A personal assistant emerged from a simple messaging hack

Transcript

From the transcript TBPN publishes with the episode.

1. 3:05:24

   I I don't know. You know, I I wake up every day. I'm like, okay. What do I wanna work on now? What would be cool? And then they was like, okay. I I wanna check with my computer on WhatsApp because because if my agent's not run is running and then I go to the kitchen, I wanna check up on them. Or, like, I wanna, like, do little prompts. Yeah. So I I just hack together some WhatsApp integration that literally receives a message, calls Cloud Code, and then returns what Cloud Code returns.

2. 3:05:55

   One shot. Yeah. And it took, like, one hour, and it worked. And I was like, oh, okay. That's kinda cool. Yeah. But I usually use prompts, like a little text and an image. Because images are like they often give you so much context, and you don't have to type so much. Mhmm. So I feel like this is, like, one of the hacks where you can prompt faster, just, like, make a screenshot so that agents are really good at figuring out what you want.

3. 3:06:20

   So I hacked together images, and then I I was on a trip in Marrakesh with, like, a weekend birthday trip, and I I found myself using this, like, way more than I than I saw, but not for not for programming. It's more like, hey. We are, like there's, like, restaurants because it it it had Google in it, and it it it could figure out stuff. And it's like especially when you're on the go, it is, like, super useful. And then and I was thinking.

4. 3:06:47

   I was just sending it a voice message. You know? But I didn't build that. There was no support for voice messages in there. So so so the reading indicator came, and I'm like, oh, I'm really curious what's what's happening now. And then after ten ten seconds, my agent replied as if nothing happened. I'm like, how the f did you do that? And it replied, yeah. You sent me you sent me a message, but there was only a link to a file. There's no file ending.

5. 3:07:14

   So I looked at the file header. I found out that it's Opus. So I used FFmpeg on your Mac to convert it to to WAV. And then I wanted to use this, but but didn't have it installed, and there was an install error. But then I looked around and found the OpenAI key in your environment, so I sent it via curl to OpenAI, got the translation back, and then I unresponded. That was, like, the moment where, like Wow. Yeah. You know, it's like that's where it clicked. These things are, like, damn, smart, resourceful beast if you actually give them the power. Sure.

6. 3:07:50

   And then I was I was kinda hooked. Like, I did all kinds of of of weird stuff. Like, I use it as alarm clock. I I let it migrate to my computer in London, but then it used SSH to log into my my Mac MacBook and turn up the volume to wake me up in the morning. I think I built world's most expensive alarm clock. Yeah. That's crazy. And it it even got it wrong because I had, like it uses a heartbeat. You know, like, the concept of you do a prompt and you get something is already if this full access inherently dangerous. But I was like, let's turn it up a notch. Let's let's automate that. Let's give it a heartbeat, and and and the prompt was literally surprised me.

7. 3:08:31

   Yeah. Wow. That's very cool. But, you know, I I see this this project as as much technology as it is, like, art and exploration. Mhmm. Because this feels in one way in one way, it's just glue. It's it's it's just putting pieces together that we already have. In another way, it's it's a whole different way how you interact with those things. Mhmm. Because all the technology blends away. You don't think about new session, compaction, which model I mean, maybe a little bit because tokens are still expensive. But, usually, all of that blends away. You just you just talk to a friend or a ghost or

8. 3:09:17

   Yeah. Maybe maybe last year, everyone was wanting these agentic experiences. You were having this experience, and and, it seemed like all the focus was on browsers. And seeing the way that people have been using sorry, Mold MoldBots, taking me a while to adapt. It just feels like all the focus was at the wrong layer. It's like, why do I care about the browser if I can just talk with an agent? Across every app.

9. 3:09:45

   Across every app, every every surface. It's like I don't care about the browser at all anymore.

10. 3:09:50

    Yeah. I I mean, a lot of the prep work I did before I built this was just built little CLIs. Because my my my premises, MTPs are crap. Mhmm. Doesn't really scale. People build, like, all kinds of weird search things around it. But you know what scales? CLIs. Agents know Unix. You can have, like, a thousand little problems on your computer. You they just have to know the name. They call the help menu. They load in what's needed. We are calling the help menu. Then they know how to use it, and then they can use it.

11. 3:10:27

    And if you if you are smart, you build it in a way that just uses what the model already expects. You know? Don't build it for humans. Build it for models. So if they call minus minus log, you build minus minus log. And so it's like agentic driven from, yep. Build build how they think, and everything works better. It's a new kind of software in a way. Yeah. So for most of the things, I don't need a browser. Like, I built something for the whole Google thing, for places, for my Sonos. I hooked up my cameras, my my my home automation system. Like, with every little CLI and skill, my agent got more power and it got more fun.