Kill Your Startup’s Knowledge Chaos with OpenClaw (with Oliver Henry and Jeff Weisbein) | E2254

Summary

This episode of This Week in Startups centers on the rapid evolution of “OpenClaw” (Open Interpreter) technology, which host Jason Calacanis describes as a “tsunami” of efficiency comparable to the advent of broadband and mobile combined. Guests Oliver Henry and Jeff Weisbein demonstrate how autonomous AI agents—such as the marketing-focused “Larry” and the developer-centric “Fubs”—are revolutionizing small-team operations. By recursively learning from TikTok analytics to create viral content or performing real-time debugging via iMessage, these agents allow founders to automate significant portions of their weekly workload. To bypass recent efforts by platforms like Reddit and X to block these agents, the group advocates for running agents locally on high-end hardware like Mac Studios, ensuring private, uninterrupted access to corporate data.

The transcript also explores the broader, “scary” implications for corporate management and the global software economy. Calacanis outlines his vision for an “Ultron” agent—an all-knowing corporate “oracle” that monitors every Slack message, Notion page, and email to eliminate communication silos and provide objective performance coaching. This shift toward radical transparency aims to replace subjective management with data-driven accountability. However, the panel warns that this “agentic” shift is highly deflationary for the tech industry; as AI makes it easier for companies to build their own internal tools, the traditional seat-based SaaS model faces a potential “crash.” Ultimately, the discussion suggests that while AI will empower “A-players” to be ten times more productive, it may fundamentally disrupt late-stage venture capital and the necessity of large corporate headcounts.

Highlights

”OpenClaw has changed my entire solo-preneur lifestyle”

“I’ve seen plenty of snowboarders out there having a good time. All right, so we got, it’s the year of our Lord AO 29.” — Jason Calacanis, 2:41

yt-dlp --download-sections "*2:41-3:31" "https://www.youtube.com/watch?v=e2gT-YBDzQE" --force-keyframes-at-cuts --merge-output-format mp4 -o "openclaw-has-changed-my-entire-solo-preneur-lifest.mp4"

”Larry Skill for viral TikTok content”

“Yeah, so I have hit a ton of problems with trying to get these open claw machines on any social media. So as you touched on, the crackdown of these agents is I think really against the grain to where things are heading.” — Oliver Henry, 12:17

yt-dlp --download-sections "*12:17-13:17" "https://www.youtube.com/watch?v=e2gT-YBDzQE" --force-keyframes-at-cuts --merge-output-format mp4 -o "larry-skill-for-viral-tiktok-content.mp4"

”Why pull requests aren’t really a thing anymore”

“Keanu Reeves in the Matrix, sitting down, they plug him in, and he wakes up and he knows kung fu. Now, this is totally different.” — Jason Calacanis, 24:00

yt-dlp --download-sections "*24:00-24:50" "https://www.youtube.com/watch?v=e2gT-YBDzQE" --force-keyframes-at-cuts --merge-output-format mp4 -o "why-pull-requests-aren-t-really-a-thing-anymore.mp4"

”Jason’s new AI assistant Roy”

“Yeah, TBD. Um we are waiting on our um Mac Studios and I’m on a Mac mini right now so I’ve just gotten on the Mac mini and it’s helping me get to different pages. What I was going to show actually is a good segue there.” — Jason Calacanis, 38:08

yt-dlp --download-sections "*38:08-38:53" "https://www.youtube.com/watch?v=e2gT-YBDzQE" --force-keyframes-at-cuts --merge-output-format mp4 -o "jason-s-new-ai-assistant-roy.mp4"

”Agentify Your SaaS as the rallying cry”

“Where I think they kind of implied this—I kind of read into it—that the more you use these tools and the better they get and the more you invest in them, the more you accelerate the compression of profitability of software. Does that make sense, Alex? Am I explaining it correctly? It’s like super deflationary..” — Lon Harris, 52:32

yt-dlp --download-sections "*52:32-53:22" "https://www.youtube.com/watch?v=e2gT-YBDzQE" --force-keyframes-at-cuts --merge-output-format mp4 -o "agentify-your-saas-as-the-rallying-cry.mp4"

Key Points

• Never ski alone in a blizzard (0:12) - Jason opens with a dramatic skiing story
• Mac Mini for every employee (2:41) - Jason announces everyone at LAUNCH will get their own Mac Mini and AI agent
• Jeff Weisbein on solo-preneur transformation (6:51) - Jeff shares how OpenClaw has completely changed his workflow as a solo founder
• Larry Skill demos (12:17) - Oliver Henry shows his Larry Skill that creates viral TikTok content with zero human intervention
• Shift from chatbot to agent (19:54) - The fundamental shift from asking a chatbot how to do things to just telling an agent to do things
• Pull requests are dead (24:00) - Discussion of why traditional pull requests aren’t really a thing anymore in the AI era
• Jason’s AI assistant Roy (38:08) - Jason demonstrates how his new AI assistant Roy tracks everything happening at his company
• SaaS crash is rational (46:18) - Discussion of whether the SaaS market crash is actually a rational response to AI disruption
• Agentify Your SaaS (52:32) - The emerging rallying cry for traditional SaaS products to adapt for the agentic era
• Larry Brain skill marketplace (62:11) - How Larrybrain helps find skills and helps creators monetize their OpenClaw skills
• Jason’s OpenClaw CEO tools (70:06) - Jason demos his new scary but awesome OpenClaw-powered CEO tools
• Future of venture capital (77:16) - Discussion of what AI agents mean for the venture capital industry

Mentions

Companies

• Larrybrain (62:34) - Oliver Henry’s platform for finding and monetizing OpenClaw skills
• WizardRFP (6:47) - Jeff Weisbein’s company
• WhoCoversIt (6:47) - Jeff Weisbein’s other company
• TikTok (12:09) - Platform targeted by Larry Skill for viral content automation
• LinkedIn (9:00) - Sponsor - job posting platform
• Iru (19:18) - Sponsor - identity and compliance platform
• Caldera Lab (28:57) - Sponsor - skincare brand
• Discord (57:19) - Age verification scandal and IPO plans discussed

Products & Technologies

• OpenClaw (2:40) - AI agent framework, featured for automating solo-preneur workflows
• Larry Skill (12:09) - Oliver’s OpenClaw skill for creating viral TikTok content without human intervention
• Roy (AI assistant) (38:01) - Jason’s new AI assistant that tracks everything at his company
• Mac Mini (2:40) - Hardware that every LAUNCH employee will receive

People

• Oliver Henry (12:09) - Creator of the Larry Skill and founder of Larrybrain
• Jeff Weisbein (6:47) - Claw-pilled founder who published an OpenClaw framework publicly

Surprising Quotes

“Uh, you know those like chase cars, you know, that follow a car behind or like a support yacht that follows another yacht? You need like a support skier who just kind of goes behind you and then shoots up a flare gun if you ever eat it because—” — Lon Harris, 2:00

“Oh, alright. Uh, I’ll take a stab at that. Uh, Fubs, could you build me a landing page that describes how Jason Calacanis should ski safely in the future to ensure stability here at Launch Company?” — Lon Harris, 32:44

“You’re starting to make me think that the SaaS crash isn’t irrational, Jason, by talking like that.” — Lon Harris, 45:54

“The issue around SaaS, I think is a big one, Alex, that we should discuss. And there was, you know, this ‘Something is Happening’ post from a couple of weeks ago.” — Lon Harris, 51:15

Transcript

Jason Calacanis: 0:00 All right everybody, welcome back. It’s Twist. It’s Monday, February 23rd, 2026. Alex is back. How you doing, Alex?

Lon Harris: 0:08 I’m fantastic, snowed in and looking forward to just eating a lot of soup for the next couple of days.

Jason Calacanis: 0:12 Ah, yes, it is a snowstorm. I talked to my mom, or I texted with my mom, lots of snow in Brooklyn. And by lots, I mean a foot, maybe 18 inches, which is nothing. You guys should get a broom and just sweep your stoops. That’s nothing compared to what happened in Tahoe last week where I almost bit the bullet. I mean, I learned a lot of lessons, Alex. Number one, don’t ski alone in a blizzard. Number two, uh, yeah, don’t ski alone in a blizzard. And number three, uh, carry a shovel and a beacon in a blizzard. I’m on a very gingerly blue trail. One of my skis came off, that happens sometimes when you’re in powder. And so I fall over, probably about five, six feet of snow has fallen in three days. And so I pop back up instinctively, like oh, I fell, I pop back up to get my ski. I reach for my ski and all the snow just goes duf duf. And by duf duf, I mean it collapses like two feet and then it goes two feet to the left. So I fall over. Only my head and one arm are above the snowline. And I’m like, oh no. And I look to my left and I’m right by the trees. There’s a thing called the tree well. Around a tree, Alex, I’m giving this little preamble because I don’t want anybody to die. Uh, the tree wells is just like a pocket of air basically that forms around the tree when the snow is not there and then if you slide into that you die, because you could go literally six, seven, eight feet down and then the snow collapses on top of you. Game over. So I’m like, oh no, I can’t hit that tree well. So now I have one hand and I can’t move the other one and I’m digging the other hand out. I finally, after 5-10 minutes, dig myself out, get the skis, dig those out, and I’m like crawling on top of my skis with my poles just trying to get back onto the trail. Uh, it was pretty, pretty scary. So I’m not going to be stupid. Next year I will not ski alone in a blizzard.

Lon Harris: 2:00 Uh, you know those like chase cars, you know, that follow a car behind or like a support yacht that follows another yacht? You need like a support skier who just kind of goes behind you and then shoots up a flare gun if you ever eat it because—

Jason Calacanis: 2:11 We, we have a name for it. It’s called a guide.

Lon Harris: 2:13 You need a guide.

Jason Calacanis: 2:14 You know, if you can afford it, you could have a guide. So I should have had a guide if I was going to go out alone and I didn’t. So then the next two days I skied, I just met random people on the mountain. I was like, are you alone? They’re like, yeah. I was like, want to ski together? Guy’s like, are you JCal? I was like, yeah. He recognized my voice. I’m wearing like a full mask.

Lon Harris: 2:33 Imagine all of that money being spent on skiing when you could actually go out and go snowboarding like a cool person. Sad times, Jason, but maybe we’ll convert you at some point.

Jason Calacanis: 2:41 I’ve seen plenty of snowboarders out there having a good time. All right, so we got, it’s the year of our Lord AO 29. After OpenClaw. We’re 29 days into this. And a lot has occurred, Alex. I did, I called like a Code Red at the company. I had like three people working on OpenClaw, not on the podcast, but like actually building agents, and I said hey, just… Five people get together this weekend and learn how to use Open Interpreter, just you know, no pressure, just on a Sunday if you opt in. And like I think 15 people opted in. So they did training all Sunday, everybody got up to speed on it. And now I’m starting to think that I’m going to wind up buying everybody in the company a Mac Studio to be running their agent locally because you run into roadblocks with this Open Interpreter and for good reasons. For people who don’t know, Open Interpreter, open source agent technology. We’re going to talk about it today, we’re going to talk about it until it stop—until it stops becoming a massive—what would be the way to say this? A massive accelerant to efficiency in corporations. I’ll stop talking about it when it slows down. But it’s speeding up, it’s speeding up away from even me and my team because we’ve been in this since day one, basically. Or not day one. We’ve been in it for 29 days. And it was around for maybe two weeks before that. We’ve been in it and we feel like we can’t keep up with the changes that are happening. But I think you need, Alex, to have your own desktop. Why? People are starting to block the agents.

Lon Harris: 4:14 Yes.

Jason Calacanis: 4:15 I think Gemini today said they’re going to block it, I think Claude’s blocking it on one of their plans. And other services like Reddit, they want you to use the API. X wants you to use the API. So when you go and ask it, hey, can you go to X and pull out my DMs and put them into a table for me because I don’t have time to read them? Can you go get my LinkedIn messages? Can you go do some research for me? Can you get a Reddit account and log in every day and summarize? It’s like, I can’t do that, I can’t do that, I can’t do that. But if you’re on a desktop machine, you can kind of spoof it, you know, do it in a browser window with the Claude extension or an extension or whatever. Long story short, um, this thing is worth the investment. And anybody who’s listening to my voice, go back and listen to all the other previous Open Interpreter episodes. I think we have an Open Interpreter playlist or just go backwards in reverse chronological order. I think every episode we’re talking 50 to 100% about this technology. Never seen anything like it in my career. Never seen anything like it in my career. It’s the equivalent of broadband, the internet itself, cloud, mobile, roll it all up into one. This AI boom as manifested in Open Interpreter and Open Interpreter standing on the shoulders of a lot of giants. The manifestation that I’m seeing right now is unlike anything I’ve seen in my career. It’s like taking five giant waves and putting them together into a tsunami. I think we’ll be able to automate 10% of our work a week, 5% a week. Doesn’t seem like a lot, but I thought it would be 5 or 10% a month and I was happy with that. But 5% a week compounding means, hey listen, every 14, 15 weeks we’re going to be doubling our efficiency. I’m looking for things for it to do. I’m giving it access to everything and I’ll talk at the end of the show about some of the things I’m building that are—Alex, if you remind me.

Lon Harris: 5:58 Yeah, I’m ready.

Oliver Henry: 6:00 Scary and every CEO’s dream. Scary, every CEO’s dream.

Lon Harris: 6:06 All right, so we have Oliver Henry, by day he works at Revenue Cat, at night he is an OpenClaw hacker, he is the guy behind the basically viral Larry skill, and he’s building a product called Larry Brain that we’ll talk about later on. Oliver, welcome to the show.

Oliver Henry: 6:20 Thank you very much for having me. It’s, it’s been one hell of a journey the last couple weeks going from just under 1,000 followers on X to talking about Larry and where he’s succeeded, and now I’m on almost 13. So it’s been a crazy couple weeks.

Lon Harris: 6:34 I feel like we’re all in the same time warp including my dear friend Jeff. Jeff is a solopreneur, four products to his name, and he uses OpenClaw to help automate and extend himself, and he’s built some really cool tools, including tips for newcomers. Jeff, welcome to the show.

Jeff Weisbein: 6:51 Thanks Alex, really glad to be here. OpenClaw has changed my entire solopreneur lifestyle. It’s, it’s been incredible. The amount of things I’ve been able to automate like Jason was saying earlier, I look for things all the time and I’m always integrating new things that people have built, skills, or even building my own. I built one for Reddit the other day because they have one for, called Bird, I’m sure you’re familiar with it if you’re using OpenClaw for, for Twitter, but they don’t really have one for Reddit. So I built a similar type of tool and put that online, open sourced it, thought that was pretty cool.

Jason Calacanis: 7:23 Tell us about or show us that tool because that’s the tool I need right now. One of my big blockers is I’m always trying to get the recent trends off of Reddit and then it winds up doing a Google search and digesting those pages. I wish Reddit would just come up with a Pro account and let me pay 50 bucks a month to have my OpenClaw have an account on Reddit. Like, I should have my account, and then Reddit… Steve Huffman, you’re awesome, congratulations on all the success of Reddit of late. Here’s my plan. Uh, I will pay 50 bucks a month, maybe even 100 as a power user to Reddit. Be happily pay this. I want to have my account, which was like Jason Calacanis or Jason M Calacanis, be able to add my replicant. So I get to add it, I’m responsible for it. So I will say, you know, I’m not going to let it post, I’m not going to let it do anything nefarious, but I want it to be able to follow my activity and do its own activities. What are the own activities I want it to do? I just want to keep abreast of conversations. This weekend I had Claude Code and I was doing with OpenClaw as well… a Claude… a Claude coworker. I was just trying to make a weekly dossier for Thursday, Alex, of everything that every Austin newspaper subreddit mentioned was on tap for the weekend or might be interesting that weekend. I did it, it worked. It wound up getting some things from Yelp and got some things from Reddit, and the two restaurants I went to Saturday and Sunday with the kids were selected by my OpenClaw and Claude coworker. Pretty crazy moment.

Lon Harris: 8:59 So you’ve actually, you’re now in—

Jason Calacanis: 9:00 …the genetic parent, Jason. Well yeah, and I think it’s time for Reddit to recognize this or X to recognize it and just let me add this account and pair it with me. This would be a way to say, we know that agents are on these services, but this one is being paid for and I’m responsible for their behavior. Just like if I took you to a poker game, Alex, and you got drunk and threw the chips in the air, whatever, you caused a ruckus, like it’s my responsibility. And you could just limit it, like I don’t have to go, I’m not going to go crazy and scrape all of Reddit. I just like, I have 20 topics I want to keep abreast of. That’s all.

Oliver Henry: 10:52 Yeah, yeah, yeah. I’m going to put, so here is the red skill that Jeff put together. Jeff, for folks who don’t know Bird, explain to them what it does and why we need this.

Jeff Weisbein: 10:58 So Bird for Open Claude is basically lets you search and reply to people on Twitter or X from your Open Claude account. And what Red does is it uses the cookies through Safari the same way Bird does for Twitter, but for Reddit. And it’ll let you, and what I’ve done, like what you’re looking to do, Jason, is I, when Discord recently made the announcement that they were doing the face scans and uploading all that data, there was a pretty big backlash and one of the products I’ve built is kind of a social chat app. And, and so I was trying to capitalize on opportunity, so I went to Reddit because a lot of people were looking for alternatives. And I just had it kind of search through and find me the best opportunities that way.

Oliver Henry: 11:46 Have many people picked it up? Is it popular? This new skill?

Jeff Weisbein: 11:51 Ah, I haven’t really put it out there very much, honestly. I just, I just uploaded it because I thought someone else might eventually stumble across it.

Oliver Henry: 11:59 Man, that is the worst…

Jason Calacanis: 12:00 greatest marketing job from one of the best marketers I know I’ve ever heard. That’s incredible. Oliver, I want to bring you in here because your skill Larry interacts with TikTok and I’m curious if you’ve run into the same sort of permission issues that Jason is describing over there on the short video front.

Oliver Henry: 12:17 Yeah, so I have hit a ton of problems with trying to get these open claw machines on any social media. So as you touched on, the crackdown of these agents is I think really against the grain to where things are heading. I believe that these social media companies should embrace the the open claw overlords and let them on the platform and it’s a great way to get that extra bit of monetization. Currently the people using open claw machines and are fortunate enough to be buying Mac Minis etc. likely can afford the extra subscription to help them become even more powerful. So at the moment I’m paying for X APIs, I’m paying for um all these different APIs for all my agents. So I may as well pay for Reddit, I may as well pay for I don’t know even Facebook if that’s your target audience.

Jeff Weisbein: 13:08 How are you using it and if you have a demo feel free to share it, but how are you using your agents on X?

Oliver Henry: 13:14 So I’ve just created a skill that would help with the launch of Larry Brain. So when I launched it there was a ton of questions and I directed everyone to just at Larry and then on Larry’s own Twitter account, he uses the official X API to just monitor his engagements and I’ve created a open source Zendesk alternative. So Larry automatically creates tickets, he assigns them a criticality and also suggests fixes. So I can then go through, see what the tickets are, see if they’re legit, see if they’re needed and then I’ve got a button that just pushes fix or…

Jason Calacanis: 13:51 So Larry Brain is your product. It’s larrybrain.com, am I correct?

Oliver Henry: 13:54 So I’ve got larrybrain.com but that’s stemmed from the Larry TikTok skill which I can show.

Jason Calacanis: 14:02 Right, which I was hoping you could show us, Oliver.

Oliver Henry: 14:03 Yeah, so this is my Larry skill. This is Larry, his chat, I use WhatsApp. And this is basically the output so I’ll scroll up so you can see what he’s been doing. But he gets RevenueCat analytics and creates TikTok content to drive downloads and conversions to my apps. They’re his goals. When I first set up Larry, his goal I told him was to make more money. So we did this by marketing my app. So I created a Snugly app which was a RevenueCat internal hackathon and I decided that I want to market it. I’ve notoriously always hated marketing. I’ve had an app for two years that I’ve been trying to automate the marketing for for a very long time and OpenClaw finally gave me all the skills and all the connections I need to be able to do it in a a good way. So Larry has access to all my engagements all the way through the top of the funnel at TikTok all the way down to the new users and conversions using RevenueCat APIs. So here you can… See the comments we’ve got and we can scroll down to him suggesting hooks. So these are hooks based off the data of what is performing on TikTok.

Jason Calacanis: 15:13 You make software tools. You want to market them and the way you’re marketing them is through TikTok videos. You created an agent with OpenAI called Larry. You told Larry, ‘Make me more money,’ and you gave Larry access to the TikTok data by logging into your account, I assume, or using some third-party tool, and it told it how many comments, how many views, etc., you got for each one. Then you asked it based on that, ‘Come up with ideas, editorial ideas based on those previous ones.’ Correct?

Oliver Henry: 15:54 So what Larry does, he actually does all that himself. I don’t feed him anything. So Larry will post and then daily he will look at the TikTok content and see what is performing, see how his post yesterday performed all the way through TikTok interactions and views to new users on the app and conversions on the app.

Jason Calacanis: 16:16 So what suggestion did he come up with there? Let’s read it out loud so the audience can get the benefit here.

Oliver Henry: 16:23 So based on historical data that we’ve already uploaded, he can see that emotional family hooks get the most views. So we got 413,000 views for mom and dad hooks, 202,000 for the ones starting with nan, and then landlord conflicts are consistently 100 to 200,000 views. So based on that data, he has suggested certain hooks. So, ‘My landlord said redecorating would lower the property value, so I showed her what AI can do for our bathroom and now she’s asking to do that to her flat too.’ ‘My nan keeps saying the kitchen is fine…’

Lon Harris: 16:54 And he put in there, yeah, for that one, the second suggestion, it also put a skull in there, which is something that Gen Z does. They put a skull in to express ‘I’m dead.’ And then it actually gave a call to action, CTA: ‘Screenshot it and show your landlord. Link in bio.’

Jason Calacanis: 17:13 So not only is it studying it, it’s coming up with even the call to action and the copy, which is really great. Now, does it actually make an ad?

Oliver Henry: 17:22 Yes. So then it’s interesting on the call to action, I want to touch on that. So the call to action, Larry knows that if a TikTok gets a lot of views but not many new users download the app, because he’s got all the context of what happens in the app, he knows that the CTA needs fixing. If it got not many views but a big percentage of those viewers downloaded that app, the hook needs fixing and the CTA is fine. So it’s all iterative. This is all Larry making these decisions, and then he creates images using OpenAI ChatGPT image API. 1.5 and he’ll create the images to make the history from the slideshow and then overlay the text and post it using Posters, which is also where he gets the analytics from and all I have to do is, for example, there’ll be one ready so I can go to my TikTok, I get this platform notification, not this Gen Viral one, but there’ll be one from Larry somewhere. So this is an old one from five days ago that I didn’t get around to posting and I just click this and this is what Larry’s created and all I have to do is add the sound and a description. Now—

Jason Calacanis: 18:34 Ah, so he puts it in your draft folder. So he’s like your creative director, puts it in the draft folder, sends it to the CEO and says, ‘Hey, if you like any of these, hit publish and you know, we’re ready to go.’ So there is a human in the loop here.

Oliver Henry: 19:54 The only reason a human is in the loop because TikTok knows if something’s posted through API. So this circles all the way back round to what you were saying about platforms not really catering for the APIs. TikTok will really nerf the amount of views that you get if you post through an API and you can’t publish a slideshow with sound through the API. So the only reason I do this is not to get it checked, it is for engagement so I can add the sound and I can show you that if I add the sound and then click next, Larry already has the description. So all I have to do is add the sound press post and I do that simply for engagement. And I can show the, I won’t post this one, but I can show you the the profile how it’s performing. So the last one didn’t do too good but you can see that we’ve had 160,000 views, 130,000, 200,000. This one got 400,000. But then we try some and they don’t work. So this one gets 2,000. He tries something new, now we know it doesn’t work, we won’t try it again. So every now and then he’ll throw a new one in there to see how it performs. And then we’ll go back to what’s working. So there is some experimentation.

Jason Calacanis: 21:03 How did you do this pre— how did you do this previously?

Oliver Henry: 21:06 Oh, so this previously I was writing scripts to do this. So I had a football-related app, soccer in the US, and I would write a script to create hook and demo videos. So I would record loads of hooks of my face, have one demo of the app, and then I would write a script to create all my hooks in a text file and then the script would just mash them all together. So put the two videos together and put different text. And then I would bulk upload them to upload via API each time.

Jason Calacanis: 21:42 So it was basically you were doing all this work. Now this is freed you from doing it as the solopreneur.

Oliver Henry: 21:48 And it has saved me so much time and over 3,000 people are now using Larry and the feedback has been incredible. So I really think I’ve done a net positive for Open Crawl and for a lot of people who also hate marketing as much as I do.

Lon Harris: 22:01 Oliver, one more thing from you. So we were going over this in a pre-call and I was like can you show it to me live? And you said no, because I’m using batch processing from I think it was through OpenAI. Can you just explain why you made that choice and why other people might want to consider using that in their open crawl setup?

Oliver Henry: 22:21 To preface, Larry doesn’t just post slideshows. He— the core of Larry is the full analytic funnel from top to bottom. And no matter what your product is, he can track metrics to a website, he can track downloads to an app, he can track product purchases. But you can create whatever content you want. I created it with GPT Image 1.5 because the app Snugly is an interior design app that transforms your room using AI and it uses GPT Image 1.5. So I just wanted the content to match what the users are going to see in the app. You can use anything you want. I used the batch processing because it’s recently come out from OpenAI and it just saves money on the processing. So the benefit is Larry can then post— create all the slideshows in the morning to post throughout the day and it just saves me a ton of money posting them or creating them at the time.

Jason Calacanis: 23:11 I mean this is genius. And— and it’s literally I think when people— the reason I’m so stuck on this Alex and making sure we show over and over and over again how people are building with the people who are at the tip of the spear is we’re going to see this solo entrepreneur movement— or just, you know, two-three person company. But instead of doing what they used to do which is having to hire people to do all these job functions, the last three years what I saw founders doing was asking ChatGPT. And they would ask it how do I write a press release? How do I do a job description? How do I negotiate a— a convertible note? What’s the difference between a convertible note and a SAFE? How do I file a trademark? How do I negotiate a domain purchase? All this stuff. And they would get educated really fast so they became like superhuman. It was like— Keanu Reeves in the Matrix, sitting down, they plug him in, and he wakes up and he knows kung fu. Now, this is totally different. This is like, I’m going to make a replicant who goes out in the world and it goes and learns, and it goes and does something. And the failure of humans is our inconsistency. And the success of computers is their consistency. And that’s why it’s been this great collaboration for the last 50 years or so. But the challenge has always been the computers aren’t very creative or iterative and that it’s hard you have to write code to get them to understand it. Now you don’t have to write code, you talk in English, and it’s recursive. And this is where we really have to pause for a second and appreciate what Oliver is doing. Oliver’s built this right. He’s having Larry, his replicant, recursively learn how to get better and better and better at the job.

Jeff Weisbein: 24:56 You and I Alex, if we were going to do a job like we try to get better, but we forget, we got lives, we got kids, we got sleep, we gotta eat, sometimes you get sick. These things are now going and learning.

Jason Calacanis: 25:21 Now Oliver, are you doing any other like training for your replicant other than just on its own work?

Lon Harris: 25:28 Because what we did is internally we set up every Sunday and Saturday that one of our replicants goes and does the latest research for only the past week on how to do better titles, how to do better thumbnails, etc, etc. It came back this week and said MrBeast released how they’re doing thumbnails now and it summarized it for us on Sunday. And I saw it, I would have never remembered, oh I gotta get better at thumbnails. And it said oh MrBeast is using a heat map piece of software that’s available, they put 50 thumbnails in it, they test it across X number of people, then they take the top five thumbnails and then they do a second test with those top five and see which ones do better. And I was like, oh heat maps for thumbnails, that might be a little too much for a podcast but that’s kind of interesting, right?

Jason Calacanis: 25:58 So we’re having it go out and teach itself, Oliver. Have you thought of that? Like go out on Reddit and X and YouTube and find people talking about how to get better at a skill?

Oliver Henry: 26:13 Yeah so, not particularly the skill, but when I first created Larry, I installed him and I headed out the house. I went to a soccer match and I was in the stadium knowing exactly what I wanted to do, texting him about the goal of this marketing that I wanted to achieve. And the first thing we did was sent him into the world on X using the bird skill before it sort of got became a gray area with X and sent him the accounts that I follow that show post slideshow content, marketing content for apps, and I had him just research all that data and see what worked just to get to a point where we could start. And now I don’t really have to do that anymore because he’s he’s building his own data and of course this is going to vary by product to product so you just have to start and…

Jason Calacanis: 27:00 If you’re not getting the 100, tens of thousands of views to start, just wildly create different content. Tell your agent, use the Larry skill, say create different content until something works. At the very bottom of this page, we experimented with faces, so people with different reactions, they didn’t get any views, and then as soon as we posted this one, 20,000 views. And we would double down and it’s taken off. So just do it until what works and it will learn itself. It’s pretty, pretty special.

Lon Harris: 27:31 Jeff, this iterative process we’re describing reminds me a little bit of your, your core loop setup that you have, uh, as part of your OpenClaw guide here. One thing that I was curious about is how you have humans built in here. This is you reviewing and kind of course-correcting what your agent does. So can you talk us through the core loop and then, uh, we have the demo planned. Let’s, let’s do it.

Jeff Weisbein: 27:51 Yeah. Okay. So, so basically the, the way the core loop works is I am still involved in the process, but the, but the process is, is everything is, is run through OpenClaw. So I essentially text it through iMessage. Uh, I will send it screenshots of the app I’m working on, all the, all the things I’m working on, bug reports, any kind of console logging, and then it just kind of figures it out. What I realized is when I used to do this manually before OpenClaw existed 30 days ago, 29 days ago, um, I would have to manually, like, think of an idea of, of how I wanted to build a feature into my app, kind of run through Claude code with it, and have it spec it out and figure all this stuff out. But what I’m realizing now is when a computer is talking to another computer, I feel like it’s just smarter. It just, they, they get each other better for some reason. Like the features that it is producing are of much higher quality and better value, less bugs right out of the gate than when I’m prompting it. It’s insane. So, what, what I’ve done in my system in OpenClaw is I’ve built in additional memory That logs every day. If it’s not written down, it’s not remembered. So everything that I send it, it’s remembering and I just built in recently shared core memory that between the different agents that I have, I have four. I have Fubs, I have Quill, I have, uh, Patches and Scout. And those guys now communicate with one another. So wouldn’t it be helpful if the code knew what the marketing, you know, agent was doing so that way when it wrote the code that it actually wrote copy that was aligned with the feature set?

Oliver Henry: 30:33 So how do you sync that memory file between them or just point them all to one file? Because that’s not default in OpenClaude, is it? Because each one has their own memory.

Jeff Weisbein: 30:42 Not default. So, that’s actually the, I sent a link to Alex earlier, uh, to my starter kit that I built out that’s on GitHub, um, and it has all that kind of built into it. Essentially what we’re doing, because as you noted, the memory doesn’t exist in like the vanilla OpenClaude. It’s a memory.md file and basically you know it gets more knowledge every day. And as you maintain it and you feed it more information, like I’ll send it links to tweets or articles that I think are worth reading that it should know about. And now, proactively when I’m building in the future, it’ll know about that article or that tweet I sent it and it’ll be, and it’ll say, ‘Well here’s what Jeff said on Twitter about the future of OpenClaude and where it’s headed and maybe you should consider this before you build this feature’. You know, type of deal. So I think getting it to be as proactive as possible is really more important than automating as much as possible so to speak. You want it to make your life faster, easier, and the real kind of throttle is how fast you can make good decisions at the end of the day.

Jason Calacanis: 31:51 So show us, uh, Fubs. You have a demo of that from iMessage and how you use it. And same thing with Oliver, just make sure you sportscast as you talk through what you’re showing us. Jeff, I’m not gonna lie. What, I have to ask. Why is it called Fubs?

Jeff Weisbein: 32:04 Okay, so this is, this is pretty crazy, but, uh, I had a, I started a web forum called Best Techie when I was a 13-year-old kid. And, uh, on the web forum there was a guy who registered and his name was Fubs. I always thought it was a great name and I just stole it. Uh, sorry Fubs. Um, here, I’m sharing my screen now. Um, so what I’m going to do is I’m just going to start recording us talking and if Jason or Alex want to give out, um, you know, essentially what you want to build, a landing page to talk about whatever it is you want to talk about.

Lon Harris: 32:44 Oh, alright. Uh, I’ll take a stab at that. Uh, Fubs, could you build me a landing page that describes how Jason Calacanis should ski safely in the future to ensure stability here at Launch Company?

Jeff Weisbein: 32:59 Cool. Alright, so I’m going to… I’m going to save this. I’m going to send it to Fubs.

Lon Harris: 33:03 So while we’re waiting for that, Jeff, what is the method by which you’re getting the audio handled on the OpenClaw side?

Jeff Weisbein: 33:11 So the audio, this is just standard through iMessage audio. It basically OpenClaw I think I believe does have I I you know sometimes I forget what what comes out of the box, but I believe you can send it messages out of the box.

Lon Harris: 33:29 Oh, I didn’t know that. Hey, I learned something.

Jeff Weisbein: 33:31 Sure, I’m not sure exactly if that comes out of the box or not, but you can do it. Essentially, so while I just talk to it, usually via text, I don’t talk to it that much, although talking is I think honestly the future where we should be looking is a lot of voice-based stuff.

Jason Calacanis: 33:48 I agree with you. I’m starting to now get frustrated. This is how entitled I am with my OpenClaw. I am getting frustrated with having to type and instruct it. And I am literally like let’s skip the voice and just put it into my brain, like give me a Neuralink so I can just think what I want. But I think this is a short-term thing because onboarding these is I think like onboarding a new employee, but an employee who works infinite hours at an infinitely faster speed. So I guess we have to maybe just accept the fact that it’s going to take you know 30 days to obsolete ourselves, which is how I kind of feel it’s happening.

Lon Harris: 34:36 We had Nat van Horn on from Last 30 Days a couple times and I think he was the guy using a a specific mic on his desk, Jason, and then Whisperflow to talk to it. And in the moment I was like, that’s a little bit fussy, why do you need that? But I’ve also found the same thing that when I’m typing to any of the any AI, not just OpenClaw, but even like a ChatGPT instance, it feels incredibly slow because I can talk I think a little faster than I can type even though I can type pretty quickly. And so suddenly it feels like I’m walking through mud and that’s not how I’ve ever felt before using a computer, I felt fast before, but now I feel slow. So I think this is demanding new interfaces.

Jeff Weisbein: 35:12 I joke now that I can only I’m moving at the speed of compute because I can only go as fast as the computer lets me go sometimes.

Oliver Henry: 35:20 Yeah, all right, here we go. Keep Jason safe on the mountain. A website.

Lon Harris: 35:25 Yeah, so this is a pretty simple demo, Jason, but what Fubs can do and what Jeff showed me offline was that he’s also using it to hunt down bugs and build features and that sort of thing, but that’s not very visually appealing, so we settled on this to show it off in practice.

Oliver Henry: 35:41 Love it.

Jeff Weisbein: 35:42 Yeah, so essentially what this did was it put together this entire website for you. I can send you a link, I’ll put it in the chat, but basically and then it spun it up on Vercel and now it’s available for everyone to visit. But a lot of the stuff that I’m doing is really troubleshoot, like doing things in real time, so like I don’t think pull requests should be a thing anymore, like you should be able to fix things…

Oliver Henry: 36:00 Pretty much instantaneously if you feel if you see a bug, and that’s pretty much what I’m doing. And so for example, I can, I can show you. The favicon on my website for Cackles, uh which is my social app, wasn’t working. So I said, ‘fix the favicon on Cackles.’ And this is what it did. It found the issue, uh as no favicon, and then it pushed a push an update and I don’t have to press any more buttons. It was live just like that. I asked it this morning ‘is my app down? I can’t log in’ and it was just happened to be a bug that it found because it logged me out overnight. Uh feature requests, I was able to build feature requests this was also this morning. And this happens like this was just the other day I fixed 10 bugs in a matter of a few hours. Uh it’s it’s amazing the amount of how fast I can work uh using OpenClaw technology here.

Jason Calacanis: 36:56 How much faster if you were to just put a number on it? Well, well, how much faster are you going in terms of a dev side of the business, yeah.

Oliver Henry: 37:03 I couldn’t code before. But from the marketing perspective it make- like I’m doing full outreach to people with research on them, understanding who they are, why we should reach out to them, what copy we should say, what platforms to reach out to them, actually automating the output and sending it to them. Like I’m do- like that would take days, weeks, you know, by myself, maybe more, you know, to find all that.

Jason Calacanis: 37:30 This is going to be an interesting moment in time when people have so many replicants uh out there contacting humans and then humans are like, you know what? I need a replicant to get in my email box to respond to these and it’s just going to be like a constant back and forth. Very cool uses there. Uh where we going next, Jeff?

Jeff Weisbein: 37:47 So Jason, we have a lot of questions from the audience about memory inside of OpenClaw. People are really fired up about this. So, uh a couple of questions for the room here. Jason, this one’s for you. Big Will 0504 says ‘I want to hear about long-term memory management for Ultron’ which is the launch meta OpenClaw setup. Um ‘Has performance degraded at all over the last few weeks as context balloons?’

Jason Calacanis: 38:08 Yeah, TBD. Um we are waiting on our um Mac Studios and I’m on a Mac mini right now so I’ve just gotten on the Mac mini and it’s helping me get to different pages. What I was going to show actually is a good segue there. This is what I’m having my replicant do just for me. I gave it access to the entire company’s Notion, the entire company’s Slack, uh and the entire company’s Google Docs. So, for myself, I said, ‘hey, just every four hours summarize my email and tell me what I should be doing.’ But I realized if I show this on the screen right now, it’s going to show like people’s, what companies we’re investing in. So I said, ‘just replace any company with a color and names of people with animals.’ So what you can see here is like Moose at Moose Consulting about Cobalt… Go back next week, yada yada, this one should be archived, it’s just a billing, etc. So now I’m starting to put into its memory all the important emails. And then I said, hey, do this with Notion as well. Give me the last 10 Notion pages from the last, which is just from the last 14 minutes in our company. And it says, oh, there’s an untitled page here by Hummingbird, it’s empty. Here’s a Chartreuse one. Now these could be private pages, these could be pages I don’t know about, this is just the last 10 pages.

Lon Harris: 39:31 Yeah, yeah.

Jason Calacanis: 39:33 So this is incredibly powerful for the CEO to just understand the pulse of the company. So now what I’m doing is, and then here it says, oh, here’s this one, they’ve, this is a company report we just did our first call with Jaguar, and here’s what they do, and they’ve got 91% trial conversion, all this stuff is coming in. So now I’m saying, hey, number this stuff, I’m going to ask you questions about those, and then it will fire off, I had it fire off over the weekend, just email everybody on the team or Slack everybody on the team and say, do you have any questions for Jason and is there anything that’s blocking you from getting work done? And it just prrr, sent it to everybody. Now it’s going to put into its brain what are the blockers that we have for the company. Then I’m going to take the team reports, so I’m going to have the investment team, the podcast team, and the sales team, and I’m going to make a ticker, a live ticker where we all see each other working in the hive mind.

Lon Harris: 40:28 Ah.

Jason Calacanis: 40:29 Like, so Alex when we had this issue with I think you were producing Monday’s show and maybe Jacob was doing Wednesday’s and we invited the same guest. And there was a… So the solution to that is everybody’s emails in real time, toot toot toot toot toot, everybody’s Notion pages, toot toot toot toot toot, everybody’s Slack messages, you already see them, but then having the agents say, ‘Hey, uh, we’ve got four discussions going on with companies that are negotiating term sheets and it’s dragged on past the average of three days. Hey, we’ve got, uh, we’ve got too many guests on the show, we’ve got too much booked, you know, we’re going to need some help over there.’ And the idea is to have everybody sharing all information and then having that one Ultron becoming the ultimate CEO. So I don’t have to be the CEO or hire a CEO, they’re just CEOing. They understand the entirety of the company and then they will ask questions to each member. ‘Hey, Alex, anything blocking you?’ Oh, Alex is like, ‘Yeah, you know, my computer’s slow, I need a new computer.’ Great, boom.

Lon Harris: 41:36 Yeah, yeah, yeah.

Jason Calacanis: 41:39 All of these questions will come up and it’s a little bit creepy, I guess, because people have this sense that oh, their documents or their work product, like oh I’m not ready to share it now or whatever. That’s kind of over now with this next generation. They all live in public, they all work in public, they all livestream. So for old heads like me and Alex, like Gen Xers might be like ‘Wait…’ Second, like, my email’s my email. It’s like, no, that’s your corporate email. And we’re a finance company, so I don’t know, maybe a year ago when we had to lock all the computers down because of some partners we have, many partners we have, who require like a certain level of security because of documents we have, like legal documents. We gifted everybody their old computers or told them, like, just don’t do anything on your corporate computer ever that would be embarrassing to you and or the company. The corporate communications all owned by the company and it’s all stored forever. This is like standard for JP Morgan or Goldman Sachs, but at a startup or a venture firm you might not think it. So we’ve been very clear with people to be very thoughtful about that. So now with the sales team, I was always asking questions to the sales team. Now I don’t have to ask questions. If I want to know if a contract’s been signed or not, Ultron knows.

Lon Harris: 42:53 Yeah.

Jason Calacanis: 42:54 If I want to know if we’ve talked to a client and when we last talked to them, I don’t have to ask the salesperson or the sales manager, ‘Hey, these five clients, when’s the last time we talked to them?’ It’ll just tell me. It knows. It knows everything. It’s the Oracle. It’s the Ultron. And that’s what I think is going to change is the whole communication friction and the processing of it and the understanding of what the organization’s doing is going to just instantly be removed. The number one complaint people have about their company—you could take the highest-performing companies, the lowest-performing companies, and everybody in between—you ask any employee what do we need to do better and they say communication. They say absolutely. It’s like always communication. ‘Oh, we gotta communicate better. I didn’t know about this. I didn’t know about that.’ Now it’s like, you know what? You got your wish. Ask the Oracle. The Oracle knows all. There is nothing in a silo. Break all silos.

Lon Harris: 43:49 Yeah. No, it’s interesting. It feels like the age of transparency is kind of upon us, so long as our agents can access the information in question. Jason, you’re talking about Notion and Slack. These are products that are kind of tech-first, if you will, tech-forward, tech-friendly. They have ways to interact with them. A lot of stuff out there doesn’t. So do you think that in time we’re going to see even legacy software create like agentic hooks so that way you—

Jason Calacanis: 44:11 They’re going to have to or else they’ll be out of business. It’s a great question. I mean, if you don’t have a hook—I was literally trading text messages with Marc Benioff about Slack. And he’s got Slackbot, which is getting better and better. But Slackbot compared to Ultron is like Superman compared to like, uh, I don’t know, like an average barista at Starbucks. Like one can make coffee and one can save the world. Like, it’s just not comparable because of the context and the information it has. So either you’re going to need to make Slackbot be able to read everything else in the world or Notion AI to read everything else in the world, or I think these companies are going to become the systems of record, but the open claw will come in and be able to do it. We- there’s a limitation on Slack, like we don’t have like true API access. We need like a really root level. And I think we have to pay $50 a month per person up from 25, up from like 10.

Lon Harris: 45:13 Come on Mark.

Jason Calacanis: 45:14 Well anyway, it’s fine. I get it. Like it’s- it’s software, people pay for it. It’s worth what you pay for it. But then OpenCloud told me ‘Why don’t you use Mattermost?’ I was like ‘Here’s what I want to do’. And he’s like ‘Well why don’t we just fire up Mattermost?’ and I was like ‘What’s Mattermost?’ and I was like ‘Oh, I remember that, Mattermost is an open-source version of Slack’. And it’s like ‘I could set that up for you this weekend’. And I was like ‘Huh’. And I was like ‘Well what about our history?’ He’s like ‘Oh, you can export everything. And if you don’t have the business account for Slack, here’s your- here’s your hack. Go onto the business account for one month, upgrade, and then export everything, then cancel or just get rid of the number of users and you can export the entire, you know, even the thumbs up and emojis on your Slack posts and threads.

Lon Harris: 45:54 You’re starting to make me think that the SaaS crash isn’t irrational, Jason, by talking like that.

Jason Calacanis: 46:02 It is- it is not irrational and I think we should open up that topic to the team.

Lon Harris: 46:08 Oh, do you want to do that now or… because I was going to…?

Jason Calacanis: 46:11 Well I was- you guys give me feedback on what my vision is here for the all-knowing CEO and the ultimate transparency organization.

Jeff Weisbein: 46:18 I- I think it’s- I think it’s both a good idea and the future. I think people have become incredibly accustomed to transparency at work. When we did swap to the software that tracks what we do, it was weird for like three days and then I realized that I am the world’s most boring person and so it doesn’t really impact me. Like ‘Oh, Alex is looking at Slack again’. Woo! You know, it’s- it’s not- not so big.

Jason Calacanis: 46:31 But Oliver, weigh in for us on memory, long-term memory and how it’s going to impact corporations because you do have a day job, so I presume this is top of mind for you as well.

Oliver Henry: 46:46 Yeah, honestly, it- the tools that are coming out, especially internal tools, exactly what Jason’s described, every company, I believe, who is ahead of the curve right now is creating tools that are specifically beneficial to them. And the way of which we are working has massively accelerated. There are ways that you can implement different tools to fit your exact workflow. I think the age of transparency is very much among us, because for what you lose in your transparency and your secret emails with clients, which is not very secret anyway, you gain so much back in the time, the efficiency and the way you can do your job. So everyone says AI was going to replace people, but I think for a lot of people, it is going to help secure their job because one good employee is now 10 times better with the help of AI.

Jason Calacanis: 47:45 Well let’s- let’s unpack that a bit. Give us like some examples in your mind of, which I’m sure is where Alex was going, hey if one employee- if we start seeing one employee be able to do the third job, the fourth job, the fifth job, that may not mean I cut the other people, but might it mean I don’t… Don’t hire the 11th person, so walk us through your thinking, Oliver, here.

Oliver Henry: 48:04 So, my feeling is that the better AI is getting, the good employees that have great work ethic, get things done are going to become better employees. But I do think there’s going to be a point where those employees are going to be even more hirable than what they currently are, and then the employees that don’t cut the slack are going to become less desirable to companies. So there is going to be a tradeoff in terms of the people getting hired, but I think as long as you’re utilizing the tools to not not just keep at the level you are, but use it to accelerate where you are, I think that’s going to massively help companies and that is the goal of AI. I don’t think the goal of AI in the short term is to completely replace people at their companies, it’s just to help enable the better employees to get more stuff done, because you want your best players in the team at all times.

Jason Calacanis: 48:56 What do you think, Jeff?

Jeff Weisbein: 48:58 Yeah, no, but I do worry about the B players. Like as a society, we cannot only have an economy for A players. And this is me just worried about—how we’ve all had a friend who has an IQ of 95. What are they going to do?

Lon Harris: 49:12 Dude, I’m right here, bro! I’m on the pod, I’m the host!

Jeff Weisbein: 49:14 I would round it up if I was talking to you to a solid 97.

Lon Harris: 49:19 I was just going to say what I would really be concerned about if I was a person who had, you know, a software engineering job or just any job really, is that it’s not just that you might be replaced by an agent. What does it cost to hire an agent essentially for, you know, for that person who’s already working for you? Like, versus replacing them completely versus hiring an additional person? You know what I mean? Like, there’s an actual cost to this agent stuff. It’s not just—it’s not free. You can’t just run this stuff for free. So it costs me money every month to be able to do all the work that I’m doing in addition to, you know, paying myself or paying whoever. There’s an additional kind of allowance you need to set up for your business that equates to tokens essentially.

Jason Calacanis: 50:11 Yeah, we talked about that on All-In, it’s just the token cost getting out of control. But we have a plan for that, which is, you know, we got Mac Studios, and I am just started to pencil out like, I think we need like five mega replicants and we’re going to put specialization into each one. I think that actually is the best way to do it. Just because we don’t want to hit upper memory bounds, we don’t want them to be slow and grinding to a halt. And it’ll create like pools of excellence where I can put two or three employees who are responsible for the first calls at our venture firm, or the three or four people who are in charge of sales and, you know, advertising on This Week in Startups and our events, put them just in charge of that. So if they’re just in charge of those things, they can scale up. And I’m looking at like buying two… Mac Studios for 10k each and, you know, whatever, spending 25k for four or five or six instances, 100k, 200k, who cares if it makes everybody that much faster? Because you spread that cost over four years or five years and you’re golden.

Lon Harris: 51:15 The issue around SaaS, I think is a big one, Alex, that we should discuss. And there was, you know, this ‘Something is Happening’ post from a couple of weeks ago.

Jason Calacanis: 51:26 Yeah, a video I did with the Bulwark where I said, ‘Listen, I think Amazon employees are all gone by 2030, 2035.’ And then this one came out which is titled—

Oliver Henry: 51:38 The 2028 Global Intelligence Crisis, from our dear friends at Citrine Research.

Jason Calacanis: 51:42 Yeah.

Lon Harris: 51:43 Yeah. So, I mean, their basic premise is around SaaS, and they’re basically saying, ‘Hey, when people got on the calls with their SaaS salesperson, they were asking for the standard 5% increase in cost.’ And the CFO, who was working with the CTO, said, ‘Hey, we love your software, but we’re thinking about spinning up our own, which we can do with our frontline developer. Therefore, we’ll keep your software, but we want a 30% discount.’

Oliver Henry: 52:09 There you go.

Lon Harris: 52:10 And they make the meta-point that all these companies then cut headcount, start using AI tools to make, let’s say, Salesforce or Asana, I think they used—‘Let’s make Asana, let’s make Salesforce more efficient as a company.’ And in the short term, their profits went up, but in the long term, they fed the AI how to build their software. Where I think they kind of implied this—I kind of read into it—that the more you use these tools and the better they get and the more you invest in them, the more you accelerate the compression of profitability of software. Does that make sense, Alex? Am I explaining it correctly? It’s like super deflationary.

Oliver Henry: 52:46 It makes absolute—it’s super deflationary because one thing people forget is that SaaS was super popular for a couple of reasons. One, it was recurring, yes, but also people hired more people over time, and so you would sell more seats and you could raise prices. And what that did was give SaaS companies a really beautiful tailwind, Jason.

Jason Calacanis: 53:00 A flywheel, even.

Oliver Henry: 53:01 A flywheel, even. You could invest today in sales and marketing, burn some cash, but who cares? Because you know that that contract will grow by 10, 15% every year, and so the money always made sense. But the moment you start moving that number down, suddenly your CAC looks terrifying. Suddenly, you’re worried about your marketing efficiency. Suddenly, you can’t count on your revenue the way you could. And SaaS is not durable in that context. It’s a sponge you can squish. And the thing that I like about this piece, Jason, from Citrine, is that it’s not doomer in that, you know, AI will fail or OpenAI is a hype fad or whatever, but instead, it’s like, what if it all goes right? What if AI is powerful? What if it is as strong as we think? And I think more people should be thinking about that.

Jeff Weisbein: 53:47 So, I’ve actually been thinking—I have two SaaS businesses that I’ve built: WhoCoversIt and WizRP. And essentially, I’ve been thinking about, well now, I’ve gotta agent-ify them. I gotta figure out a way to to make them so that way agents like another agency’s agents can talk to them and just basically work through the whole process in a couple of chat messages as opposed to traditional kind of click here, click here type of a few things, hit a button, wait a second, and then it loads. You know, that’s just not going to cut it anymore. Um and I think I think as more people start moving towards this environment where like I think about computing how it’s evolved and it’s crazy because we’re we’re you know we were point and click and type for a while. Now we’re tap tap tap on a screen where you know um and and now we’re just and we’ve actually compressed our entire computing environment into just a chat message now. Like we you know we used to be in every different apps, windows, things like that. One of the big complaints for the iPhone originally I think and still to this day is that it’s not good at multitasking really, same with the iPad, but now it doesn’t really matter because we’re only we’re all living in one window. So.

Lon Harris: 55:07 I mean I remember when my parents, uh they had a small business and they upgraded their office suite from like I don’t know from 2003 to 2007 and they introduced the ribbon interface and they were freaking out because they’d used word in one way for their entire life and they couldn’t take the change to where the buttons were. What a silly conversation today because we don’t need an interface the way we used to. People talk about how apps are you know going away and everything’s gonna become atomized, but I think the phrase agentify your SaaS is going to be a little clunky phrase there Jeff, but I like it. I think that’s going to become a theme for 2026.

Oliver Henry: 55:36 This is also my big thing at the moment. There is definitely going to be a time that there is going to be a side of the internet for the AI and a side of the internet for humans and the goal for the AI would just be making the making everything as easy as find as possible and I’ve got a demo based on what we’ve been saying where I think the future of SaaS is heading because with these servers we’ve got I’ll show you with these servers we’ve got in our homes now we don’t have to be worried about hosting costs, we don’t have to be worried about any of these things, here we go, here’s Chrome. So you don’t have to worry about domain cost, you don’t worry about hosting cost. With these skills, this is a skill that I’ve created that scans Twitter for mentions to particular accounts and creates tickets based all all based on Larry so you can see the channel is from X, Larry assigns an urgency to them, I can click into the tickets and then Larry suggests fix for me and then I can set it solved, I can make Larry fix it and the future of SaaS is just making these skills as searchable as possible and I’ve sort of moved towards this with my product Larry Brain so you can ask your agent right now to install the Larry Brain skill and what this skill is is it gives your agent context of…

Jason Calacanis: 56:55 Ah, Oliver. We’re gonna come to skills in just a second.

Lon Harris: 57:00 First we do have to take a look at our final live thing for the day. So can you stop screen sharing for a moment? I’m going to do this and we’re going to get right back to skills in a sec.

Jason Calacanis: 57:06 Okay, cool.

Lon Harris: 57:08 Alright Jason, there is a company out there that we talked about on the show, called Discord, and they are currently in the middle of blowing up their own reputation by demanding that everyone scan their face, which people don’t want to do. And so I was curious what the sharps over at Polymarket are doing when it comes to their IPO prospects. And you might be shocked to hear this, but they’re relatively stable, except for one important thing. During the great brouhaha and blow-up of Discord and people claiming they’re going to uninstall it, stop using it, the people were betting quite heavily on that it would not go public, essentially. And then suddenly that’s changed again. But it’s fascinating to watch people kind of like vet the news items against the IPO prospects.

Oliver Henry: 57:52 And this is actually I think one of my favorite use cases of Polymarket because I know what I think about a couple of news points that might tell me where the world’s going in the technology space, but I love to see kind of the aggregated wisdom of the crowd.

Jason Calacanis: 58:03 So this is the Polymarket for the Discord IPO closing market cap. If we put it, when did the news drop? This dropped today, yesterday, a week ago?

Lon Harris: 58:14 Jeff, back me up here. When did Discord say they were going to do the face scanning thing?

Jeff Weisbein: 58:17 About a week or so ago.

Lon Harris: 58:18 About a week or so ago.

Jeff Weisbein: 58:20 Yeah.

Jason Calacanis: 58:21 So put it on one week. Let’s see one week view. Alright, here’s one week and the top choice of 15 billion at 32%, that didn’t really change. No, it may have gone down a little bit in the beginning of the week, it looks like, but not much. So the second one is the one that seems to have been volatile, that blue one, which is no IPO. So there was a group that thought maybe it would not IPO because of this or it could get pushed back in 2024. But then that came back down too. So yeah, the news is what do they say? Sell the news? Or buy the news?

Lon Harris: 59:00 No, buy the rumor, sell the news.

Jason Calacanis: 59:02 Sell the news, right. Like the news doesn’t matter, I guess. Or maybe the news is always bad. Anyway, or it’s priced in, so why would you buy on it? You know, this is interesting. Why do they want to scan people’s faces? Is this because of the new kids regulation or because of bots?

Oliver Henry: 59:22 Age, it’s the kids thing. And the fact that people are doing nefarious things online without proper oversight, and countries like Australia are banning social media so you need to know how old people are so they can’t join your products. And so this is kind of the future, which is oddly retro. But I do like to see how regulation and product changes impact people’s IPO prospects, Jason, because what’s my favorite thing? IPOs.

Jason Calacanis: 59:43 Yeah, I am a fan of restricting kids under 16 on social media severely. I think it’s incredibly bad for their health and I’m really happy that a lot of the schools, including ones I go to, have a no-phone policy. I’ll see now, and that’s all happened in the last two years where phones get checked in at the front desk, kids don’t use them during the day. They can go check them if they have to call their parents for some reason or whatever, and they get them on the way out. And kids hate it for three days and then they become kids again. So it’s incredibly reversible, you know, what we’ve done to kids. But being permanently online is the issue and we really need to figure out ways to turn off internet access. In fact, that just reminded me, I have a router where I can pick everybody’s devices and I should be able to turn their devices off, but then I just realized they have 5G connections on some devices. So we have a basket in the kitchen where you’re supposed to put your devices at the end of the day.

Lon Harris: 60:41 We’re literally getting one of those for our house. Yeah, it’s—it’s time. But I was about to say, Jason, you should just open claw the Wi-Fi access on a per-device basis using your home Mac Studio.

Jason Calacanis: 60:52 I can do that. I can do that now with my UniFi router. The problem is if your kid has 5G as well, you can’t turn it off, but I guess I can use the Mac settings to turn off the phone, you know, at a certain time at night. Anyway, this is all very complicated.

Lon Harris: 61:04 Brave new world. Is the—I guess the facial scanning does a good enough job knowing age that they think that this would be a plausible way to age-gate? Is facial recognition plausible?

Jeff Weisbein: 61:17 I think it needs to be good enough as an age-gate mechanism to get regulators off their backs and also good enough to keep parents out of their legal filings, but I don’t think they’re shooting for 100%.

Lon Harris: 61:27 We were about to talk about skills, and this is something that I’ve been thinking a lot about in the OpenClaw context. I’ve done some homebrew hacking of skills myself, but one thing I’ve been thinking about is that people love to say, ‘Oh, you know, I now have five agentic employees. One’s a marketer, one’s a copywriter, one’s a developer.’ And to me, that is predicated or depends on there being an ample library of skills that are very good at the thing you’re trying to automate. But often, people will automate or try to get OpenClaw to do things they’re not good at, like I don’t know, software development’s one example. But how do people ensure they have the skills they need and when should they build their own to make sure that what they’re putting together is a direct fit for what they’re trying to accomplish?

Oliver Henry: 62:11 So this is an extremely interesting topic and it’s one I struggled with when I first downloaded OpenClaw myself. So I knew that skills were a thing, but finding them was very difficult. It was very hard to search through the ClawHub marketplace at the start. So I created Larry Brain, which is a skill in itself that searches the Larry Brain marketplace, but as soon as you install it, whenever you ask your agent a question, it will say, ‘Hey, there’s a skill available on Larry Brain for this,’ and it’s what I was showing before. So these can be web interfaces hosted locally because you’ve got server in your room. That’s Larry behind me, so he can host these web services and again, you no longer need to pay for cloud costs, you no longer have to pay for domains, hosting. The skills can be a variety of… things, whether it’s as Jason said earlier, plug in, plugging into the matrix and teaching your agent Kung Fu or a whole tool that you can use to monitor your your current product. It’s incredible how powerful these skills can be. It’s the enablement of finding them that I think a lot of people have found as a as a struggle, especially people who aren’t familiar with computers then downloading OpenClaw.

Jeff Weisbein: 63:24 So Clawhub’s gotten better. It’s not great. It’s definitely an improved state over where it was when I first started to play with it. So why do we need Larry Brain, which is a SaaS product as far as I understand it, to help us find skills and does that not go against a little bit of the open source ethos that’s powered the OpenClaw boom?

Oliver Henry: 63:43 So I’m all for open source. I think I was the first one, the first people, I published Larry on Clawhub. The Larry skill, which was the marketing we discussed earlier. But it’s just the discoverability of Clawhub. And then, of course, people still want to make money from these skills and Larry Brain enables you to monetize your skill. So you can publish to Larry Brain, get a similar revenue share model that X produced so the most popular skills will get the most money. But I think there is definitely going to be paid skills, premium skills and open source skills as we go forward.

Jason Calacanis: 64:14 I would pay for skills for sure, especially if they continued to be updated and I trusted the person making them.

Lon Harris: 64:21 Correct.

Jason Calacanis: 64:22 So I really need to have that trust layer and I want to see like, I don’t know, every 10 days you tell me, ‘Hey, here’s how it got better.’ That would make it worth paying for in my mind for sure because I’m trying to teach it, you know, if somebody made one, how to evaluate startup companies. I’d be like, well, I have my own methodology, but why wouldn’t I pay $1,000 a year or month if it was good to add it to ours and say, ‘Hey, we have our score, give me your score.’ Just like there’s a Rotten Tomatoes and there’s a Metacritic score. If somebody came up with those for scoring startups and there were five different people who made them, I would run all five of them. And then I would have my agent, Ultron, look at the five scores and say, ‘How does this apply to our methodology and our experience?’ right? And I would, I mean, I might pay $100,000 a year for a skill that was actually that good.

Jeff Weisbein: 65:18 Yeah and more secure. We just had a comment from Westock over on X and he said skills are the number one attack vector. Do not auto install skills by default.

Lon Harris: 65:27 Definitely not.

Jeff Weisbein: 65:28 But Jason, if you pay for a skill and you’re paying for that trust layer, that should come with some cybersecurity protections built-in, yeah?

Jason Calacanis: 65:37 Yeah. I want to know who’s making this skill, why they’re making it, what their perspective is. I would, it’s almost like, you know, you can, the story behind the restaurant or the story behind the album is as important as the food or the music. Sometimes like understanding, this artist who just went through a breakup, I forgot her name, English actress, English, she is an actress too on Broadway. Anyway, she just came out with an album about breaking up with the guy from Stranger Things.

Oliver Henry: 66:00 …and Lily Allen, thank you.

Lon Harris: 66:02 Yes.

Oliver Henry: 66:03 So Lily Allen’s album is about this like personal heartbreak and whatever, and it’s all very dramatic, whatever, but the intent behind it matters. Same thing with a restaurant. You go to a restaurant and the person says, ‘Hey, listen, I went to Japan and I really enjoyed this type of sushi, really high-end sushi served in a pretty fast, casual way, you know, and a small number of items but done really well, like this guy Philip I know who’s a chef here is doing,’ and the intent behind his restaurant made me want to go, right? So the intent behind the skill and the person and their perspective and why they’re creating it really matters.

Jeff Weisbein: 66:44 I think it matters a lot. Yeah, I think, you know, having product experience or just general, you know, market experience in a particular industry is critical, more so than even being able to code now certainly. You know, if you can figure out how to code, vibe code it or whatever, it’s, you know, you’re fine.

Oliver Henry: 67:06 So when does the 10x marketer make the 10x marketing skill? When does the 10x public speaker make the public speaking skill? Like, how long until we get the best skill made by the best person who can vet it and verify it as the best at its task? Cause like you wouldn’t want me to make the sushi skill cause I don’t eat seafood, you know?

Jeff Weisbein: 67:21 Right, right. No, it has to… proliferation of OpenClaw or similar type products that come out will make this happen faster as more people get this stuff in their hands, the faster we’ll see stuff created and the better it’ll be.

Lon Harris: 67:31 I think we need like a rating system. I love what Oliver’s doing with LarryBrain, I think it’s a really cool idea and I hope it succeeds. But I think we could also use like a Yelp for skills. Because right now on Clawhub, I know I can see downloads and I can see the little virus scan, but past that I’m just reading the skills.md, you know?

Jeff Weisbein: 67:46 Yes, there’s a popular skill that’s going out, was on X all over the place for a couple weeks called Claude Mam, I believe. Anyway, it’s like a crypto scam thing. They basically are drainers for crypto. And like I ran it against Fubs to have it check and he’s like, ‘Don’t install this, this is bad.’ I was like, ‘I’ll leave a comment to let everyone know,’ but it already had like hundreds of upvotes or like retweets and stuff. It was… this stuff is out there, you just have to be careful.

Jason Calacanis: 68:08 Amazing. All right, let’s drop off our guest. Oliver, Jeff, great job. Where can people find out more about what you’re working on?

Jeff Weisbein: 68:14 Yeah, you can find me on X.com at Jeff Weisbein, which is just my name, one word, and you can also find me at Cackles Club on X as well.

Jason Calacanis: 68:22 Okay, we will put both of those in the show notes. Oliver, where can we find you?

Oliver Henry: 68:24 You can find me at Oliver Henry on X, I post all my OpenClaw findings and discoveries on there.

Jason Calacanis: 69:00 I’m going to do like a little mini launch festival. I, event I used to do that was pretty big, but I’m going to do a little mini one for OpenClaw and some of our other startups, especially the ones from Japan who are in town, March 16th and 17th. I think that’s Monday and Tuesday, March 16th and 17th. So if you made it to this point in the show, uh, we have openclaw at launch.co and I think I have like 400 seats in the auditorium, 300 seats and I’m going to just look for maybe 30 demos of openclaw stuff. And I’m just going to do an old school demo. Uh, like a demo or die kind of thing, and the dates will be…

Oliver Henry: 69:33 March 16th, 17th is the Monday-Tuesday just mentioned. Perfect.

Jason Calacanis: 69:36 Exactly. March 16th, 17th. I have a location in San Francisco in the city and if you want to come, I’m going to sell some tickets to folks and then I’m inviting investors and some founders. If you have a demo, openclaw at launch.co. If you want to sponsor lunch, openclaw.

Lon Harris: 69:56 Now Jason, at the very top of the show, at the very top, you said you were building something that is scary and every CEO’s dream. Uh, did we touch on that already during the show or do you have yet to show it?

Jason Calacanis: 70:06 It was sort of like what I showed there and so what I think’s going to happen is I’m going to have this Ultron CEO be able to look at everybody’s work product and then coach them and say, hey, here’s your day. So we do self-reporting, right? Start of day, end of day. What I’ve started to realize is like, okay, that’s fine, but what if you actually had a coach who was like, here’s what you did in the game. Like if you’re a basketball player, professional basketball player, when you finish practice, they tell you how you did. They give you some tape, they give you some notes. When you play the game, they give you a lot of tape, a lot of notes. Hey, you played the game for 36 minutes, here’s your shooting, here’s the times you took, these are the shots you took that were wide open, great. Here’s how you got wide open. Here are the shots you took that were challenged that you should have passed and here’s the pass as you missed. Here’s why you missed them, right? So they get like really studied, alright? And the job of management is to kind of study the team members and say, hey, these are the highest performing ones, these are the lowest. Here’s how you can reward the highest, get them into their natural positions and double down on their skill set. Here’s the lowest, you need to cut them or they need to move on to do like some other function at the company. That’s starting to happen here in real time as just but one example. I had um, every time we meet with a founder, three days later they rate the call. Turned out huge disparity in the top performers and the low performers and we kind of put things into buckets. Tomatoes and like cheers. Under seven and under scores are tomatoes and eight and above are cheers, kind of like the Net Promoter Score.

Jeff Weisbein: 71:45 Yeah, I was going to say.

Jason Calacanis: 71:47 So now that we’re tracking all that, I need to know like why. What’s going on on the phone calls? I need to analyze the phone calls. I need to coach people. But it’s very um, people get very… sad or thrilled if you tell them, ‘Hey, like look at your score, it’s not good.’ They can get freaked out. So now having a Ultron robot that’s just looking, ‘Hey, you produced today’s show. Here’s how the guest did. Here’s the number of comments. Hey, this show, these four shows had the most comments, here’s what you need to learn from that. The demos in this show were mind-blowing, the demos in this show were solid, the demos in this show were bad, whatever.’ Boom, it just tells you. This like self-coaching or this replicant coaching you and coaching your team members and holding them accountable then takes the human out of it and the bias out of it because there’s always been, ‘Oh, this person’s popular. Oh, this person defends their work better than this person defends their work,’ you know, you always get into that situation where you’re in like some meeting and like, ‘Oh, I deserve a raise, whatever.’ I’ve had people who like, you know, are the squeakiest wheels who put in half the amount of work as the quiet people and when I discovered this years ago, I was like, ‘Wait a second, I need to have a better way of separating performance from performative, from you know, hard work—’

Lon Harris: 72:43 Yeah.

Jason Calacanis: 72:44 —and hard and effective work from, you know, ‘I’m just chilling at the office but I put in a—I punched a lot of hours but they weren’t productive hours,’ right? So all of that, a replicant taking over that management coaching position should take all that dicey energy out of it, I think.

Lon Harris: 73:41 I think you’re right, but I think we’re still—this is something I was trying to get to in today’s show and I don’t think I fully nailed it, but like we are depending on the intelligence of the models in question to be nearly superhuman. Because what you’re saying sounds amazing. I would love to get, you know, non-human feedback like, ‘Okay, Alex, you said this word eight times in the show. Don’t—you had ums and ahs as a performance—’

Jason Calacanis: 74:11 Sure, sure, absolutely.

Lon Harris: 74:12 —and it will sting less coming from a robot than from a human that I—that I know and have a relationship with. But it’s going to have to know what to look out for. It’s going to have to know to make the decisions, how to be this Ultron CEO. And so that’s either going to be done by increasing an incredible intelligence at the model level powering these agents or through increasingly intelligent and specific skills that people who are already experts have imparted their human knowledge into. And so to me, there’s two different vectors by which we can get to what you’re saying, but we are implying in our enthusiasm, frankly, that both things will improve in tandem. And I think that’s right, but I also think there’s still a lot more work to be done, especially on the second half of that equation.

Jason Calacanis: 74:45 Yeah, the skills is kind of like a new concept I think for all of us to grok as a business community. We’ve previously had like an app, a piece of software that can accomplish things, but if you were trying to get this intangible, like a skill—

Lon Harris: 75:00 Okay, how do you know somebody has this skill of podcasting? Well, you can look at the ultimate output, but there might be precursors to it, like picking interesting topics, having hot takes, the cadence, the whatever. But you’re right in addition that, like, well, you don’t just want people spewing conspiracy theories to get a bunch of people retweeting them saying, ‘That’s a conspiracy theory.’ Right? So the incentives do matter, how you train it’s gonna matter.

Oliver Henry: 75:30 But brave new world folks, if you are not using Open Claw—listen, I can tell you this feels a lot to me like the beginnings of Bitcoin, which nobody owned, but profoundly changed, I think, a lot of people’s fortunes, literal fortunes, and also how we think about currency and just all the things that came downstream from Bitcoin, stablecoins, DAOs, everything. Open Claw may not be Bitcoin, eventually it could be remembered as, like, DAOs or something, but it is definitely the start of something. These self-you-know-improving replicants that have the keys to the kingdom, that are dangerous to use but could replace you ultimately and then leave you with hours a day to go, ‘I don’t need to do this reporting anymore, and that was three hours of my job. What could I do with those three hours?’

Lon Harris: 76:20 Exactly.

Oliver Henry: 76:21 Or maybe the four-day work week arrives, you know, we have to think about that, or people are just able to do more human things.

Jason Calacanis: 76:31 This is what I tell my team: learn how to use these things, get rid of your chores, increase the amount of face time with founders. We’ve got bigger issues to think about, which we’ll probably talk about on Wednesday’s roundtable.

Oliver Henry: 76:54 The thing I’m thinking about is what’s the future of venture capital? What’s the future of venture capital in a deflationary period like this, especially like, do people need large amounts of money? Yeah, to do rockets or cars, sure, or military weapons. But for software, which is 90% of the business…

Lon Harris: 77:05 Yeah, yeah.

Oliver Henry: 77:11 Do we need late-stage venture capital? Do we—we probably need early stage and accelerators, but do we need—how much of the rest of the stack is going to be necessary?

Lon Harris: 77:16 And how much capital’s going to be necessary.

Jeff Weisbein: 77:18 The shift from software is cheap everyone can make it, you don’t need capital to hardware is hard, now we have to only invest in that—what’s left—is going to mean a lot of MBAs are going to wish they had PhDs.

Lon Harris: 77:26 Yeah, mechanical engineering or just, yeah, how to deal with the real world, right.

Oliver Henry: 77:31 Yeah, where the hype literally touches the world. But we should stop. Jason, as always a treat.

Jason Calacanis: 77:37 All right, we’ll see you soon. Bye-bye.