AI Tinkerers - "One-Shot"
00:04:11 · Luke Harries, Joe Heitzeberg
[00:04:11] S0: Yeah. Yeah. When I was 19. [00:04:12] S1: Well before ElevenLabs. Okay. And then you also did, so, again, this Renaissance background with lots of interest, but also united by building and tinkering. You did, you were a YC founder? You were [00:04:25] S0: Yeah. [00:04:25] S1: Head of growth or product at PostHog? Yes. Okay. And that's one where it's like, I think PostHog's story is partly like every Y Combinator company just used them. That was [00:04:34] S0: like Yeah. Was was moving on from my previous company. Yeah. I phoned up Dalton, our YC group partner, and I was like, Dalton, what should I do? I'm working out my next move. He's like, postdocs in London, they're doing pretty cool stuff. You should chat with them. [00:04:49] S1: There you go. The power of community. [00:04:50] S0: Exactly. You know, jump in [00:04:52] S1: with a community where great things are happening. You can always jump ship or do something or combine forces. So you don't know what's gonna take off. No. Right? But you're in the middle of it. That's really cool. Let's talk about the tinkering. I wanna see something like... I know you've done a bunch of stuff that blew up pretty recently on Twitter with MTP. [00:05:07] S0: Yes. [00:05:08] S1: You know, again, it became popular to kinda talk about MTP, and you were like, oh, I need to go build something. So what did you build? That's
00:05:08 · Joe Heitzeberg, Luke Harries
[00:05:08] S1: You know, again, it became popular to kinda talk about MTP, and you were like, oh, I need to go build something. So what did you build? That's [00:05:14] S0: that's Yeah. Well, a little bit of the backstory. Yeah. I I did a tweet. The the funky thing I think about the current wave of AI is there's so much innovation on, like, the business and b to b side. Mhmm. But the consumer side is terrible. And I think it's just because there's companies like Meta and Google who own the consumer interfaces, who are fantastic at AI, but just, like, seemingly aren't shipping the product. So I did this tweet, which is like, I think you can basically solve consumer AI by connecting a AI agent to your WhatsApp, your calendar, and your Gmail. And if you do that, you basically have the entire context on someone's life as well as actually being able to execute actions because you could create group chats, you could send emails, you have all their invoices. So I tweeted this being like, why has no one done it? And I thought about it a little more, and I was like, well, the hard bit's actually the WhatsApp bit. So, like, how do you connect WhatsApp to AI? So that's why I was like, okay. I wonder if I can actually do this. [00:06:13] S1: What what makes that hard? [00:06:14] S0: But the hard bit is WhatsApp has business APIs, but doesn't have personal APIs. And so just being able to query your messages is actually surprisingly difficult to send messages. [00:06:26] S1: Interesting. They don't have just a basic API for, like No. Search
00:06:26 · Joe Heitzeberg, Luke Harries
[00:06:26] S1: Interesting. They don't have just a basic API for, like No. Search [00:06:29] S0: No. It's very like walled garden. Interesting. [00:06:31] S1: So how did you get in? [00:06:32] S0: Yeah. So I went down a couple of different routes, like, basically, you know, did the long tail search of GitHub repos powered by by Deep Research. And the first time I tried, there's this protocol called Matrix Okay. And which tries to abstract all the different messaging apps into one. It's like very cool open source project. Yeah. Beeper was built on top of it. Okay. And Yeah. I tried to get that running and build on top of that abstraction, but I'm not sure if it was me or the docs, but that was a disaster. Okay. And and this was all on a Saturday, by the way. So I did, like, our trend matrix. And then I was like, well, okay. What's actually underneath matrix? Oh, there's this cool Go library, which is written Okay. And called what's meow. Uh-huh. And I and was like, oh, like, I wonder if that works. Yeah. And so I pulled up Cursor, start hacking away. I've never written go in my life. Right. And yeah, just like with Cursor, managed to get the fur... My first goal was like, can I get a local copy of my WhatsApp messages on my computer so I can do all the querying? Mhmm. And so we wrote... I say we, me and cursor, me and Claude, point 3.5. Yeah. Wrote this thing so when it connects... The library connects to the WhatsApp device API. Mhmm. So it pretends to be the browser client. And actually, when you start the client, the first thing it does is load all your messages.
00:06:32 · Luke Harries, Joe Heitzeberg
[00:06:32] S0: Yeah. So I went down a couple of different routes, like, basically, you know, did the long tail search of GitHub repos powered by by Deep Research. And the first time I tried, there's this protocol called Matrix Okay. And which tries to abstract all the different messaging apps into one. It's like very cool open source project. Yeah. Beeper was built on top of it. Okay. And Yeah. I tried to get that running and build on top of that abstraction, but I'm not sure if it was me or the docs, but that was a disaster. Okay. And and this was all on a Saturday, by the way. So I did, like, our trend matrix. And then I was like, well, okay. What's actually underneath matrix? Oh, there's this cool Go library, which is written Okay. And called what's meow. Uh-huh. And I and was like, oh, like, I wonder if that works. Yeah. And so I pulled up Cursor, start hacking away. I've never written go in my life. Right. And yeah, just like with Cursor, managed to get the fur... My first goal was like, can I get a local copy of my WhatsApp messages on my computer so I can do all the querying? Mhmm. And so we wrote... I say we, me and cursor, me and Claude, point 3.5. Yeah. Wrote this thing so when it connects... The library connects to the WhatsApp device API. Mhmm. So it pretends to be the browser client. And actually, when you start the client, the first thing it does is load all your messages. [00:07:57] S1: Mhmm. [00:07:58] S0: So we wrote a hook in order to just save those into the SQLite database. [00:08:02] S1: Yeah. Yeah.
00:08:02 · Joe Heitzeberg, Luke Harries
[00:08:02] S1: Yeah. Yeah. [00:08:03] S0: And and I was like, okay, great. I've done that, but I don't like Go and I don't I don't really know Go that well. So then I was like, okay. Let me spin up like a simple Python MTP client for that. Yeah. This was when MTP became the rage and I was like, okay. Maybe this could actually be a really cool interface to Yeah. Query So then the Python and the Go library are just connecting to And the SQL and then the next thing I did was expose the ability to send messages. [00:08:32] S1: Mhmm. [00:08:32] S0: So it actually has a little local host server where they go in the Python library [00:08:37] S1: Yeah. [00:08:37] S0: Library chat. Yeah. And and I built this, like, basically on the Saturday. Mhmm. Showed my girlfriend on the Sunday or my now wife. Mhmm. Showed my wife and funnily enough, my mom. I And was like, oh, look. I can send a message. That's pretty cool. And my wife was like, nice. Like, but what are you gonna do with it? Right. I was like, oh, maybe, you know, I've learned a few things about launching products. Yeah. Maybe I should actually launch it. Yeah. Maybe a brief side note for all the Tinkerers out there is like, I actually find launching your own products so much harder than launching your company products. Because when it's your own product, it's like your ego. You wrote this. You're putting yourself out there. [00:09:13] S1: Yeah. But I
00:09:13 · Joe Heitzeberg, Luke Harries
[00:09:13] S1: Yeah. But I [00:09:14] S0: was like, oh, well, just like lock in for an hour, made a quick video, wrote a tweet thread, and launched it. And, yeah, I I think quite early on, we got a retweet from Gary Tan. [00:09:24] S1: Nice. [00:09:25] S0: And that was the first initial spike. Nice. And then it like plateaued... It like very quickly got thousand, 2,000 stars, and then plateaued. [00:09:32] S1: Yeah. So I [00:09:33] S0: was like, oh, that's pretty cool. Like, what other channels could I share it on? So I wrote a hacking news post, and then that took it like past 3,000 stars. Cool. And and then there's been a few things after that. But the biggest thing was like, I genuinely didn't think anyone else would use it. Yeah. But kind of the learning from myself. And do you know Simon Willison? Yeah. [00:09:50] S1: Of course. [00:09:51] S0: Yes. He has fantastic blog. And one of his pieces was always like, you owe it to yourself for every project you do to just write about it. So I saw it as like, okay. Let's launch it, get it out there. And it's been really cool to see all [00:10:03] S1: these different [00:10:04] S0: people using it. [00:10:05] S1: Well, I have a lot of questions about your motivations going in and then what has actually happened since with the usage. [00:10:09] S0: Yes.