:48 the Greg Eisenberg workspace to accomplish this um automation with
43:54 Tik Tok. Let's demo it just working. We can spawn a uh sub aent and VM in this workspace.
44:09 And let me just give it a API key just in case it needs that.
44:14 What really blows my mind about this you know, I was going to say we we're
44:21 we're building a business in a very short amount of time, but it's really
44:25 like we're building an asset. like the amount of assets that people are going
44:30 to be be building um using tools like this is is going to be crazy, right?
44:39 >> It's insane. Honestly, it comes down to like honestly like it comes down to
44:46 taste now. Good ideas. Um [laughter] cuz like if you have a good idea, it
44:51 could just build it and yeah there's there's going to be like oh my goodness
44:54 what's going to happen with all of these assets like you said that people are
44:58 just going to build and build and build. There's going to be so many assets. I
45:01 think is this what is this what we mean when we talk about the um the abundance
45:07 that AI will bring you know of solving all these problems? >> Yeah. Well, I think what ends up
45:14 happening is, you know, unfortunately, there are going to be more and more layoffs as AI helps
45:22 with productivity. At the same time, I think there's going to be a
45:27 renaissance, the golden age of entrepreneurship and people creating
45:32 assets, products like this, oneperson businesses, and uh that's how I see it playing out.
45:43 >> Yeah. even even maybe may we don't know officially but maybe it's already
45:46 happened with Peter Steinberger the creator of openclaw you know he he just
45:51 um officially announced he's joining open AAI >> like I think it was just him who built
45:57 Open Claw how much did he get you know aqua hired for I think it was a lot
46:01 >> um >> so that's that's I mean that's that's cool I think this is the best time to be
46:09 a builder tinkerer to get creative Um, I I'm excited to see what people
46:14 build with Open Claw and computer use agents in general of like
46:18 there's so many things, you know, whether it's a a super fast chess
46:23 computer use agent or if it's something that's genuinely driving an outcome in
46:27 your business. Like I just think there's so many things that can be built. Um,
46:31 playing around with these tools, getting familiar, learning how to how to
46:36 leverage them. Yes, AI is going to be replacing a lot of jobs, but also it's
46:41 going to enable a lot of people to do things, you know, that they've never
46:43 been able to build before, and now they can do it. So, um, yeah, I'm I'm just Oh, here it is. Okay,
46:51 so you can see >> it spun up this computer, Tik Tok Trend
46:56 Hunter, and my screen's just refreshing. Uh, let me just click into it
47:03 and and I could just tab back and forth a little bit to see. Okay, VM is up.
47:07 It's opening Firefox. Let me wait for the agent loop to start. Here we go. So,
47:11 it's spun up its own computer. This is insane, Greg. It's spun up its own
47:15 computer. OpenClaw did. And now it's it's using its own Python script that it just made
47:22 just now. We took it from idea browser. And now it's going to go do this thing.
47:26 [laughter] um that we just built out. I don't know how long this took. Last like 10 minutes
47:32 kind of just, you know, in between we're talking, having our coffee. It's like
47:35 [laughter] this is insane. I don't know. It kind of It makes me It gets me giddy. It's like
47:41 [laughter and gasps] uh and it's going to figure this out. It's like, you know, it's going to debug
47:47 like, okay, what's going on? Why am I on the ads.tiktock.com? Let me reroute
47:51 myself. I'm sure it's going to figure all this out. Uh, but it's just cooking,
47:55 you know. >> Crazy, dude. Crazy. Anything else you want to cover
48:04 before we head out? Yeah, I think um as far as other things to cover, I mean
48:13 I I just want people to start thinking about these tools, you know, yes,
48:18 they're they're once again they're great personal assistance, but if you start
48:23 thinking of OpenClaw as an NAN or [clears throat] you start thinking of it
48:28 as um you know, like a Lindy AI of like you know people there are real
48:31 businesses right now like we can go to Upwork right now and find jobs that are
48:38 being posted around, you know, things like this. This is what you do. You just
48:41 go to Upwork. Upwork's great because you get to see what the market's asking for
48:46 and you just type in uh robotic process automation. you know, this old outdated
48:53 way of of programmatically automating tasks that is like clunky and it breaks
48:59 and it's not intelligent. If the button isn't in the exact UI space that you
49:05 just delegated it to, it won't work. And you go here, look, posted yesterday,
49:09 Android RPA automation posted yesterday, automation pipeline for client upload.
49:14 You go here. You could just do let's let's find $500, $1,000, $5,000. Let's
49:20 look at all these projects. Okay, this one. Boom. Right here. Um $1,000 budget.
49:27 I'm looking for experienced automation engineer to build desktop automation
49:32 computer use uh for my software business. We sell a specialized dynamic
49:37 PDF. You take all this context, give it to open cloud, give it to Claude code.
49:40 How much of it can you build out as a demo based off of this context alone?
49:45 Send a proposal. You have your first customer right here, $1,000.
49:51 Um, get get some case studies, leverage that. Maybe go deeper into this industry
49:56 that this person's in. Start building out specialized workflows using cloudbot
50:00 open claw for all the different vertical use cases in that. Create a workspace of
50:04 it. Um, I think that's where we're at right now and I'm excited by this. So,
50:09 um, yeah, I I guess we'll just see where it goes. And yeah, okay, this needs a
50:14 little debugging as to be expected. We spent 10 minutes on it, but I think you
50:20 get the gist. Um, and yeah, I'm excited to see what everyone builds.
50:25 >> From your lips to God's ears, baby. I I'm I think your approach makes complete
50:31 sense. Um, it's the exact approach I would use. Totally recommend it. people,
50:37 get your hands dirty, get tinkering. I'm excited for what you build. Um, Nick
50:43 doesn't do a lot of podcasts. Um, I think I was only able to find him do one
50:47 live stream before. So, show him some love in the comment section. Uh, like
50:54 the video. Um, show him some love and, um, Nick, I hope you come back on. Uh,
51:00 share more use cases. I'm going to be sharing more use cases that I'm using
51:05 that my you know I haven't done uh too many publicly but I'm going to be
51:09 sharing more of my use cases both on virtual machines and on my own Mac Mini
51:14 uh for my open claw stuff. So get ready for that folks. And uh Nick wait is
51:19 there anything is there is there anything else before before we go that
51:24 you want that you want to share? I just want to share with you, Greg. You don't
51:29 know this, but I've been a longtime follower. >> This is my YouTube Rewind 2025 top.5%.
51:39 So, if you're watching this and you love Greg's podcast, you love his videos, you
51:44 you get building on top of it, you put cool stuff out there, you know, you join
51:48 your you join your top YouTube channel on your on your re your rewind of the
51:51 year. >> I love it. I love it. Nick, you're a legend. You got to come back on. You're
51:59 one of us. You're one of us. So, uh, appreciate I appreciate that. Thank you,
$

Making $$$ with OpenClaw

@GregIsenberg 52:03 19 chapters
[AI agents and automation][content creation and YouTube][open source and self-hosting][productivity and workflows][revenue model and pricing strategy]
// chapters
// description

I sit down with Nick Vasilescu, founder of Orgo, to break down exactly how people are turning OpenClaw — the open-source computer use agent — into a real revenue stream. Nick walks me through live demos of deploying OpenClaw for business clients, shows how sub-agents and parallelization multiply output, and shares his design-thinking framework for identifying and automating high-value workflows. We even build a TikTok trend-hunting agent from scratch during the episode to prove how fast you can

now: 0:00
// tags
[AI agents and automation][content creation and YouTube][open source and self-hosting][productivity and workflows][revenue model and pricing strategy]