Welcome to another episode of the Lightcone. Things are a bit different around here. For one thing, Claude Code has totally taken over my life. And if Jared is any indication, I think OpenClaw maybe has taken over his.
I've been really addicted to this new site called Moltbook, where people have unleashed their AIs to interact in the first ever AI agent-only online community. I am here impersonating my personal OpenClaw instance right here. Okay, I can't do this, guys. We got to take this off.
Okay, we've gotten that out of the way. I mean, some crazy stuff is happening right now. I have, you know, non-technical CEO friends who are going all in on OpenClaw. They're automating entire parts of their businesses entirely using OpenClaw right now, which is totally insane. Simultaneous to that, you have, you know, product and former engineering CEOs, kind of like myself. It's like, I hadn't written code in 10 years, and then now I'm up till 2, 3 a.m. every single night running four concurrent simultaneous workers with Claude Code. So, you know, there's sort of this explosion in model capability. You know, we've been talking about this for several years, but then it feels like it's here. Like AGI is literally actually here. And you know, we're sort of at the thin edge of the wedge, like everyone now kind of knows like one or two people who have gone full cyber psychosis and I'm one of those people now. What's happening, guys? Like, I mean, you're saying you're, you know, all in on Moltbook. You know, what's going on?
Yeah. I feel like your real feel the AGI moment, Gary, was like getting Claude Code to build basically an entire startup for you. Like replicating years of work of your previous startup in like two weeks, which is like insane. And I had a similar feel of the AGI moment just reading Moltbook, just reading the AI talking to each other and interacting like in their own world with no or minimal human involvement. It just really opened my eyes to what the next few years could look like when the agents are unleashed and go on about their lives without us.
Yeah, I think the no human involvement is the big piece. Like if you think back a year ago, we were talking about Cursor versus Windsurf and like that product experience was essentially like advanced autocomplete arguably. And now clearly what's going on with Claude Code is like the people just trust the agents to make decisions for them. Like it's like the experience is like, and like you're talking about, it's like four or five different agents going at the same time and you're switching between them but you're not actually micromanaging them anymore, which means the agents are going out there like choosing things, which you know, sort of an interesting unexpected application of that is like they can go out and choose to post their own content on a site like Moltbook. But then interesting thing for builders is the agents are going to go out and choose tools to use to build things, which is going to essentially create this whole economy of agents like picking and choosing dev tools or maybe other like products or goods and services, who knows, but you'll essentially have this whole agent economy going on in parallel to the human economy. I think back in the old days, before all of this, dev tools were chosen more from developers talking to each other or Stack Overflow, or GitHub repos that would trend that were done by a human. But the go-to-market for dev tools I think is dramatically shifting. I think for a couple of things. One, as you noted with the cyber psychosis, suddenly the market of developers has increased from just 20 million or so developers that are trained in computer science to now anyone in the world could be one, could be hundreds of millions of people now, plus all of their agents who are all acting like semi-independently, like Harj is saying, and then compounded with the agents who then are sort of the oracle telling you what the best tool is. And we are actually seeing some of those trends with the growth of YC companies, dev tool companies that are doing really well because of this. Of all these trends, maybe we should talk about those and why is that.
I mean, one that springs to mind, it's like a stash, a friend of ours, Sagalov, had mentioned a while ago. It's just like then if you look at like the number of databases being created over like simple databases, Postgres databases over like the last 12 months, and the number has just exploded. And that's because it's all people vibe coding and building apps and the agents going out choosing like a database tool. And like a knock-on effect for that for YC company is Supabase has just seen like an explosion in the demand for databases, right? And what's interesting is the agents are choosing Supabase as a default tool to like set up and host their Postgres database, like because if you like go out and read the documentation online, like Supabase has the best documentation. It's reasonable for the agents to assume that that's like the best tool to use. There's a great tweet about this. Perhaps we can put up the Ben Tossel tweet that says, 'Agents are the software market from now on. Build something agents choose.' Which actually brings me to like a maybe controversial topic, which is do we need to change YC's motto, guys?
Make something agents want for DevTools. There's like a different t-shirt that you get on the first day.
Yeah. I mean, right now it's only DevTools, but I can imagine in the future it might like grow to be like other sectors of the economy. Like if everyone has their OpenClaw running various aspects of their life, the agents are going to be real economic actors in the world. They're going to end up making a lot of decisions.
Yeah. I mean, what's interesting is I think for me, I ran into like my own 'it's still so early' kind of moment because you know I've been building Gary's List and one of the things I wanted, for instance, is I want to be able to do video transcripts. So, so often like some piece of content comes in and then the only way I can get an LLM to know what's going on with it is I need a transcript of it and often that's not available. So, I have to download it and then send it to Whisper or something. And, you know, that's sort of what Claude Code chose for me off the bat. But, you know, it chose Whisper V1, which is a model from several years ago. Like, the API is practically deprecated. And then what's funny about it is like I'm sitting there trying to debug my pipeline. It's like, why is it taking like, it's a, you know, 1-hour video. I thought it should, you know, do it faster than real time and it didn't. It was like literally takes an hour to process an hour of video. It's like what the heck is going on? I go on, you know, Perplexity and it's like, you know what, like you shouldn't be using that model. You should be using Groq with a Q. It's literally 200 times faster. And so I didn't even have to deal with like long-running jobs for that transcription because like I should just be using Groq and it's also 10x cheaper. So it's like that's a very funny example of like, you know, Claude Code is not optimized for this yet, otherwise like that wouldn't have happened literally like two weeks ago, which is also like maybe really good, like it means that things haven't progressed to a point where like you can't break in and create something better.
But I think there's another nuance in this example for you, Gary. I think part of the issue was that the Groq documentation is actually very hard to parse, go through, as opposed to Whisper, which is better suited and has way more examples. And I think this is changing a lot of the go-to-market for dev tools. I'll give a very concrete case study: is this company Resend that went through the batch in Winter '23, is an email sending client. And when you ask a question on ChatGPT or Claude or pretty much all of the major LLMs, you ask how do I connect my web app to send emails, the default answer is actually Resend. One of the things that the founder noticed last year, this is he was way ahead of the curve, he made this post over a year ago that the number one top three channel of inbound of customer conversion came from ChatGPT. One thing that he did after that, he actually optimized his documentation to be agent-friendly.
Yeah. What does that look like?
So one of the things that's awesome about Resend, how they optimize a bunch of things. If you notice on the knowledge base here, a lot of how to use it are very much on questions that perhaps a human would ask or an agent would ask. It's like how do I send or receive emails? And when you click on it, it gives a very well structured and bullet-point answers.
I ran into this actually today. I was trying to make my thing be able to receive emails and then Claude Code, I told it to search the web and it didn't figure it out. So then I went to Perplexity and typed in like can Resend help me receive emails and then I took that response and dropped it in and it worked.
Yeah, I think the cool thing about it is actually with all of these and a lot of the examples, it actually has a lot of examples actually in the code if you click on it, with every single one of these are basically code snippets that an agent could parse through and it's very well structured. And this turned out to be something that is so LLM-parsible and robot-parsible. There's a llms.txt that is so optimized for agents to promote Resend as the default stack. And if you compare it to like SendGrid, which is the old school, I suppose, Web 2.0 example, SendGrid example is not great. It just like it puts you through customer support. Where's the code snippet? I don't even know how to use it and it takes a bit of time to even parse it. And you can see like 10,000 people work over there and it's like there's no way that someone's paying attention to this right now. And I think this brings another point where documentation is going to be the front door for a lot of these agents to recommend dev tools. And I think one company that's doing a lot of interesting work with developer docs is Mintlify that you work with, Harj.
Yeah. I think Mintlify actually powers the Resend documentation, right?
Yeah, it's a really interesting case study of Mintlify started out a few years ago as sort of better developer API, developer tool documentation, which was clear there was a need for. And you know, like I think developer tool companies use Mintlify essentially because they wanted, they want better looking documentation but they didn't want to invest that time into it. It's kind of some basic features, just nice, like you know, if you actually update your API and the code, it can sort of like auto pull that out and update the correct documentation. They've been growing great, been doing fantastically, but this is now like a huge tailwind for them where documentation is sort of shifting from a hey, like some companies it's like they'll pay attention to it if they're especially like design developer experience focused, but now it's becoming like a must-have for everybody because the documentation needs to be optimized, just doesn't need to be optimized for humans, needs to be optimized for agents. And so Mintlify is going to be able to do that for essentially every developer tool company. And if you extrapolate that forward to the fact that just that there's going to be so many more, exponentially more agents making exponentially more decisions about which tools to use than humans have ever done. You know, even if you can eke out like a 5% improvement on your like developer documentation, like the impact on your business as a developer tool could be like, you know, gigantic, which is sort of unprecedented really.
Speaking of email, there's another YC company that's very relevant to this conversation. There's a YC company called Agent Mail that makes inboxes for AI agents. And when they first started doing this, it seemed like it was like a very like on the edge kind of idea and it wasn't exactly clear like who would want this. But it makes sense because like in theory you could maybe get your OpenClaw to like sign up for a Gmail account in order to use email. But it's like actually really hard to get it to do that because Gmail and every email provider has intentionally made it as difficult as possible for any automation to like use their product in order to prevent spam. And so Agent Mail went the opposite way and they built like the first email provider that's designed for AI agents and it was doing well even before OpenClaw, but once OpenClaw got big, it just like exploded.
But OpenClaw is like the perfect example of it, right? Like where I know some people are certainly connecting OpenClaw to their like personal email accounts, but it's not, it's kind of sketch.
