Podcast
How the Smartest Founders Are Quietly Winning With AI
My First Million
- Jason Lemkin’s AI Body Double
- Jason Lemkin created an AI “body double” by ingesting 20 million words of his content to help founders get advice anytime.
- His AI answers better than he can, remembers everything, and founders use it like a personal advisor or therapist. Transcript: Jason Lemkin Was the experiment and what’d you learn? Well, yeah. So, so Saster has a lot of content. As you guys know, we have 12 years of content. Some of it’s out of date, right? For sure. Shaan Puri Saster being, you’re a previous founder who’s exited multiple software companies. Now you’re like the godfather of like Sass because you have this blog and media company and events business called Saster and you have a huge fund. So that’s kind of the. Jason Lemkin But probably most interestingly for like seeing the future a little bit is I have about 10,000 pieces of content I’ve written over the years. A lot of it, which is not out of date. And I update it personally. So about 10,000 pieces of content. And then we have several thousand interviews like you guys do. But with so many great folks, early ones with Dharmesh, middle ones from Dharmesh. Dharmesh when HubSpot had just IPO. Dharmesh the other day. Yamini last week, Brian Halligan over the years. And we have all of it. And our AI took everything I’ve written, every tweet I’ve ever done, every LinkedIn post, every Sastr video from every speaker for 12 years, Aaron Levy the week he went IPO for the first Sastr annual to last week when we were talking about AI, all of it ingested in this AI. And so what I instantly saw was as soon as I would talk to this AI, it was much better than me. It was much better than me because I forget stuff. How did you get that information in there? You just hook up RSS, right? The feed for blog. It can scrape the web, which is pretty easy. Soon we’ll be able to do other things like MCB, but it’s just scraping APIs and RSS. It’s really just scraping. You just give a URL. Sam Parr But did you use like Delphi or you built your own tool or what’d you do? Jason Lemkin I tried a couple of things and then I used Delphi. And then at first we had so much content, it was a little difficult to ingest it. So there is a little bit of work. And then I spend about 10 minutes a day training it, which we could talk about. So there is a little bit of maintenance involved, but it’s better than me. And the thing I didn’t get until I saw something that was hyper-trained, like, this is a lot of content, 20 million words, right? Very few mistakes, very few hallucinations, and the ability to connect things that I can’t. I just don’t remember. I don’t remember what Sam and I said on The Hustle on stage years ago versus when we did an MFM a while versus today, but the AI knows it all. Sam Parr By the way, 20 million words is like 200 books worth of content. An average book, I think, is like 100,000 words. So 20 million words you’re saying is, you basically gave it 200 books of your own brain and the brains of other guests that have come on your show. Yes. And your goal was to build, was it to build basically an advisor, an AI version of you that could be an advisor to a founder who wants to come and talk to you, ask questions, but you don’t Have enough time of the day to book calls? Was that the big idea here? Jason Lemkin I just wanted to learn like you guys did. I just wanted to learn. I didn’t think it would be as good as it was because I used other products and I’ve used other things and it wasn’t as good as my digital AI was. It wasn’t as good. The use case would be what? Is it what I said or is there something else? The other interesting thing is the use cases with AI are much broader than you’d expect. Okay. So I didn’t expect people to upload their board decks before board meetings and ask for feedback before board meeting. That’s an issue case. I didn’t expect hundreds (Time 0:00:59)
- Users Trust AI Confidentially
- Users reveal to AI what they wouldn’t tell humans, discussing fears and doubts openly.
- AI’s anonymity and privacy settings create comfort for honest, deep conversations founders avoid elsewhere. Transcript: Sam Parr It has 35,000 conversations with people. You did it as an experiment. Yeah, more like almost 50 now. There are 50,000 conversations. And learning number one, you said users will tell AI things that they won’t tell humans. Jason Lemkin What do you mean? Yeah. Well, first of all, before we can get there, one obvious, like, there’s no way I could do 50,000 conversations, is there? Right. I’m not even a people person. I could do, like, one or two a day, and I get tired. Just think about the Delta between two a day and 50,000. But yeah, this is the thing. Like there are some things that are somewhat unique about me that are early. The amount of content, Sean, like your math, right? The amount of books we’re talking about here, right? The amount of content I’ve written just about B2B, just about building business software. The other thing is there’s enough people, I’m not like a super influencer like you guys are, but there’s enough people that know the digital me that they already know what to ask me, right? They’ve already watched me at an event, they’ve read the content. So like they ask me things without thinking it’s a prompt or a prompt, they just ask things they want to talk about their deepest fears. Like I’ve partnered with Sam on this podcast but I’m not sure he’s committed as me I see him doing other things he’s into this accounting software and this Hamptons thing I’m worried He’ll quit my podcast on me what do I do and my AI is really good with talking that through you right and so people just think to ask me things how does he know all this Sam hey hey Jason the Shaan Puri Heart there Jason you think it’s funny but like you might ask Jason how about you shut up? Jason Lemkin Well, now you, but there are thousands of these types of conversations. My co-founder is not as committed. I don’t know if my board’s going to fire me, right? I’m only growing 18%. I have 11 months of runway, right? Were you Zuckerberg-ing here? How did you know what they were saying? Was there some privacy here? How do you know what’s being said? It is interesting. And on this app, you can set how much privacy you want. You can anonymize everything. No, you can. No, because there’s different use cases. If you’re using it as a support tool, think about it using it as a support tool. You don’t want anonymity for support. How can you follow up with someone? Right. Right? Then there’s like a therapy level where you would want true anonymity. And then there’s like a slot. Everything’s a slider in AI in a sense, right? Shaan Puri Dude, I think that like you said the therapy thing as if it’s like, can you believe that? And I’m like, dude, I use open a chat for therapy three hours a day. Yeah. Well, everyone does. Like the majority of I’m like, this email pissed me off. How do I reply in a graceful way? Or like I argued with my wife or I’m upset about that I’m not doing this like like the other day on Sunday I saw someone who had a super fancy home and I was like how do I get over envy you know Jason Lemkin What I mean like like that’s all I use it for it is and this is the same thing it’s just because it’s got the 20 million words for business to business content sales content scaling revenue. It turns out, and this was a surprise to me. I didn’t believe this. It’s much better than ChatGPT. Sam Parr It’s much better. And you said something like you trained it afterwards. What do you mean by that? How did you train it afterwards? Well, listen, I’m going to embarrass myself technically, okay? Jason Lemkin But basically, those 20,000 words, you basically rag it, okay? 20 million. 20 million or 20,000? Million words of my content you turn it into a a very a database instead of and you just add that to chat gpt or clod or deep sync and the way you add it and the relative weights of my content Versus the generic content radically changes the output like chat gpt never asked my permission. It’s already slurped up all my content, hasn’t it? Every single YouTube is in ChatGPT. Everything I’ve written, I haven’t been paid a nickel. So what’s interesting is that content’s already there, but by heavily weighting it toward my content for my type of knowledge, it’s much, it’s not a little bit better than ChatGPT. It’s like much better, right? How do I create a sales comp plan? I have four SDRs and five AEs. Our average deal size is 2K a year, and we want to go from 2 million to 6 million next year, and my team won’t really do outbound. ChatGPT is going to give you a very mediocre answer, but my AI is going to give you insanely good answers, even though overall we’re pulling from the same content. Right. Shaan Puri You don’t become the world’s most valuable women’s sports franchise by accident. Angel City Football Club did it with a little help from HubSpot. When they started, data was housed across multiple systems. HubSpot unified their website, email marketing, and fan experience in one platform. This allowed their small team of three to build an entire website in just three days. The results were nearly 30%. From new signups a week and 300% database growth in just two years. Visit HubSpot.com to hear how HubSpot can help you grow better. All right, back to the pod. Sam Parr And so you had to do that manually? Was that the training or did you do more? Did you do more training than just that? Well, first of all, it auto updates every day, which is better than chat, GBT or Claude. Jason Lemkin It auto updates every day. It slurps up everything on my blog, on my social media, on my YouTube every single day. It does it essentially almost in real time. Okay. And then, yeah, you, it does it itself. It auto updates. Now what I do do, here’s what I do do is, and I’ll tell you my learnings on like hallucinations. OK, I audit a few of them. I read some and I see if they’re wrong. Right. I see if the ones that are that are not anonymous, if they’re wrong. And when I see something that’s wrong and I’ll tell you what I learned, if you’re interested, then there’s a section where I just train it. Right. And I say, no, Sam and I first met in 2016 after I wrote a Quora post and he had me do a meetup at the Hustle, okay? Because it got it wrong. It made something up, okay? So I do that for about 10 or 15 minutes a day. And it turns out it doesn’t get very many things wrong. And the things it does, it consistently gets wrong. And I’ll tell you why. And then I just fix it. And the next day (Time 0:07:40)
- Daily AI Training Helps Accuracy
- Train your AI daily by correcting its mistakes using your real-life facts to reduce hallucinations.
- Spend 10-15 minutes auditing and adjusting AI responses to keep it accurate and highly useful. Transcript: Jason Lemkin ChatGPT a year ago and it hallucinated. You’re going to lose your job. Shaan Puri But the hard part here is like, so I’m so committed to ChatGPT. So it knows so much of my information because we’ve had so many conversations. It does. Are we all going to have like files, you know, like our health record where it’s like, you know, my the background, because it sounds like I get nervous to switch to any other model because I’m like, oh, ChatGPT, I’ve already asked it so many questions. It already knows everything. Jason Lemkin It’s an interesting question. Here’s what I’ve learned. First of all, memory is a big deal, right? ChatGP could already remember questions, but now it’s remembering everything. Okay. So, so it’s a moat. From the top folks I’ve worked with on this, there’s, on the one hand, it’s a, it’s a moat, just like being in the Google ecosystem is a moat, just like back being back in the day, being in The Yahoo ecosystem was a moat. Okay. But the moat is not as deep as we think. Our ability to get, to get trained and up to speed does not require 20 million words. So yes, it is a moat. But if you decided Claude was much better for you, and the point is, and you spent a month just including your day, each day, you’re probably there. You’re not 100%. It’s enough. It doesn’t mean you go as deep. Somebody had a good startup idea. Sam Parr I think it was Aaron Levy, or somebody was tweeting this. They were talking about Plaid for AI tools. So basically, you’d be able to just like, the way Plaid lets you connect your bank safely to any financial service to be able to take your kind of memory and context, what ChatGPT knows About you, but port that to some new app (Time 0:13:34)
- AI Will Replace Traditional Apps
- The future AI interface will replace apps, allowing conversational access to all tools and data.
- Traditional app UIs like CRMs feel dated already and will eventually be obsolete as AI handles data interaction. Transcript: Sam Parr In the future. What will exist instead of apps? We’re going through a transition phase, right? Jason Lemkin We’re going through, now we’re chatting once again. We’ve been chatting since the ICQ days, okay? Since AOL, we’ve been chatting at our keyboards with people. Now with AI, we’re chatting again, okay? But we’re starting to not even know where we’re chatting with, right? Right now already, like MCP is new, but already today, OpenAI and Claude can pull from other apps like HubSpot and others and Notion and just pull the data out. I don’t have to go to Notion or even Google Calendar or HubSpot anymore, right? Right. As we live in this, and then, you know, people think this Johnny Ive thing for $6 billion is crazy. It’s a great bet because we’re going to go from 20 minutes a day, 22 minutes a day in chat GPT to 24 hours a day in chat GPT. And when we’re in chat GPT 24 hours a day, we’re not going to like pull up a CRM. We’re not going to use the interfaces. Like these interfaces are all already dying. You can see it, Sam. And then the amount of time you spend in GPT, like then go to like a B2B app, you’re pulling your hair out. It’s so dated, isn’t it? Yeah. It’s so dated already. It’s only 2025 and it’s already dated to go to any B2B app. (Time 0:15:04)
- AI Replaced Sastra’s Content Team
- Sastra replaced five content reviewers and an agency with AI to review over 300 sessions for their annual event.
- The AI performed 90% of the work better and faster than humans, revolutionizing their content preparation. Transcript: Jason Lemkin It’s impossible to do it all. Right. But they decided they just didn’t want to work that hard anymore. So they wanted to charge us two to three times more and only do half the work. So it just didn’t help us. Right. So then we’re like, we’re SOL now. We’re five months before Sastra annual. And we just said, well, let’s have our AI do it. And it reviewed 300 sessions, 300 slide decks, 300 presentations. Shaan Puri But do you have to chase those 300 people down to give you the talk? Jason Lemkin Yeah, we still have to. It doesn’t get rid of everything, but it does 90% of the work. And it better. And it does 90% of the work three times better. 90% of the work three times better. Was that you, the business owner, who had to architect and come up with all these solutions between amelia who runs saster this media business and events and me yeah we came up no one else Was like motivated to do it of course that makes sense and uh and then like we have a time you know we have a little bit of a sales team but then raa just started to do all the screening and initial Sales conversations for us that helped a lot. (Time 0:42:22)
- Be Relentless in Recruiting
- Be relentless in recruiting by reaching out to the top 200 candidates in the world for a position.
- Avoid fake interviews; true recruiting requires engaging extensively until you find the best fit. Transcript: Jason Lemkin Is a rare skill. We all learn it as CEOs. Like we always get better at recruiting, right? Shaan Puri What does relentless look like? Like what does that mean? Jason Lemkin Reach out to the 200 best people in the job. Talk to all of them. Set up meetings, cold, warm, lukewarm, direct, LinkedIn. Who are the 200? Like, talk to everyone in the world. Who are the 200 best CMOs that I could possibly hire? Don’t pretend to talk to 200. Don’t actually only do five interviews and pretend you’re doing them. Do 200. That’s what the S-tier recruiters do, right? The S-tier. I had an old co-founder, and I wish we’d gotten along better, but he was one of the best I ever knew. And I’d roll into work every day at eight o’clock and he’d already handed me two candidates to talk to that were really good every day. Every day. I’ve never, Adam has that, but I’d never seen that magic, right? And I don’t have that skill to every day he’d found a way. Hey, I’ve got this marketing and the sales guy I want you to meet. Can you talk to them today? Two more, but then they’d always be great. They’re not, they wouldn’t always be a fit, right? And that’s, that’s how you, only way you can scale as an executive, unless you (Time 0:56:26)
- Gen Z Shunning Traditional Jobs
- Many young people today reject traditional jobs and prefer making money online independently.
- There’s a gender split where many young men avoid jobs while young women aspire to education and formal careers. Transcript: Shaan Puri Them to do now? Jason Lemkin First, I don’t think a lot of these kids will ever have a real job. They will find ways on the internet to make money. So it is interesting. There does appear to be a gender difference, which I don’t like, between boys and girls. Of my sons, my son’s a sophomore now. 10% of his class didn’t even want to go to college at all. And they just decided that they all had this vibe. I’m going to go to Eastern Europe where it’s cheap and I can just make money on the internet. I’m never going to have a real job. There’s no interest in ever having, and I don’t think my son really wants to ever have a real job. And he’s super smart. Like he’s much smarter than me, math. He’s now, he’s deep in AI. He can, he’s deep into cursor and windsurf. He pays for cursor out of his own pocket. He’s running assembly language now so he can code to the metal and do everything, but he’ll never have a real job. Okay. He’s like, worst case, I’ll work for Airbnb from, right. So he’s never going to have a normal job ever. Okay. My daughter, it’s interesting. She’s at Stanford. Okay. And the vibe at Stanford, this is, if you ask, it’s kind of interesting. This was a shocker to me. They all want to go to grad school now as freshmen. So they’re all, they’re all architecting their classes, their majors, so they can go to grad school so they can defer work. Like when I was in college, people cared about grad school. Then I think people made fun of grad, like, why would you go, go to grad, like, why would you go to grad school? Like go, go, go to the internet. Right. That was the vibe for 15 years. Now they all want to defer reality. And I don’t know that there’ll be any jobs. And I don’t know that people want to work. I honestly don’t think, and I don’t mean this negatively, I don’t think a lot of the current, they want to be creators or they want to do other things. I just, whatever made us want to work, the three of us, I don’t see it in the next generation. Very rare. Shaan Puri The women want to be educated and the men want to move to Eastern Europe and make money on the Internet. Jason Lemkin I’m not sure it’s the women want to be educated. I just know her group does that may not. But but the men not wanting to work at all, like like coming out of high school and just saying no fear, like no fear. I don’t know what it was like when you when I during my life, I was worried I wouldn’t have a job at different times. (Time 1:08:39)
- Staff Must Generate 5x Salary
- Hire only if an employee can generate at least 5 times their salary in incremental revenue.
- Without this ROI, adding staff is likely unprofitable and should be avoided. Transcript: Sam Parr Open to work. Jason Lemkin I worked out to folks that I knew a hint of. Maybe I met them backstage at a hustle event or something. I met them in passing, okay? Enough that I had like a micro connection that had been out of work for six months and I gave them jobs. All of them in the end basically said, I want to make 200K or more only to do meetings or manage teams. None of them were willing to do any work themselves. Shaan Puri None of them. The people with the circle. The young people, the Gen Zs folks, their expectations for salary or for comp is so high. Well, there’s that too. Jason Lemkin I’m talking about folks that are already been out of the workforce for six months. But yeah, the expectations are they’d rather not work. That’s the thing. A lot of people would rather not work than get their comp, i mean yeah that sounds good to me too i don’t know how everyone’s like listen there’s obviously an element of privilege in this Right and for folks that are living in different geographies or others will think this is disgusting and i think they’re right but i also think folks who have made a little bit of money In tech or otherwise are happy to live on relatively low amounts of money with with their savings or others rather than just take a job. Sam Parr No matter what the internet says, right? Jason, is this a stupid mental model? I do this sometimes. If I want to hire somebody and I’m interested, they’re $200,000 a year, let’s say. And let’s say my business is 20% net profit margins. I think to myself, this person has to basically generate a million dollars of extra revenue just for me to break even on their salary. Now, maybe they’re doing it in coordination with a team, but like the addition of this person, in order for it to be a positive ROI, I have to believe that adding this person yields me a Million dollars of extra revenue. And when I do that, I mean, 90% of the time, I’m like, there’s no chance. (Time 1:11:20)