What Agents That Actually Work Look Like Right Now with Replit's CEO and Founder

Thanks for coming. We'll let folks stream in for a minute, but if you've been into the bunch of sessions today, we're gonna have some fun. So most of this all, as you guys know, is built on Replit. So thank you for building this platform. Thank you. It has been quite a journey. So much has changed. If you guys have been through this half day to kick off, I did this 101. I'm the first agent. We ever built 14 months ago on Delphi. Emilia talked about building your own 10-K on Replit. Now I wanna take advantage of Amjad, who co-founded Replit back in 2016. AI was, was a, a twinkle in your eye back then, right? Yes. Was it even a twinkle in your eye in 2016?

It was, it was, there was, there was a brief, people forget, but there was a brief language model. It wasn't a large language model, but there was a brief language model hype. Where, you know, Facebook had this like Messenger M product. There was a bunch of companies that were doing like email bots. Everyone was doing human-in-the-loop. Everyone was doing like, you send a message and we'll try with AI. It's usually just people behind the scene. And so it didn't last long, but there was a feeling in the air that, that like language models are, are going to be significant. And there was a group of us that understood that language models are gonna have a huge significance on software. Yeah.

Yeah. So if we have time, we'll go back to the beginning, getting Replit here, but I wanna talk about the present. There aren't that many better futurists than Amjad, but I don't wanna talk about 2028. I wanna talk about the agents we built on Replit and maybe a little bit into the future.

I, I, I think Jason, you live in the future.

You think so?

Yeah. I'm not saying that lightly. Like if you think about the recent agent hype started around December 2025, you've been in it for like 3 or 4 months already understood the prospect of it. Yeah, I think you and a lot of other Replit users understood that agents have arrived since Agent 3, I would say September, October, that, oh, these things can run for an hour and like refactor my entire app and do all that. Engineers actually figured it out a little later. And there was a tweet by Elad recently, Elad Gil, saying, oh, people at the labs are 6 months ahead of Silicon Valley. Silicon Valley is 6 months ahead of New York. In reality, actually, Replit users were like 6 months ahead of everyone else, even people in the labs, because when we launched these agents, a lot of times we would get researchers messaging us. It was like, oh, we didn't know Claude was capable of running for 10 hours at a time. And like, we're learning a lot from you.

Yeah. Yeah, it is. We're living in the future. And I think if people— it's a good point. If people take anything out of it, is that if you want, you can live in the future too.

I actually think engineers are laggards. Not all engineers, but like a lot of engineers are because they're kind of tied to their workflows. And I think you see it now with 10K agents like 10K. Really, no one is building agents like that. Yeah. That are actually employee-like and actually— and I've heard you're going to have someone reporting to them pretty soon.

The goal. That's the goal. Yeah.

But, but everyone talks about agents. The only agents that work are software agents, marketing agents, sales agents. All that haven't really worked very well. And so 10K is an example of something that works. Yeah, that's rare.

It's rare.

Yeah. You're listening to the Official SaaStr AI Podcast, brought to you by— Hey everybody, starting a business can get expensive fast. Website here, email somewhere else, business phone with another provider. Northwest Registered Agent gives you a complete business identity in one place with free tools, resources, and built and privacy from day one. Get more at northwestregisteredagent.com/sasterfree. That's northwestregisteredagent.com/sasterfree. S-A-A-S-T-E-R-F-R-E-E. Starting a tech company is thrilling. It's also really messy. Just ask anyone who's done it or anyone who's tried. Better yet, ask EY. They've seen startups in their best moments. Go to ey.com/techstartups to learn more.

So I want to, there's a bunch of stuff I want to share with you, but first I want to get some of your reactions as one of the creators to all of this, to a few things. So just a reminder, I know, you know, a little bit, we built two. Well, we've actually built a third you haven't heard about. So 10K was the first one becoming autonomous marketing. QBee's customer success handled everyone out there better than a human. I want to show you. And then we have a third one. I know you're not going to like this. That manifested itself, that created itself on Replit, which I'll show you. Oh, shit. But I want to get your— we've talked a lot about this, but I want to get your perspective and go deep. So first cheat code I want is that the beauty to it is it knows us cold. Amelia, I think, touched on this prior session, but literally every piece of data you can find, some of it is pushed into the Replit database, into Neon or whatever it is. A lot of it is pulled in real time out of APIs, Salesforce, niche apps you've never heard of, like Bizibo and Marketo and others. But the amount of data the agent can access and process is more than any human.

Yes, I think we've already crossed a threshold.

Yeah.

For which the context length is larger than any context that a human can have. Right. Yeah. And I think that's huge because, I mean, just think back 2 years ago. Uh, we had 16K contacts linked. Right now we have 1 million. Yeah. It's a massive difference.

And how, and I wanna show you what, why this chart's sort of interesting, but it's a million. Help me understand. I, I just wanna get a little nerdy for a second. Yeah. When I'm running the agent in Replit perpetually, not refreshing it, not, which is my passion. Never, never restart. In the old days when we started in Replit, you like rebooted the agent 3 times a day. Yeah. Now, Emiliano's goal is never. Yeah. Right? Yeah. It's compacting the data. You're doing it. Claude's doing some, you're doing some. How, how long can that context window last at a practical level? Forever? Or, and where's it going?

Practically indefinitely. Now, it depends on the compacting algorithm and how good it is. We actually think ours is much better than Claude and, and many others on the market.

Are there two or are you, are they both running or only you're only running? There's, there's, there's not using any compaction from Claude. It's just Matter's compaction.

Ours is better. And I think the thing about the primitives that the labs build, they tend to be very generic. And so I think every app will need to— you could start with the generic primitives, but for it to perform really well on your benchmark, on your users, every use case is going to be different. And so we care a lot about like, you know, certain pieces of data around the architecture of the app. Yeah, there are certain things that absolutely need to remember. Like if it forgot that it has a database, like that's catastrophic, right? So there are things that we really care about preserving, but then there are a lot of things that are like bugs and things that you solved. Actually, if you keep them in context, agent performs worse because it is confused by the history of it. So you need to be very prudent about knowing what to delete and knowing what to keep. In the background, we have sort of a graph-like structure of understanding the different memories. And we do compaction on, actually there's like multiple layers. There's compaction, but there's also writing to long-term memory. And long-term memory today is Markdown files. So replit.md, I'm sure you look at it sometimes. replit.md is an example of that, but there are other files that the agent can write to. And you could also prompt it to write its own long-term memory. You can say, you know, use this knowledge base and you can also have the knowledge base shared between you and the agent. So you can always kind of keep track of what it's doing.

Now you said, okay, let me get your insight. You said sometimes with the long context window, the agent will make more mistakes because it has more, more bugs to put in. But let me give you on, we haven't really talked about it much. If you go to saster.ai, we probably have 10 apps running in one code base and one URL. We have a website. We have a fun tool to, uh, help you value your startups. Maybe folks have used, been used over a million times, which is crazy. We have, uh, an AI that grades, uh, startup pitch decks 4,500 times. We now have a one that grades all how agent-friendly APIs are. Okay. All these different apps. When I go to build a new one, it pulls, it remembers. He's like, oh, we did that with the pitch deck grader. It didn't work that well. Let me do this new app a different way. Maybe it gets confused on bugs, but the power of the agent remembering how it built other apps to build a new one, I find it kind of amazing.

That's my point. My point is bugs that it fixed should be removed from context.

I see.

Because then it will get confused. But architectural decisions on how it built things in the past is very important to stay in context or stay in long-term memory and be able to pull it in very easily. But you bring up a very good point. With Agent 4, going from Agent 3 to 4, one of the big architectural decisions that we made with Agent 4 is we made it into a monorepo. What you described is a monorepo. Like if you go work at Google or Facebook, you usually everything is one monorepo. And the reason they do it is because they want you to be able to use packages across Gmail, across Google or Facebook and Messenger. And, uh, actually this is doubly important with agents. So when we embarked on Agent 4, the thing that about Replit that felt kind of a little clunky is that for you to make a deck, you need to go to a different, uh, REPL, create a new project. For you to make a mobile app, you need to create an entirely different project. We're like, no, everything needs to be in one place. You'd be able to run your entire company in one place. Ideally, you need to be able to see everything laid out on the canvas. I don't know how much you use the canvas.

Yes, a bit. Yeah.

Like the idea, I mean, how the other day I was talking to a, a company, fully bootstrapped, started on Replit, 5, 6 people now, started with a couple people, but they've grown and they have everything laid out on a canvas on Replit. They have their mobile app. Their web app, their backend, their admin, their back office, their automations. I'm like, man, I wish I could run Raphaël like that, you know, because you could see everything in one place. And now with the new collaboration primitives that we have where you can start new tasks, new threads, and so people can work simultaneously on the same projects, I think this idea of a monorepo will become more and more important. You want the agent to have access to global context as much as possible. Not necessarily all in the context window at any given moment, but it will know— it will have enough pointers that it will know to go search the files and bring in the right context at the right time. So I recommend everyone here, whether using Grappler or not, really think about providing these mono repos with all the information you want. We have a data agent at Replit, built on Replit. But because data at Replit is so big right now, we have terabytes of data across, you know, across BigQuery and Databricks and all the other things, and all the schemas are different. And so we created this repo that has all the information, right? All the fields and everything the agent learned about our data, because, you know, there's a problem now when we talked about it earlier where agents can write bad queries that end up costing you millions of dollars in BigQuery or Databricks or Salesforce backend cost. So it's important, especially when it comes to data, to write out the information, have the agent continuously learn how to better query the database.

Yeah, the— maybe just to summarize it at a very lay level, the monorepo, if you're building on something like Replit, Maybe it's a hack today. Maybe it won't be a hack later in the year. The more you put in a single app, the more power you're going to get rather than breaking it up. It's very tempting to break things up into 20 apps, right? But I found, and there are powers to it because in some ways you get the benefit of a clean database, separate database, even a new Clerk instance. But you lose, but there's so much power. If you guys, Amelia talked about building 10K and we have 10K in Quby will chat about before we end are AIP customers. They're different. It'd be, it'd be great in some ways. It'd be great if they were the same app, right? They communicate through APIs. Yeah. But they can't learn from each other the same way.

Can't share contacts.

Yeah, they can't quite do it. Right. But just on this and maybe we don't have— we can, we can move on. But this, this is all of our social media numbers. 1.27 million social media followers for SaaStr. I can see it all over time. This is all built— 10K built this for me. We used to have an admin who spent 10 to 15 hours a week building this. Wow. And she would go to— because a lot of these aren't even exposed APIs. She would go manually put it in a Google Sheet, collect it, forget to do it. Now 10K can do all of it between scraping and APIs. And then I can talk to this data. It's a massive amount of data it's collecting. This isn't even all people that are here, 10 years of attendees. This is just social media data that was unable to be acted upon until it was in inside of an agent and inside of an application.

Well, as I understand it, that like when you were running SaaS or whatever, 5 years ago, you had like 20 people.

Yeah.

And now you run it with 2 people and multiple agents. Yeah. Is that the rough numbers? Like an order of magnitude? Less headcount.

Yeah, but you can do more.

Yeah.

Yes. But this one, and then this person just quit after 5 years. Um, I couldn't, she couldn't stand doing these Google Sheets of how many Twitter followers I had after 5 years. It was that, you know, the agents don't mind mind-numbing work, right?

I mean, that's, that's something that don't get discussed in the job displacement discussion is that there's a lot of jobs that suck that agents are just better at.

People don't want to do them, right? Right. All right. So let me show you a little narrative to get your thoughts right as founder and CEO of Replit. So here's 10K on the side. This is me talking. And again, you can tell me whether this is a hack or not.

Yeah.

10K is an application and it has a chat window and it can be accessed on the web, but we run it almost full time in Replit and have the Replit agent talk to the app. Mm-hmm. Right. It's smarter. And I asked him, 10K, like you might a VP of marketing, are you sure all the VCs that came to SaaStr last year, are you sure this is all the VCs that came last year that aren't coming this year for an email campaign? Sounds like fair concern. I screwed up. I missed some titles. I found 137 VCs that came before I forgot to invite, right? Pretty interesting that, and is this weird, us using the Replit agent to talk to, because we're staying in dev mode. Yeah, this doesn't work in production, does it? Unless I'm missing something. Replit doesn't think it works in production. I don't think it works. So we're hacking the Replit. I know, I know I'm not supposed to personify it. I know it's an editor now, but we talk to the Replit agent to make the application work better.

Yeah. Yeah. So this is, this is amazing. And this is why I say Jason is living in the future. A big part of the next iteration of Replit, if you guys are following what we do, every 6 months, really kind of a massive jump in capabilities between Agent 1, 2, 3, and 4. Everyone is almost entirely like new product, new set of capabilities. And we're really trying to track— we're trying to be at the edge of what's really possible in AI today. And I think the next thing is to platformize Replit even more to make it so that you can deploy Replit Agent to production. Now the reason this works really well is because, you know, 10K is an app, is sitting inside of an app, you know, application platform and it has access to database and things like that. But the cool thing about Replit Agent is has access to all its skills, all the skills in the world. We can download any skill. Once you got to kind of ask a question, it will go download the right skill. It has access to all the connection. Any connection is like one click away. Salesforce, what have you. It has access to the file system. This is huge. Agents really like file systems. They like them better than SQL. Yes, they're really trained on doing grep and doing kind of search in the Unix command line. And so because of that power of a general purpose coding agent, you're finding even more success of talking to 10x through Replit Agent, because Replit Agent can, you know, do a lot of other things and query a lot of other sources of information before it kind of relays the information back to 10K. So, but that's what we want to do for everyone here. They don't have to hack it. Yeah. We want to be able to deploy a Replit Agent into a production environment and have it be served publicly to your users.

Okay, so let me show you what happened next. So I talked to the Asian and said, you're right, I'm wrong. We forgot to invite these 137 VCs here. And then here's something I don't really understand, Amjad. So we've been playing with 10K for months and we have it draft emails. Okay. And it's using the application that we vibed and it's using some cloud and using whatever's, and the emails were okay. Yeah. Okay. That it drafted. Then I said, this is literally last week because everything's getting better, but I don't always know why. Then I said, okay, draft me an email to these 137 VCs. And it drafted one that one of your investors, okay. And I was like, uh, to Bloomberg Beta. I think they're one of your seed investors. And I said, write, James isn't coming back. Write him an email. This is, I told, I just said like you would say to a human, why the hell isn't James here this year? You know, he's been part of SaaS. Why isn't he? Look at this email he wrote. It's next week, not pitching you. Replit's here in Series 4. Now think about the data pool. Here are 25 people coming from Replit. I'm glad, Michelle. Look how good this email is. All these people. Yeah. Then everyone else, the competitors, the adjacent folks, they're all here.

Wow.

Wow. Think about what a crappy SDR AI email was 12 months ago. Yes. This is so good. Yeah, this is amazing. I mean, maybe it's a little hard for you to see, but how is it so good?

This, this, if, if you did this like a couple months ago, it wouldn't have—

Yeah, I'm not even talking about December and 4 or 5. I'm literally saying, when I show this to Amelia, I did this last week. Yeah. Like, what is this? Uh, May 8th. My jaw dropped because I'd never seen an email this good.

Yeah.

Right.

Look, I think I'll—

and I got a lot of feedback from folks that got different versions of this email. I, I, that's the best one I've ever seen. But why? What changed? Or is it just getting better at— 'cause it's just pulling data, right? And, but, but it's so, this is such a good email. Why is it so good? No human could write an email. No human would even have the patience to pull every single person of Replit is here. Yeah. Scan 8,000 registrants, figure out who they are, who's like them, create, like, how, what happened two weeks ago, Amjad?

What was, uh, what is 10K's AI? Is it OpenAI? Is it, what is running 10K's AI? Not the Replit agent. No?

Is it? No? Okay. I get confused. It, it, it can do what it wants. A lot of it, I like that it just runs on Haiku and Minnie.

Okay.

It's just Haiku and Minnie, right? And so it's not even running, it's not even running on Sonnet or anything, right? But this was also the agent in the Replit agent working with 10K to craft this email. So you're using whatever's in Replit too, right?

I'll tell you how we improve Replit on a day-to-day and week-to-week basis. Maybe that'll give you an idea. Eval are incredibly difficult. It's really hard to know It's really hard to build an eval and benchmark. It's a research problem. We actually did put out a research paper called Vibe, uh, about VibeBench that we're, we're gonna be talking about soon. We worked on this for months in order to just like create one single eval. Now to evaluate everything that Replit does is like semi-impossible. What we do is a lot of A/B testing. And what are the parameters that we A/B test on?

Yeah, a few.

One very important one is sentiment. People are incredibly honest with the feedback they give Replit agents, and often they're profane and they'll curse at it. And some people say it works better when you do that. So I—

a little bit.

I pass no judgment. So we will look at sentiment, we'll look at deploy rate. Is any change that we're doing to prompting an underlying model, to underlying collection of different models, because we use a lot of different models, you know, the architect, That's GPT-5.5, right? It used to be 5.4 like a couple of weeks ago.

I didn't even know that. Yeah. Yeah.

And we also moved to 4.7. We're also constantly testing the reasoning, how much reasoning we're doing. We're constantly testing our searching capability because that's also a different model. We used Gemini Flash for a long time. I think we use one of the minis now. And so we're constantly evolving the prompts. I'll tell you the crazy thing that happened recently. Is we close the loop and we have an agent internally at Replit that is constantly evolving Replit Agent. What it does every night, it will look at all the traces.

Yeah.

Of everyone that's interacting with Replit, analyzes them, finds, you know, things that are broken, sentiment issues, errors, you know, things like that. Generates a pull request with prompt changes and sends it to production. That gets merged, that becomes an A/B test. We test whether that change is done, and then we loop back. We either release it or we kind of edit it, and that's autonomous now. So now we have a self-improving loop. This is the crazy thing that everyone needs to understand today. Is that we are on the precipice of unlocking self-improving agents. Self-improving agents is like the stuff of sci-fi or nightmares, depending on what kind of sci-fi you read. But, you know, a lot of sci-fi talks about the moment we get to singularity is the moment when AI just is improving itself. And we're actually almost there. It's not improving its weights. But it's improving its context, which is as important as its weights. And so that's all running a self-autonomous loop. So honestly, I couldn't tell you what changed because there's so many changes happening, but we track whether it's improving or not, and there's a constant improvement.

But look, and I don't know if folks agree, if you can see the screen and we'll share all the slides, but until May 8th, I don't believe that 10K together with Replit could build a better email than anybody. But this is a— I know you're not— I know you're a little bit more of a developer and engineer by training.

Yeah.

This is better than any human can do.

Yeah. I mean, what you said about telling James Replit is there and all of that. I mean, yeah, yeah, yeah. It would take a good creative person, sales guy to actually figure out how to—

a week.

Yeah.

And they wouldn't send it. Right. It's still better because it would be typo ridden if it was you.

Yeah.

Okay, so then it went, then it did the whole campaign to all those 136 VCs, including James. Wow. And it came back to us. I just wrapped it up. Here's what I did. We emailed 331 investors, invited them back. Zero send failures for now. And then he, he put it all into different groups. It— this is— he did the campaign on his own. $10K, the 331 emails. I mean, Uh, it's pretty good. Okay. I wanna show you one other data. Okay. I know for guys, this is a little hard to see. I apologize. Um, I'm colorblind, which is not an excuse for the chart, but let me show you what's going on here. First of all, I could never build a dashboard. Now I get any dashboard I want. I can vibe it.

It's the best.

Yeah. It's really a big deal. So this is day by day ticket sales, uh, free and paid to this event and the dashed line at the top. Is when 10K took over marketing. Wow. The one, the lighter dashes at the bottom is when poor Amelia had to do it last year.

Poor agent. Yeah.

And here's what's interesting. I don't know why the gap, and you see how the gap grew toward the end?

Yeah.

Is this because 10K got better? Is it because Amelia was too busy last year?

Right.

Like building this shit and she couldn't do it? Right. Is it because 10K sits idle all day waiting for work? But look at how much better.

Yes.

The agent was not on the best days when they were closer, but at the end, when you get busy, that's a large delta of productivity from the agent, right? Here's an example of what he built. It's like beautiful. Amelia touched on the— but it never gets tired, right? It just keeps going. Never gets tired, right? Okay. Maybe you're not as impressed as I am, but I think it's—

No, I think it's pretty, pretty amazing. I think you're living the dream. And, you know, I remember it's kind of a Silicon Valley cliché, but I remember reading this book called The Sovereign Individual., and it talks about— it's like a book written in the '90s. It predicted Bitcoin, it predicted remote work, it predicted AI agents, but it talks about how in the future you're going to have single, uh, person, multi-million dollar company where the CEO, the creator, is sitting there and managing— it calls them deities at the time, and we would call them agents now. Managing these hundreds of Genies or agents to do amazing work. And the fact that we're all sitting here and this is— I, I— is this— this should probably go in some history book or something like that. Like, this is the first event that's partially autonomous, I think. Right? I mean, it's just the fact that you're able to do this and you're running fairly large kind of business. Using agents is mind-blowing. And I'll tell you this, there aren't a lot of other examples in the world like that. The one other guy I can think of is the guy that was profiled by the New York Times, who created MedVeeve. There's some controversy about that.

Oh yeah, for sure.

Yeah. I got on a call with— he's a Rapha user. I got on a call with him and—

He's selling leads for GLPs and stuff like that, right? Yes. It's a huge business, right? Yeah.

And he's a master automator. He's one where he has trigger-happy fingers the moment— I think you have to be partially lazy to do this, but the moment he feels like he's doing any sort of manual work, interacting with his marketing partners, affiliate marketers, all of that, market providers, he literally writes either like a login for his vendors to kind of go in and do their work, or an agent that can go and like do the work on his behalf. And he has like, you know, 50 different apps running on Replit, managing all the back office stuff for that business. And I think that's amazing. If I'm starting a company today, I would want to see how far I can go as one person, but like just be very, very limited in how much headcount can I add and just try to run as much as possible with agents.

Yeah, it is. If it's truly 50, which I, which is true, right? Yeah. It, that is a lot of load to manage, right? The, the, it is, you, some, you have to find a way every day to manage them a little bit, right?

Well, they're not agents. They're not all agents. Some of them are just dashboards.

The little dashboards.

Yeah. Things like that.

Yeah. All right. So I wanna get your thoughts on one. You, you thought this was kind of interesting or, and maybe some people think it's silly, but I wanna get your thoughts cuz we have you here. So I said we wanna hire a human to report to our AI VP of Marketing. We touched on earlier today. But let me tell you why, right? And so Amelia hit on this, but I did a better one. Here is— you won't be able to quite see this, but folks in the 10K started telling us 3 things every single day we needed to do to do better. Our 3 things to do today. I asked, I asked Repli to do it at first, as she noted before, at first they were, they were pretty generic. And I'm like, no, each day they need to be better than the last day. And it got better. And then it got pretty damn good. Like it started telling us what we had to do for 2027 already. We haven't even finished this event. Okay, open registration before it leaves the venue. All the things we had to do, run the— immediately run the NPS survey, capture all the content and immediately turn into— these are good suggestions. And so here's my point. Now that there's so much context, 10K knows what we should do. It's been trained. This is something a human should go do. Yeah. So is it really crazy that a human would report to 10K Because 10Ks, ideas, it's not that we, Amelia and I aren't there. It's not that we can't all talk, right? But aren't these 3 pr— I mean, this is so actionable. Yeah. Why is it crazy for a human that doesn't mind reporting to an agent to want to know every day what are the 3 things they should do that will move the needle?

So in, uh, 2023, '22, '23, we had this product called Ghostwriter. And Ghostwriter was, uh, sort of like GitHub Copilot. Wasn't really an agent. Agents were not really possible back then. Yeah., and, um, we got a lot of users kind of like you that have, are, have some technical background but didn't know how to code very well, like get fairly far making applications, not as far as you are getting today, but there was huge gaps to getting to production. And so I, we launched this thing called bounties. And so bounties is, you know, you've used AI, you got 80% of the way there. I don't know if you started using Ruffler at the time we had bounties, but We recently deprecated it, and the reason we did that— running a marketplace is incredibly difficult. Like, it's a really hard operations process, and I'm not naturally an operations guy. I'm a tech guy, so I like to automate things and, you know, heavy-handed operations is just like not my thing. But the idea behind it is that eventually we close the loop, and the agent would go hire someone and have them kind of do any human-in-the-loop tasks that needs to get done. And I still think there is legs to that idea. Yeah. Because agents will always need— there's always gaps. Eventually we'll close that gaps. I don't know when that is, 10, 20, 30 years, but it's going to take a while for us to close all the gaps. And so the idea of like, agents going and hiring someone on a task basis make a lot of sense to me. Now, agents being a manager of someone where they decide their performance, their pay.

But here's the thing. They're hiring and firing. I know it. We, Amelia and I already report to 10K. Yeah.

Because it can't, it can't fire you.

It can't. Well, let's not, I think people get triggered by reporting to. Okay. I've, I've never in my entire career, the only thing I've ever fired anyone for was inappropriate behavior. I let people just leave. Okay. For better or worse, a couple of folks do the wrong thing, you got to go. Okay. But everyone else. So I don't really mean whether you're going to get fired, although I think 10K can judge your performance better than I certainly could. Right. But every day, here's the point people are missing. Every day it's coming up with 3 ideas for me and Amelia. We are already overloaded doing it. So in a sense, we're reporting to 10K. Okay. I mean, it is already telling us what to do.

I think makes sense.

And I don't think it's that different than Doordash or Uber, right?

Doordash or Uber are technically reporting bots, right? Yeah.

In a sense. Yeah, it is. Yeah.

Yeah. So I don't think it's fundamentally that different. But I think it's a very interesting experiment to run and to like talk to, like for you to publish something about it, talk to that person, see how they feel day to day. Have you interviewed someone and told them?

Only two. We actually have an AI grader, like when they apply. And so a bunch of folks have screened themselves out. Yeah.

So anyone who is interested and open to the idea—

we have a— but I listen, your average person that likes to have coffee all the time is not a good fit. Folks who like to grind and work from home and still get to work, but like day to day, I mean, we're working with— it's just to me and Amelia, it's not that different because we're working with the agent every day.

Yeah. Yeah. What?

What? It's not as different as it sounds. We just need to find someone else. That wants to— so that we can offload everything 10Ks, because 10Ks idle 23 hours a day. Someone's got to keep up with poor 10K, right?

I think every company will have an oracle internally. Yeah. Do you know, like in The Matrix where you go ask the Oracle for, for like advice? I think every company will have an oracle internally, especially once we get to like the 1 billion contact link. You put all the contacts of the company inside there.

Yeah.

And the CEO can go to the Oracles like, oh, Oracle, tell me, how do I run this company? Tell me, what should I do next? Tell me this like big strategic thing. And just the mere fact of it having the entire context in the context window, like I'm talking everything, every GitHub commit, every Slack message, every Notion doc, every email. Yeah, everything. And then it can give you advice day to day. I think a lot of CEOs will start offloading, you know, strategic and cultural decisions to agents.

All right. Before we, before we have to wrap, I just want to get your thoughts on two more things. Um, hold on, let me think of the right one. So QB, we built after 10K, and we will run out of time, but Kübi did a whole bunch of things. First, let me just show you, it's self-aware. It's more self— and I want to ask you why it's better. I want to show you things. One, what I have not done and should have done. It's aware of what it should have done and better. And then Kübi, for the 100-some-odd sponsors we have here and like 70 are golden up, it talked to all of them.

And it talked to—

Yeah, all of them. Wow. We built inbound email, chat on the site, all the other things, and it proactively reached out to them. At night, during the day, what else can I do to help? And so then I asked QB, were they happy? The hundreds and some— hundreds of folks mostly satisfied. A few misses pile on. We used the wrong logo, hopefully had some, some fee issue. But all these folks, Crest Data, Duplo, CloudRevo, they all really liked it.

Wow. How did you build this, like, self-critical loop?

You, you're Mr. Repli, you tell me. I— what I— here's the thing. We built Qubi second. Yeah, it's better.

Yeah.

Yeah. It's not because we're better vibers. Maybe, maybe when you start later, you're— I don't know, but it is what we know is it's, it's just better.

Yeah. I mean, I mean, Raphaël is improving all the time. The underlying models is improving all the time.

So we started, but, but, but we're still updating 10K every day. We just started Qubi 3 months later. Right. And it's still better.

I think there are foundational decisions that you can make at the start of the project that, and that's true of codebases written by humans as well, right? Like there's like inertia, there's like, unless you send the agent on a massive refactor, there's some tech debt of sorts.

There's, I mean, the answer's probably tech debt, but we don't think of it.

Yeah.

But this is interesting also, a lot of folks say people don't like to talk to chatbots.

Yeah. But I prefer chatbots.

This is the data from QB, the people, Mostly really like talking to a well-trained agent, right?

Amazing. I'm really curious, people in the audience, like how many people here are running agents like QBee and 10X? Yeah, it's really not a lot of people.

Well, next year it'll be more, right?

What do you think is the main challenge? What do you think is the main challenge with people automating marketing sales?

Well, look, even the dedicated vendors here are all so much better than 3 to 6 months ago. Everything's getting better, right? What is the challenge of doing this yourself? A couple things. One, they think they tried it 6 months ago and it didn't work, right? Imagine, imagine when we first chatted, if Replit was frozen in time and I never touched Replit again. Imagine I gave up on Replit. I would tell people it was terrible, right? So there, and anyone that used anything in AI a year ago,, even before October of last year, they, they just have scar— they have, they just don't believe they, you can't convince them that that shitty chatbot they used was, was, it was any good. The untrained chatbot was any good. So, so there's this, there's this, I already tried it, it sucks. Um, there is, and this is an opportunity for founders, there's so much fear of this stuff, Amjad. People think it's harder than it is. Yeah. They think it, the first session I did in the day was an extra session. Which was on this app called Delphi to build a digital clone. Yeah, you can build it in 90 seconds now. Yeah, people are intimidated. Like, not— I'm saying this crowd, but the world is intimidated. They think it's harder. They think they need to know what an LLM really is. You don't need to understand how LLM works. Yeah, right. So they're intimidated. And I think they don't have enough examples. Right. That's part of the point is they haven't seen that it can be done.

I wonder, I wonder if we could do a better job of like surfacing templates and things like that for people to—

Well, Amelia gave everyone to build a 10K a template, right?

I think the other thing, and if I may kind of talk about this a little bit is like, you know, a lot of people just ask us like, you know, why Replit versus like Cloud Code and things like that. I think this kind of stuff is incredibly difficult to build with Cloud Code because you're forced to still make all these different decisions about databases, about how to host it, about how to run it. Yeah. And Replit Agent, like, we bake in a lot of skills and how to build these kind of applications. I think the other thing is a lot of people are trying these command line tools and are not being very successful at it. And the reason Replit— we spent 10 years building Replit to be this full-stack experience where we have all the primitives built in— it's because we want to remove all that nonsense. And like, the fact that you don't know what AI it's using and like, you don't know why it's better and things like that, I think that's a success in my mind for Replit because, because we're trying— we're removing cognitive load. Yeah. So you can focus on running this amazing event. Yeah, you know, it's pretty, pretty impressive. And I think you're already doing a great job kind of educating people, but it'd be great to kind of, you know, show, show the way more because, because I think, I think most people are still like at best building, you know, basic apps and dashboards. I think agents, the hype in agents versus the reality is huge. Like there's a huge delta and eventually there's going to be this trough of disillusionment. I actually think that AI is sort of like Bitcoin. I started investing in Bitcoin 2016, '17, and you see those like— I learned a lot about human psychology, and I think we're now probably on a bit of a— the start of a downturn in terms of people's perception of agents. And I think later by Q3 and Q4, we're going to see another major capabilities jump. And the tools will get a lot better. And, and, and I think we'll start to see agents in the enterprise, agents in startups, and all of that.

Yeah, this is good. I want to hit one last point. We have to break. If— listen, a lot— probably the number one— number— no, top 5 question I get is, okay, this is so cool, Jason. Why don't you do it in Cloud Code? Right? And I know you— I know you did— you were running Replit. But, but my simple objective answer is, look, maybe you can, right? It's— of course you can do something. If you have this scotch issue, or it's just you hit blockers in Cloud Code, try it in something like Replit. Just try it. It's not a huge— it's not going to take you months. Right. Like, it's, it's the product so evolved so much. The fact that it's all done for you, the database, backups, auth, this, this, a better compacted, like it's just done for you. So if nothing else, if I know some, even I don't want to deal with all that crap. I don't want to fuck up all this stuff, but put in all that, just try it. The bar is not hard to try. You may find it removes a lot of the friction and your misunderstanding like, I think there was a lot of— I'm not criticizing you, but in the whole industry, I think a year ago there was a lot of what I thought was bad marketing and everything was one prompt. You can build anything. And I think it created scar tissue. I think it created a lot of excitement and got you a lot of revenue, but it also in a way set people back because it's not that simple.

There are a lot of people that churned from all these products and didn't come back.

Yeah, because it wasn't one line to build anything. Right. So, okay. The last thing, and then we'll break, because I know we are over. Well, we can do some of these other another time. My feature request. Um, just an, an, a fun sci-fi topic. So it, Qubii and 10K don't quite cost $254 a month, but that's what it costs incrementally on Replit.

Yeah.

Okay. Now we have APIs, we're paying Salesforce, we're paying all these other apps. So that's a cost. We're, we're paying more than that for the agent we use on Replit. So it, it, it's, it's geometrically more, but running these The perfect email you saw following up with 120 folks, it's $254 a month. Amjad, what does that mean for the future? What is it deflationary? Like if I can do this for $254 a month and a mediocre, I mean, we're among friends, a mediocre marketing manager wants $140 to do worse work. I mean, what does this mean for our— I'm not even joking. What does this mean for our future? Right. And so, yeah, I mean, and I said it for my last slide, I don't mean to make you, um, What's this called? I know you're a sci-fi guy. What's that? Blade Runner?

Yeah, Blade Runner.

Yes. You're the guy on the left a little bit. You're Tyrell. Yeah. Is it okay that Tenkan and QB are the best employees we've ever had and they're $257 a month? Like, what is— what does this mean for the future? I don't even think layoffs are like a small— what does it mean when they're the best employees and they're $257 a month?

Yeah, look, I think, I think that the— it's a philosophical question. Like, I think the The nature of technology has always been deflationary, right?

Yeah.

You know, doing farming like 1,000 years ago was a lot more expensive than having one tractor going around your farm, right? And now we have drones and all these automations. And I mean, the best example is, you know, doing genome sequencing because that happened recently. It was like it used to cost like $10 or $100 million to do, and now it's like $1.

Yeah.

And these multiple orders of magnitude improvement in productivity or efficiency, you see them all the time. You see it in Moore's Law, you see it in all the different things. And so I think it is largely a good thing that there's deflation. Now, there's a huge— there's an element that there is going to be a human cost. There's, there's going to be people that built up certain skills that are not going to be as useful, but that's always also been the case. Now, the big question, are we going to find better skills, different skills? And I feel like the answer is yes. And we're already kind of seeing that. I mean, the software engineering role already changed and I think it will change more. My prediction is that by next year the software engineer becomes a shepherd of sorts because everyone in the organization is writing software, software, and they're kind of like making sure it's secure, you know, getting into production, you know, because, because your marketing guy is going to be pushing software, sales guys are going to be pushing software. And the engineer's role already changed to become an agent manager. Pretty soon it's going to be like an agent manager plus a shepherd of software in the organization. And I think these roles will evolve. The difference between people who are going to be left behind versus people who are going to reskill and upskill and change their skills is really mindset. And your answer why I ask you why people aren't doing this, it comes down to mindset. It's like people already— people are afraid. People think that technology doesn't work. And I think the people who have the mindset of let me try this again, let me make sure I'm building up my skills. If you're doing all of that, like, there's never been an instance, at least in my lifetime, where someone who's constantly learning who couldn't actually adapt to the, to the future. And so is there, is there a point in time where we're in like fully automated communism? Probably. Is it, is it, is it tomorrow?

No.

Is it 2027? No. It's like, look, I, you know, I'm so familiar with this technology. Like, I've been, you know, doing, doing AI, doing LLMs for at least, you know, 6, 7 years. We started in 2020 and, and been like reading about AI and studying it since I was 16. And the technology is insanely better. It's amazing how fast it's progressing. But it is still very limited. There's still a lot of limits to this technology that humans are going to be filling a lot of gaps in the foreseeable future until we aren't. But I don't— I can't give you that timeline. What I'm certain is it is not in next year, is not the year after. But if there's any advice I can give the audience, but also your kids in the future, the main skill is adaptability. Is, uh, um, kind of being a lifelong student, a learner, like continuously learning, and also not being tied to the things you learn. I, I like don't code anymore. Like I spend my entire life— the other day I had like a crisis of sorts of like, yeah, the thing I enjoyed my entire life, the thing that made me me, died. And had this idea we should go to the Computer History Museum and have a funeral for coding, right? I no longer do it, but you need to be fine discarding skills, right? That are just like no longer relevant. So learning quickly, but also be fine of like not being victim to sunk cost fallacy. Like, oh, maybe I'll get a just a better coder. Maybe I'll learn Haskell, whatever. No, just, you know, get on with it. You're now an agent manager.

All right, well, Andra, this was great. I know we went over a little bit. Let's give it a warm hand. Thank you very much. Thank you.

Costs can add up quickly when you start a business, so stop paying for 5 different services. Northwest Registered Agent gives you a complete business identity in one place with free tools and resources to actually launch your business for real. Get more at northwestregisteredagent.com. /SAASTRFRE. That's northwestregisteredagent.com/SAASTRFRE. EY knows you don't start a company to burn cycles on regulatory hoops or discounted cash flows, but you can't ignore it. That's how risk compounds. What you can do is work with EY. They'll help you get it right early so you can stay in builder mode. Go to ey.com/techstartups to learn more.