The creator using AI to go viral | Will Weinbach

0:00While I was trying to improve upon my agent, I prompted it wrong. And this is for the company account. And it posted like four times at 4 a.m. And I was like dead asleep. But I wake up in the morning. I'm like, thank God this is on brand. Because if it wasn't, I wouldn't be going into work the next day. I don't want to spend the money on a Mac Mini. But if my television can connect to Wi-Fi, why can't I just run an agent on it? And then I called Parth at like 3 a.m. What do you think about this? He's like, I'll believe it when I see it.

0:25Parth Patil:I was like, I don't know anyone that's put an agent in their TV. And then I'm like, why would you put your agent in the TV? can't you just control the TV with your agent? I get a message at 5am. He's like, it worked hard. Holy shit. The token grantee program gives$1 ,000 a week in tokens to high potential creators already deep in AI across film, gaming, comics, print, and digital art. There are no tool restrictions. The freedom to choose is the point. Grantees also get access to my own custom fleet of agents and an ongoing collaboration with me. The goal, to close the gap between an idea and the world.

1:02Will, welcome. And Parth, introduce Will to the folks who are listening.

1:10Parth Patil:Will and I met about five years ago on Clubhouse, the audio social network that went viral during the pandemic. I was working on the app. Will was one of our most prolific users, one of the most creative people I met through the platform. When I met Will, I realized pretty much immediately, oh, this guy is, this guy, when he puts his mind to something, he will put more energy towards that, achieving that than anyone else. It doesn't matter if he starts at zero. What he gets done in a year, how much energy and time he puts into getting good at something, you can only guarantee that he will have something figured out a year from now that you haven't because of how passionate he is.

1:44Parth Patil:And I saw that with Clubhouse. I saw that with, like, him learning music, didn't know music, then became very good at producing music within a year. When Language Models came out, he was using Google and whatnot. But a couple years ago, I think I basically added him to my Replit Teams plan. I have a personal, I have a family plan. I call it my family plan. It's just a bunch of friends that I put on Replit because I'm like, guys, guys, Vibecode, just make something. Just make something. And I put Will on it and I was like, I'm just excited to see what you'll make. You're a creative guy and not an engineer by trade.

2:16Parth Patil:So it's sort of an experiment for me. It's like, what happens when you just put someone that has an idea or like a bunch of ideas and you pair them with some of the most interesting, powerful new technologies available? You know, what can you create? And I'd say since then, it's been quite a journey. So knowing I picked Will even before this program, I was like, oh, I can't wait to see what he does with these new tools. Now, I feel like Will teaches me more than I teach him almost every single day with these tools. So welcome, Will. Excited to have you on, man. Thanks. That was a great intro part.

2:47I love that. Yeah, I'm excited to be here.

2:51Parth Patil:Okay, so just a few years ago, Will, you hadn't written a single line of code. And I guess how did you fall into Vibe coding, making software, making apps? Like, what's your kind of like road into this world? Well, like you said, I use ChatGPT for everything, but not in the way I use it now. Like I would ask a generic question like, you know, hey, I have this essay that I need to write and you're going to write it for me. And that's that's how I was using it to start off. And then obviously I had met Parth through Clubhouse and Parth came to me and was like, you're using this all wrong. And I was like, dude, no, I'm not.

3:36And it turned into, hey, think about your pain points from workflows and then just build the answer. I was like, what do you mean build the answer? And it turned into like learning codecs, not how to code. I still don't know how to code to this day. I don't know anything that it's doing, but I think that also serves as an advantage. You know, I learned how to prompt better and now it's turned into how to create automations and how to build workflows and then how to enlist agents to do my everyday tasks that I don't really want to do. And it's a superpower at the end of the day. So how does the kind of being able to bring a beginner's mind, like the coding thing you were talking about and things like the ability to kind of start fresh, and where does that then build into important expertise?

4:27I think the best analogy I can make is when I was starting to learn how to produce. I would just put a bunch of things in the DAW and it would mesh together to make a sound, whether it was good or bad. And then you start learning the technical details of, okay, I need to EQ and this curve needs to be here and now my DBs need to be at this level so the mix sounds a certain way and it clouds your creativity and judgment in a way where now you're getting no progress because you're fixated on certain details. And I think for me, when it comes to coding, I don't know the backend. Like, I don't know what a lot of this means.

5:01I don't know how to pronounce half of the things that I'm reading on the codex screen. Parth constantly corrects me.

5:07Parth Patil:Actually, I stopped correcting. I stopped correcting Will because I think it's funnier if he gets five years into his career as a vibe coder and still doesn't know how to pronounce like repo. He says, he calls repo repo.

5:21Replite. All of that to say, I think that not having any expertise in the industry allows me to be more creative. And I think the creativity plays into new learnings and a willingness to try new things that others might be hesitant to because they know the potential damages that I don't. And frankly, there's a bit of an internal perception where I don't really care either. Like I'm willing to take the risk to get the reward. Kind of along those lines of Will's willingness to experiment is if I think about like this year, especially since February.

5:59Parth Patil:February was when OpenClaw kind of went viral. It came out in November, but it went viral in February. And I was kind of hesitant installing it. I was telling Reed, we had our own remote agents. And I was like, we got agents that are not insane that are working pretty well. I'm not sure I want to bring OpenClaw into this yet until I figure out a way to contain it in its own little Mac Mini. And so I installed OpenClaw on, I think, February 15th of 2026. And at the same time, Will was deep in codex. And then I told Will about OpenClaw. And him, you know, him being more experimental, having maybe less existing stuff that he was worried about losing, he installed OpenClaw immediately and started studying this new paradigm, this new kind of remote agent paradigm.

6:45Parth Patil:More, and I think deeper than anyone I know, even technically. So I'm now relying on Will to help me unpack this new paradigm of the remote agent that works 24-7, that can use its own computer. And then six weeks later, Will introduced me to an even better agent, in my opinion, the Hermes agent. So Will, maybe speak more to that. Your exposure of you went from coding agents, vibe coding, to this like, now you have these remote agents, the Hermes, the OpenClaw type systems, which are now becoming more mainstream in industry as the rest of the year starts playing out and these systems get more productized.

7:23Parth Patil:But they're very hacker projects. What's your experience been with these? Right. I mean, I got into OpenClaw originally because, you know, Parth, you recommended it. And originally, Parth also recommended installing OpenClaw on a VPS because I could potentially lose everything on my computer. And I did that, but then I realized I don't know how to work a VPS. So then I was just like, you know what, whatever's on the computer can be replaced. So we installed it in Terminal, and I was fascinated by how you could program the language to speak a certain way to match your work style. But I thought that OpenClaw lacked a certain amount of visibility that I think Hermes gives you.

8:04So then, you know, we were browsing X one day and came across Hermes as a recommendation. And obviously there's just a fascination with installing and trying. So installed Hermes and found it to answer a lot of the questions that were left not answered by OpenClaw, such as, you know, hey, I have visibility into exactly what this agent is doing. Not that I know what it means, but I have visibility into the agent, why it's creating a skill, how it's improving on previous skills based on my prompting all autonomously. And I think what these agents offer that maybe Codex doesn't is A, a user's POV as to what features could enhance agents in the first place.

8:54I think obviously once OpenClock came out, Codex started to implement a lot of the features that they had created on the platform. And it feels ahead from a feature basis than, you know, some of the premier model platforms. B, working style. I think the working style, you can obviously program it to be however you want from a personalization standpoint. And I think that's another value add to a user. and then C and one of the most important aspects is the memory context and being able to tell it virtually anything and being able to save that and store it in memory to then act upon it at a later point, which I think some of the frontier platforms are getting better at.

9:36But at this point, there's so much memory logged into my Hermes agent in the last three months that it's too valuable to give up, even though it's just files. I don't want to transfer it. So one of the things that's very cool, because I now understand where Hermes came from and entered into our ecosystem. Sorry. No, great.

10:00How did you discover, you start with OpenClaw, then jump to Hermes, why Hermes? And then from there, what are the other things that you're seeing that are part of some of the, call it the agentic framework of the near future? I'm on social chronically. In the best and worst ways. And I think, especially in the world of AI and the age of AI, social is a great place to find user recommendations as to what are the pain points, what needs to be solved, and what are the solutions. And I just love to test anything I possibly can. So that's how we came across Hermes, test it, and then it was the answer for us.

10:41A lot of the things we test are not the answer, but Hermes was. In terms of what I see the future being, I think a lot of these different agents can serve different purposes. And right now I run one agent because I think it's easier for me from an understanding standpoint. But we're starting to get into the realm where, and Parth, you know this because you have eight Mac minis. But we're getting into the realm of creating different agents for different parts of your life or business. So my agent's obviously very good at social media because that's what I specialize in. So knows a lot of the retention strategies or content creation strategies from a GPT Image 2 and CDance 2.0 standpoint.

11:27point. But I was introduced to GraphRag, and it's basically the democratization of

11:37information that you can basically use to prompt, and these agents can prompt for you. So I think it's starting to build different agents in different spaces with GraphRags that run these autonomous loops to complete the tasks that I might not be so knowledgeable in, but they can do it 100 times better than I can. Right.

11:57Parth Patil:So this is the idea of getting an agent to construct a knowledge graph on a domain before you begin working in that domain, right? Right. Very powerful new technique, especially when we have these Hermes agents every five hours going, oh, by the way, I learned this new thing, and I've saved it. Oh, I've learned this new thing. Oh, let me update my skill. There is something about the speed at which it constantly improves, tries to improve itself, which is something that I haven't seen in, you know, it feels like for like the frontier models, the agents we're talking to kind of stay at the same level, and then when they get a model improvement, then they improve.

12:32Parth Patil:But skills and memory are a huge way to improve before the model even improves itself. It's just, you know, writing down the important information about the tasks that the user wants. So I think the Hermes ecosystem of like rewriting your abilities constantly on the fly as the users' needs change and evolve is very interesting. Will, tell us about a project. So you're used to running a lot of these types of autonomous systems, maybe a little bit more liberally than me, but I always learn a lot every time you figure something out or run yourself over. Can you tell me about a project where you gave an agent autonomy and it paid off?

13:16My favorite project ever built is this autonomous social agent. So it'll pull videos, create videos, it'll edit the videos, it'll add copy and text overlays, all based on the parameters I give it. It'll then, you know, post to a content calendar and post to social autonomously, and then it has API access to read analytics and improve upon that. So while I was trying to improve upon my agent, um i prompted it wrong i guess and uh and this is for the company account right like this is for like work at at a company that i work for um and it posted like four times at 4 a.m and i was like dead asleep but i wake up in the morning i'm like thank god this is on brand because if it wasn't like i wouldn't be going into work the next day um but like it posted four times or however many times it was like by the way not like spaced out hour to hour it was like back to back to back to back and like a four minute period um and then i really like my biggest takeaway was just i need to i need to learn a little bit about setting the the parameters in my initial loop prompts um because then it'll it'll understand the context of what it should and shouldn't do and and maybe i I should give it more structure.

14:35We'll say a little also about, because obviously we're anticipating this wave of agentic participation in social media. I mean, it's already there, but it actually hasn't really picked up to the full tsunami that is very soon. What are some of the positive lessons? What are some of the negative lessons? What do you anticipate happening within kind of the social sphere given, you know, kind of, you know, AI autonomous, you know, social systems and everything else beginning more and more present? I think in part of what you were talking about this last night, there's, I think, a skepticism around AI content right now, which limits the number of templates that you can post on social without a certain amount of feedback.

15:22For instance, I'm a lot safer posting an animated video of a dog than I am AI generating a dog. But that said, like at the root and core of social and read, you probably know this better than anybody for that matter. I think it comes down to an aspect of human psychology of why does somebody watch a video? Why does somebody skip the video? Why does somebody share a piece of content? And I think that, you know, AI makes this a lot easier to understand for the, you know, the social media manager or the, you know, 22 year old that just graduated college and wants to be an influencer. It allows you to figure out what works from a content perspective based on like true performance data.

16:08and then you can even create a graph rag now for human psychology for it to give you very structured recommendations based on the content that you put out and I mean for me I'm just running a fully autonomous social workflow from the whole funnel of social and ultimately

16:30Parth Patil:talk about the entire feedback loop yeah let's do it So basically we have a licensed content library because obviously at the moment, like we're limited to a number of content templates that we can use across social. So it has a list of parameters of what it can and can't pull. For instance, if we're going to edit content and add text overlays to tie back to our company, I don't want text overlays on the video already. I want a certain duration of video. I want a certain like look and feel of the video as well that I know performs better in the industry that I'm in. So based on all these parameters, it'll filter through this licensed content library that has thousands and thousands of videos and images.

17:12It'll create selects and reject others using the dupe logic so it won't ever see the same video twice for efficiency. And then it has a copywriting skill. So this is basically a workflow with a ton of skills that come together to create one very improved loop. So then it'll analyze the video again that it just chose. and it'll say, all right, you know, this dog is doing, you know, this thing in this location, and it'll come up with a text overlay based on the guidance I gave it of what works from, like, a text overlay perspective, feeding it hundreds of arm top performing videos and hundreds of videos from other companies.

17:53And then the caption will always tie back to our company and like a very uh like a nod like it's not in your face as an ad um you know from here it'll then schedule it on this great platform called ordinal which is basically an autonomous content calendar with an api so the api is the game changer because now i can schedule and it'll

18:13Parth Patil:post on its own and this is to every single social platform at the same time right right that's like the power of ordinal i can connect it to linkedin tiktok youtube and i can cross post at the same time. And then from here, I'm connected into the backend through an API key with Instagram's meta, I guess. So I can now read the performance data, which I learned goes very in depth in the API key. Like you can see average watch time, you can see skip rate, which in my opinion are the two most valuable assets for distribution. And then based on that feedback and the content it generated or selected, it'll create like an auto-revisement loop.

18:55So it's constantly A-B testing new formats to see what works, to hit a specific goal I gave it. Yeah.

19:04Parth Patil:And so this is like a longer running goal, like a more autonomous system. What agents are you using to orchestrate this kind of thing? Oh, this is running on cloud because basically this started as me saying I don't want to scroll this licensed content library because it's time-consuming, so I want the agent to do it. And I was like, well, I don't really want to edit either. I want the agent to do it. I was like, well, copywriting takes a lot of time. I want the agent to do it. Scheduling stuff on the calendar is pretty time consuming as well. And then it just turned into this workflow of all these different skills built together, which I suppose I could feed to a Hermes agent and it would do the same thing if I gave it the file.

19:44Parth Patil:So the workflow is encoded in skills and skills are essentially portable across agents anyways. Right. Yeah. Yeah. You can put this anywhere. What's been, like, you know, one of the things that is very cool about everything, not just Discovery Hermes and the other, like, you know, not just Pioneer, but like, you know, bleeding edge Ford Scout. You know, you had the 4 a.m. rogue posting story. But what's been some of the, like, huh, I thought that would work. It doesn't yet. Or, huh, that was really weird or unexpected delight. Like what's some of the things from the scouting? This is a wild project and I don't know why I did it, but I set out to put a Hermes agent on my television.

20:29And this was like three days ago. And my idea was, hey, I don't want to spend the money on a Mac mini, but if my television can connect to Wi-Fi and it can do all of the things that my computer can do to an extent, like why can't I just run an agent on it? And then I called Parth at like 3 a.m. I was like, dude, what do you think about this? He was like, I'll believe it when I see it.

20:53Parth Patil:I was like, I don't know anyone that's put an agent in their TV. And then I'm like, why would you put your agent in the TV? Can't you just control the TV with your agent? There's open questions. And he's like, not interested in that. He's like, I'm just going to put the agent on and let you know when it works. Which, by the way, it worked. Very, very surprise and delight. And I think. I get a message at 5 a.m. He's like, it worked hard. Holy shit. But I think the bigger learning for me was like, if I can do that, then there's no reason we can't put AI in basically anything that can connect to Wi-Fi or Bluetooth, which is incredibly interesting from a hardware perspective.

21:31Because I think like as we shift to AI and everything becomes more about trust, like being able to have like something handheld is going to become like a very valuable asset.

21:41Parth Patil:Do you think that we'll end up in a world where more and more of our appliances all have some sort of agent inside them? I think we're already seeing that now. Well, certainly our podcasts, but yes. Is there an agent in here? I mean, I'm sure there's an agent in here. It's probably like granolaing on the back end. Yes, exactly. That's an interesting point. It goes back to a theme of Reed and I have been talking about. I was like agents listening in on a, on a, on an episode or something. Imagine I have friends that have, have, have been like, they'll be on a sales call and then the agents vibe coding the feature that they're talking about while they're selling the feature.