Where AI Agents Are Going, Our Agents Are Truly Becoming Autonomous, and Much More with Replit

last year was like a transition year. So if you came last year, it was like half AI, half old school B2B. And the energy was fun on the AI side, but it was really early on the agents, right? A year ago. I mean, so much. And we had one, we had this one app called Delphi that built digital clones that we deployed and it was it. But, and so half the audience, I was, you know, I love these guys, but I was talking with the CEOs of Dropbox and Calendly and it just was the past, you know, I love those. And then the other, and then Amelia and I were like, should we do it again? And we're like, yeah, but the AI half was so energetic.

And then when we—

honestly, part of it was when we got really deep into rep on the agents, we're like, we have more value to add to the world. And this whole year, the energy here is like, in some ways it's like 2015 again. Like, there's such high energy to deploy agents, to figure out your AI strategy. We banned all talks about the history. Even with Omjad, if I had more time, I'd be like, how did you— How had you done this for 10 years? Like, that's a great— but we banned it. We can only talk about the present and the near term. I don't want to talk about AI in 2028. I don't want to talk about the singularity and post-AGI.

Yeah.

We have—

No more intro to LLMs talk.

Yeah.

No, no. I said assume the average attendee is at $5 million in revenue growing quickly and wants an effin' Agentic strategy. And they're halfway there. They have an AI feature, but it's not great. And they're here to learn today. And I think that's the energy here. And honestly, if you're a builder, You know, this is both the most stressful and the most exciting time in our careers. If you're a builder, it's stressful because it changes. Even if you're at Replit, I bet you're stressed. What's, what's Anthropic going to do? What's my comp— like, it's so— the level of competition is so high. But the fact that we can honestly build at 100 times the pace, and that is, um, if you, if you're not having— if you're not at least invigorated or something, you should just quit the industry. Right? Fun is a weird term if you're, you know, because it's so intense, but If you're not, if this isn't the one of the most engaged times of your lifetime in tech, you ask question, you should exit because there's nothing like this. Right? It's a moment in time. Like in a couple of years, this stuff, I mean, even Replit is so much better than a couple of months ago, but in a couple of years, everything that we think is agentic and crazy will become normal. It'll be, we really will have swarms of agents around us. It won't just be talk and this will become commonplace, but the next 18 to 24 months, like this is going to be the best times of our lives.

Yeah, totally.

I really think it's going to— we're going to look back and be like, fuck, we were part of this.

So exciting.

We weren't— we were on the sidelines, right? It's just—

what's interesting is that I think most of the— most of us that are on the ground floor kind of understand what's happening, and we see a little bit of that. We talk a little bit about this technology as it fades, and it kind of becomes just second nature to expect this to have. And there's this whole generation that's kind of coming up now, and they're just going to get used to talking to this thing. Like, for us, it's kind of new. You know, just like maybe texting was the generation before, whereas now, you know, like, it's just going to be expected that this is there. And, and, and again, that's going to elevate what the expectations are for what the products can do and what people can do with these things. And so I think it's very exciting, and all of us understand that, that like we are on the front edge of this future, like we're all surfing the front edge of this like new thing that's coming. And conferences like this, the energy in the room, it's like it really is just a, an early indicator, I think of, uh, it's a bull indicator to me of what's coming.

Yeah. I think, I think that, um, uh, we're just entering the era, you know, Amjad talks, I think Amjad talks specifically about programming in English. Doesn't he say that? And you know, a couple months ago I would've been like, whatever, that's the marketing of the CEO. Uh, I think it's, it's kind of true today, programming in English. Um, I mean, literally I, in Replit, I don't need to design a prompt anymore. I don't need to have a spec. Now, I think most people should have a spec that are new to it, but I've done enough with the agent and I've done enough of my apps, honestly, with one sentence. I know it's not best practice. I can do a lot of stuff now if I have the context in the agent. Okay, so we're just entering the area of programming in English, but to most of the world, it's really weird. Okay. I think in 2 years, and it's really hard in AI time to predict 2 years, this is honestly what I think deep in this community. Is the line between software and non-software will blur. Yeah. Like our lives will be prompts and no, because just like agents, you know, prompt, prompting's becoming just English. And then, but what's the line between software and just talking to an agent? Like we won't know, like the agent may spin up and create software for real and we won't even think about it as software anymore. The agent will make a decision. Should I plug into an API? Should I just answer it? Should I just grab some data? Should I grab it from Databricks or Claude? Or, you know what, I'll create an application. You won't even know I built it. It'll be an invisible application.

Yep.

Right. So that's what I think. I don't mean— I'm not a futurist, I'm a presentist. But I genuinely think we will enter not the democratization of software, which is what enables, but post-software. Like, we will be— we will be in the ether of software. But I don't know exactly what it is going to look like. Right. But the rate of change is so fast. You can see that happening in 18 to 24 months. Right. The software is just invisibly created behind us by agents.

Can you talk a little bit about the attributes that are going to differentiate the winners and the losers in this next era? So like if you had to handicap folks, like what are some things that you're looking for in terms of like these companies are going to make it, these other companies maybe not so much?

For startups, you mean? For like any tech company?

Yeah, yeah.

B2B companies, the folks that you see.

I mean, you can smell it. You can smell the decay. Yeah.

Tell us more about that. How do you sense that?

Look, pre-AI, you know, as silly as it is, I really think you could measure startups and scallops by story points. Whatever version of story points you used, okay? I don't care how you measure story points. I don't care how you calculate it. Come up with some consistent way to measure your output by story points, and then each quarter drive it up. And I would go to board meetings and startups, and I would see ones where the CTO would say, I don't believe in— I don't believe— we're just— I believe in best efforts on my team. They would fail. Okay? And then you would see folks that would do it, but you wouldn't see the growth. You would want to see whatever metric you want. You would want to see story points increasing faster than headcount. Okay. Then there was a linear correlation before AI, but you wanted to see however you measured that it would going up each quarter. And if you had a really good team and your story points increased, you would pull away from the competition. Like that was pre-AI, right?

Let's explain to people, because I think a lot of us understand maybe what a story point is, but a story point is like a user story. And then a developer typically assigns points to this user story. And that's like how long they think it will take to kind of accomplish that task in development. And so in software development, when you're scoping out work, which means like How long is it gonna take us to build this thing? You have these user stories, and that's what defines the features. And then those points get assigned. And so like what you're talking about is the ability to stack those points basically is showing, hey, our development cycle speed, or we're getting more user stories, we're getting more features shipped.

Yeah, you might, for a feature that took a day, it might have a certain amount of points. For a feature that took, you know, the better part of a quarter would have a certain amount of points. And sometimes you'd even do them in a Fibonacci-esque, uh, scale so people could intuitively understand 1, 2, 4, 8, like how much energy went into it.

And it didn't matter.

It was as long as you could, Roughly, as long as your smartest engineers could agree this should take a day, that was your scale. And it worked pretty well, right? Um, and I bet it still works. I bet Replit doesn't count story points. Maybe it does, but I bet you could do it. Like there's probably, and I, I, you, I couldn't, you could tell me what Replit uses, but what I want to see is radically increasing velocity, radically, radically. Right. And I, I think even for the best startups that I've worked at and invested in, the true radical acceleration only happened at the beginning of this year. It only happened after 4.5 and 4.7. And then, then people had to take up— even the best CTOs that weren't at the front lines, like, they had to take a beat and really understand, what does this mean for my team? Not just me. And a lot of us were fracking around and doing stuff before December, but most of the mainstream tech world was not taking a lot of this seriously before that level. And obviously with Replit, with v3, you could actually do amazing things before that. But I think only until— only in like January, February, March of this year did even the most agile teams start pulling ahead. But now it's compounding. Yeah. Now it's compounding. So what you guys probably saw a year ago or two years ago at Replit, you're now seeing across the world. But if you just— however, I don't care how your CTO or your founder is measured, if you don't see radical compounding of productivity, you're— I mean, I know this sounds brutal, but you're gonna lose in the marketplace because there's so much competition.

I think this— I mean, this is one of the things—

we've never had competition like we have today. You guys are in one of the most competitive spaces.

This is one of the things we talked with both Michaela yesterday and Amjad was about the amplification of the efforts and not specifically like, sure, Replit's kind of building tools that allow other people, but we internally, how we use it, we talk about dogfooding a lot. We talk about eating our own dog food, drinking your own champagne. We brought up yesterday, and it is this idea that regardless of what happens externally in the market with these tools, what's happening internally inside of Replit as we like build and all are empowered to kind of amplify our work with these tools.

Yeah.

Is really the case study for what a future-looking organization will look like. As we amplify the output.

I hope so. It'll be interesting to see it. I'll tell you though, it'll be interesting as, or when I was, when I was at your Replit all hands last, I don't know, let's make it up September or something like that. It was still, even though you were on fire, it was still a small company. Everyone was in one room. Everyone knew each other and it's easy. And I, and if you'd asked me what that dog food standard was that day, if you'd asked me in the all hands, I would have said, guys, everyone in the room should be better than me and Amelia. There's no excuse. We're 5 months into this journey, and I would have said— sometimes I'm tough, it's tough love— but I would have said, I bet half this room is not as replete fluent, the term I used back then, as me and Amelia. And I'd give you a gentle F. Yeah, there's no excuse. But fast forward, but as you scale, your challenge as you scale is you go to 1,000 people, 2,000. Even as efficient you are, let's see if you can maintain it, right?

And by dogfooding, it's not necessarily like the sessions that we have, which we do these dogfooding sessions where it's like, let's get everyone on the team to kind of like test this feature that we're about to launch and make sure we put it through its paces. What I'm talking about is everybody in their day-to-day role, at least one person on every team, two people on every team are kind of tasked with how do we implement Replit inside of this like group, this team here. And so like before we think about grabbing some SaaS off the shelf or some off-the-shelf solution, we're like, how do we build the pieces we need internally? And we, you know, that becomes again, these internal tools that maybe not everybody, right? Cause I agree with you at a certain point you have to scale other departments who are maybe not as important for them to be intimately familiar with the tool, but somebody on their team is and thinking about this. And we have people that are literally tasked with going into those teams and solving those problems with Replit.

I guess.

And that's, and that's pretty cool.

For what it's worth though, I, I, I, I would challenge you for fun.

Yeah.

If, if you have these tools and you're designed to do more with Replit internally, if you can buy a better tool, you should buy it. Otherwise you're going to come to the wrong conclusions.

Sure.

Don't waste your time rebuilding any tool that exists here.

Sure.

It, the cost is irrelevant. What you, what you'll challenge your team to do is instead of vibe coding their own CRM, they will find the n=1 app or the app specific to Replit and they will learn so much more automating something this then if you go to compete, if the team goes and rebuilds something that's great externally, you learn nothing.

Yeah, but that's not right.

So let me go ahead and jump in here. Yeah, I don't really think things that I think are good evidence of this. So one of them was I pulled a little while ago. So before, like back when we first met, I was kind of like the only guy on the sales team. Like now we have like a big org. I pulled like their Replit usage versus like their performance of who's hitting quotas one to one. Right? Like the people who are using Copilot more themselves were able to do it. And like part of the reason why is because everything's changing so fast. I think you've seen this because you're on the edge of this and you're using it all the time. You're able to see like kind of where the puck is going. And also this might not work today, but like the next model release, this is definitely going to work and I can hop on it immediately, right? Of like the AI SDRs. And originally when they first started, like, like they weren't exactly working well. Then you get a model release and suddenly the same things that you had done then are now working. And so we're seeing a lot of that where people who are using Replit more themselves because they're more in touch with it, more touching it more themselves when they're working with the customer, they can see that, oh hey, this is something that dropped last week or a new feature thing that enabled this thing that you thought was impossible is now possible. And then you can show it to them and like immediately get them moving.

And then—

or even better, man, I'll tell you the best thing you did in sales. No, no, this is something— this is why you have the one-to-one correlation for Replit experience. It's one thing to tell me something that's going to drop. Okay. It's another thing in a very fluid environment to tell me why something doesn't work the way it's expected and when and why and when at all it might change. So like, listen, you're, you have, you have incredible revenue growth, but it is competitive. There are bake-offs. Okay. And I could talk to someone that's never used their product and get stupid answers. Or I could say, okay, like for example, I mean, you, you guys natively integrate with Databricks, right? Maybe it's weaker with Snowflake. Okay, I'm making this up. I, we come up with a why. Okay, give me an honest answer during the sales. And I understand what are the odds that, that your Snowflake integration will be parity with Databricks? Honestly, it's low. It will work. I'll tell you 10 customers that do it this way, but Databricks is just more native to the way. And here's why. And all of a sudden I like one of the best things that Cody did. I, I told Amjad that he's a 10 outta 10 and to promote him. Um, boom. No, no, because yeah, yeah. I will tell you what Cody was able to do that probably no one I've ever talked to was able to do, um, once he got over the shock of the first conversation he had, is he would just tell me the why. And in a, in a, in, in a world where the software doesn't change for 10 years, I don't ask for a why from sales. I use Salesforce. Listen, Salesforce is changing. I'm excited about it. But for 15 years, it was the same. So you didn't go into your sales and ask your rep a lot of questions about why. When I know, when I'm buying, I'm buying a journey with Replit, not what, not just how it works today, how it's going to work next quarter, next month, tomorrow. And when someone in sales, broadly defined, can tell me the why, I want to work with that vendor.

Yeah.

Why?

Why? Why is this hard? Why is this easy? Why does— I've been using Replit because a lot of your folks can try the product before they write a big commitment. This worked really well for me, like, but I have a question about how the security audit worked. Why? Why does this work this way? Why did domains work on single accounts but don't work on enterprise and team accounts?

Why?

Why can I not buy my domains I just get them to move over to my enterprise team. But, and, and, and well, that's the way we architected it. There's some security and other issues. Okay. I'm not happy, but I get it.

Right.

Versus I don't know, I'll have to check on that.

Yeah.

We have domains in our product. I did. I didn't know that.

Well, that's a piece that you've always brought up on a lot of these agents podcasts that you've been doing is like, hey, like the forward deployed engineer or like the person that you're working with on that side, like if they can't answer your question, I'd rather just talk to an agent. Right. And like, you know, like the focus should not be just kind of throwing warm bodies at this. It's more of like, take your best guys and then make some sort of process, an agent, a workflow, like some sort of architecture hub, some documentation that an agent then can use where you can get it to like 80% of like the best guy. And like, that's the thing. I've seen this across the industry and you're so good at it is that like the people who are just kind of throwing warm bodies at it as opposed to like, this is the best guy, this is the best email that's ever been written. I'm not going to try to replicate the best, but I'm going to try to get to something that can do 80%. And maybe you get to 70% and then a model drops and then suddenly it's able to get to the 80 or 90%. Like you were talking about with Amjad yesterday where you were like, that was the best email I have ever seen written for a re-engagement.

Yeah. I don't think Amjad is so good. I don't think he totally got how profound that was to me and Amelia. Like we felt we were in the middle of the night working and I just did it as a test. I didn't know how good it would come and it was immediately we got it. Like this was the best thing I've ever seen in email, in marketing, and by far out of repertoire. Like something happened. Even Amjad didn't know what it was, right? Maybe I didn't describe what happened enough, but if you read that, go back and read that email carefully. No human could do that well.

Is it the email you generated for your prospects or something you got from, from Replit?

I asked 10K, our almost autonomous VP of marketing we built in Replit. Yeah. I said, uh, I was just curious. So I'm in the, to be clear, I'm in the Replit agent talking to the app. So we're getting a boost. We're getting a superhero power because the agent is talking to the app. But I said, I don't understand something, 10K. The world's changed in the AI world, but we do not have as many VCs come. This is a, this is a, it's not a record, but it's certainly the record for the last 3 years attendance. But VCs are light. Why is it? And at first the agent was like, oh no, they're great. I'm like, try harder.

Okay.

This is my, this is my version of working with the LLMs. And he is like, you're right. It, it, it is, it is really light. And I'm like, okay, how many VCs are coming? 152. Compare that to the prior year it did it. Okay. And he's like, you're right. This is a real issue. Something's going on. I'm like, okay. I don't know what's wrong. Write me the world's best email to these VCs to come. And I said, start with Bloomberg Beta. They were an early investor in Replit. Let me see what you could do. And it just built the best email. And the other emails we had 10K or Replit build or Claude or whatever, they were always fine. And maybe it was because it finally had so much context from the APIs from the, and maybe even it's all the accumulated context in the, in the, in the chat, in the window, but, or maybe it's something Replit pushed out the hour before. I don't know, but. It cropped— I mean, I really hate the term AGI, but when it writes a better email than any human could, it's like—

it's just a pat on the back, right?

No, because the thing is, the reason it's so good is it's not just the quality of the email, which was undoable. It was the amount of hours it would take to find, okay, who— what's every single person from that company? And then it assembled all the adjacent competitors for Replit. They're like, here are all the people coming to James M. Lemkin. You might want to go meet with all— and it assembled in a way that made it seem like James had to be there. He's like, You invested in the company. Uh, the half the company's here, the adjacent folks, the other players, the new players, the competitors are, how could you not be here, James? And it's like, yeah, how could you be not? Right. And then I had 10K write it to the CEOs of all the companies I invested in. And it went through all 8,000 people coming and assembled the best list for them to all meet. And it destroyed all the matching software we have here. Destroyed it, destroyed it. Now it did take like 2 to 3 minutes per email. That's a lot of compute and going on. Yeah, a lot. But, and I sent it to one of my founders and it uncovered that 8 of his competitors, 8 of the entire management team of his top competitor were coming and all the adjacent folks in his marketplace. And then it summarized why maybe it wasn't worth his time to come this year.

Oh yeah.

No, he's like, he's like, no, he was like, you're a part of vertical SaaS is relatively lightly attended, but here are the 20 people you should meet. And he's like, that's the best marketing email I've ever gotten too. I don't know how we got on that track, but that is, Yeah, well, it's super powerful, right?

Well, that's, that's a piece that I've seen that you and Amelia are extremely good at with working with the agents. We were talking earlier about like instead of focusing on the how for it, focusing on defining the end state and the goal that you're going for and also giving it context on the why of like, and that's why I think like my kind of like pet theory about like why you've been so successful with this is that you have a lot of scar tissue over like a lot of doing this is, you know, what right looks like for a lot of these things. And so you're also when you're communicating with the agent is that you're very good at describing what right looks like. And then kind of leaning in and letting it figure out of like the path that it's going to go down to achieve that kind of like right end state for it. But then you're also feeding a lot in because like you always talk about it as like training the agent for it and sort of like, hey, you can feed like the world of context into here, but also what— how should that context be looked at? It should be looked at from this perspective of choosing this end state, this goal. This is the actual problem I'm trying to solve for it. And then not getting in the way of the agent figuring out the interesting way to solve that problem.

Well, I have a theory. I'm sure you're right. And you can, you would know better than me, right? I have a theory. I think that's all true why we're successful, but I have a different theory.

Okay.

And maybe I'm giving too much credit to the context history of the agent.

Mm-hmm.

I have a theory that we try a lot of things, like we tried doing emails, doing campaigns, and sometimes the agent doing a good job, so we abandoned it. Okay. I have a theory that when you tell the agent that was an incredible email, let's do another, somehow in its memory or the context window, it starts to understand what for a certain type of action the user, right, thinks is high quality, learns from that. And then when you do something somewhat similar, it knows— I don't— my prompt is not very good. My prompt was write a great email to Bloomberg Beta. That is not the best prompt. Like when I would show my prompts like 3 months ago in our Workshop Wednesdays, people would mock me in the comments. That's not a prompt, Jason. Yeah, that's it. I'm like, well, I've been doing this for a while, right? But I have a theory that whatever it is, the agent, the context, the memory, it knows what's good, remembers that at least within a session, which we try to make infinite. And that's why our software is better. Yeah, because we train— I use train in a vernacular sense, but we inadvertently train it on what is good by giving, by taking, by doing it 100 times. Yeah.

Positive reinforcement on many routes.

That Bloomberg email and send it to 100 people. The agent knows that that was good.

Yeah.

If I say that was terrible in all caps, I think maybe, maybe the agent somehow marks down to memory or something and says this was not a successful action. Right.

Well, I was, I was talking earlier today with someone else on the sales team and they were thinking about building out a new agent for like, I've been having them read a lot of your materials. I pass around the Agents Podcast for everybody within the company. And, um, one of the things he was talking about is like, hey, here's the specific aspect of engaging people on LinkedIn who engage with posts that Omjad posts that do these things or whatever, how do we do sort of these enterprise follow-ups for it?

Yeah.

And I looked at his process for it and I was just like, hey, where's the step where you're reviewing the output before it gets sent and having the feedback loop for saying this is what right looks like so that it gets better over time? And I was like, that's the most important thing. And then like his plan when he was going to go through it, he's like, hey, I'm just going to kind of turn it on and then we're going to start running with it. I was like, hey dude, you got to listen to Jason's podcast, right? It's like you have to have 30 days. You have to say, Hey, we're gonna start it this way and every day I'm gonna take an hour, I'm gonna make it better. And I'm not just gonna fix the one email it wrote, I'm gonna fix the process by which it creates that email so that at the end of the 30 days I have something that can reliably do something 80% as good as the best guy at this company of doing their reach out, of doing their responses, all those kinds of things.

And in the data science world, we used to call this labeling. So you'd have a bunch of at-bats or rolls and then you would mark them as like, yes, this is good or no, this is bad. And so you're kind of doing that on the fly. You're seeing an agent rip through something and you're saying like, yes, good. No bad, and agent somehow is keeping track of that to, to optimize for that yes outcome.

I think that's what's happening. I think that's my theory. I don't, I don't, I don't work at Replit. On the, on secret sauce, on Cody's example of, I think, trying to, trying to do social selling, respond in LinkedIn to comments, right? And maybe try to do it programmatically through the API or something like that. Is that what they're trying to do?

Yeah.

I think you could do 120% better than the best human. Which I think is what you should think about. Here's why.

Okay.

So social selling sucks. Okay. It's so lame. It's so, it's so pre-AI. Social selling is I, I, I, I write something about how I vibe coded an agent who was really successful and I get a comment, uh, great job. Wanna try our product? Like, I don't, I don't, I don't care. Like that's not adding value. Right. What real social selling would be is, uh, I was, I was, I was vibe coding last night on Replit. I'm using the non-native Clerk integration, not the native one, the non-native one. I'm struggling with the private key. I'm not getting it to work. And then through AI, not only do you see that and spot that it's in your ICP, you answer the question.

Yep.

I've seen— we've— I've seen that problem many times. Don't get upset with Clerk. This is the nature of how OAuth works. Here are the 3 things I've seen. Boom. You have added value to that relationship. And if it is an actual prospect, The odds I buy from you versus some dumb rep that writes a smiley face on it or great. Oh, I see you went to Harvard. I know someone that went to Harvard. Oh, I see. I see you watch basketball. Right. I mean, I get these ones. So I tried to do a Product Hunt launch like a while ago when we built, um, we built a product on Replic. What was it? Uh, uh, our API grader, uh, oh yeah. I'm so busy. I just set the date up wrong on Product Hunt. So I didn't realize I launched it like a month after the product. And I get these delayed emails from sales reps. Congrats. I saw you watched on Product Hunt. Would you like to try our— what, how are you?

Right.

Do a little research on me.

Yeah.

I have budget.

Yeah.

Okay. I built this. Find something that your product does that will help me, but not like, I saw you launch on Product Hunt. Would you like to buy my HR management process? So I'm rambling, but like, right. I think I firmly believe, and I'll add this to the agents, that 80% as good as your best human when that's good enough in sales. But there's this type of social selling, this, this, this agentic selling where you can do better than a human. You can add more value. And I have never seen this done until this podcast, but I think this should be the goal because anyone that has an issue in the entire vibe coding world, there's so many issues. What if every single time Digital Cody, right, or Digital Manny or whoever answered the question as best it could for real? It would be a tremendous value add, right? And you train it and you, you read every single one, you build it, and then you read it every 30 days and then you let it go, man.

So I'm going to go back to the point earlier where you were talking about don't you know, rebuild the SaaS or the thing, you know, which, which I think is very good advice. I think most people who are being successful in the Viacode space are not trying to like recreate what exists. What they're trying to do is plug gaps and connect holes between.

Yeah.

And so going into what you're talking about, we're building things like that, let's say, that take context from lots of different places to help shape some of that messaging. Because a lot of times what happens is your best salesperson or your best customer service rep or whatever, is getting input from lots of places, not just this ticket that the AI might try to respond to, or this email that it reads and has zero context about. Okay, what's being talked about? What's actually happening? Is there— is the system down right now?

Right?

Can I check the uptime monitor before I respond to this so that I know, oh, there's a little thing here? Oh no, systems are up, but there's some things happening. Oh, there's some noise on Reddit. Oh, they're talking about this in Discord. They're— oh, listen, so-and-so just made all these posts on. And so you can kind of see. So we connect little pieces to then bring in and amplify the work of the individual, which creates a dashboard that doesn't exist on the market. Right. So when we're talking about dogfooding and building some of these things and where I think the enterprises are really winning, it's like, hey, we have this system, this system, and this system, but those systems don't really talk to each other, you know.

And I think—

and now we can kind of—

dashboard is one of the best use cases for Replit.

Yep.

I tried to show Amjad one, but I threw so much at him. It was a lot for him to process. But the first slide I did, Um, uh, is, uh, our social media followers at Saastr. So I have like 1.2 million social media followers. But to your point, um, some of these, like some of the things, um, so I, we had an executive admin who her only job frankly for 4 years was to spend hours every week going, logging into WordPress, logging into Twitter, logging into LinkedIn. How many followers does Jason have? Put it on a spreadsheet and track it. Some weeks she hated it. As time went by, it was menial and she didn't do it. And then I, and then, you know, it's funny, one of the first Replit projects I did was, can I automate it? But not enough of them had APIs.

Mm-hmm.

Right? Then I did it again, like, a couple weeks ago, and it's epic. And now I have dashboards for the first time in my entire career as an entrepreneur, executive, founder, and CEO with Replit. For the first time, I have dashboards that, that benefit me.

Mm-hmm.

We run a headless Salesforce inside of Replit. Inside of our 10K agent. So I can see every sponsorship, every deal, every ticket flowing through the system in real time. I'm not going to log into Salesforce and find a dated dashboard that's never been updated. I'm not going to log into a marketing— I, I just don't have that patience.

There's all this other noise around it where it's like, I just want this metric.

Now I have dashboards for me, yep, just for me, for the first time ever. I mean, um, and all that context now is powerful. What?

And all that context is now in agents. So that's the reason why agents can write something that's better than a human, because it has this entire, like, context for our social media.

So agents are a big part of it. Um, and then also underdiscussed, I mean, it's your talk, we can talk about whatever you want. I think underdiscussed is how tools like Replit democratize APIs. Completely underdiscussed.

Mm-hmm.

APIs until 6 months ago were for developers. It's under discussed. I mean, I got us going on Replit, but as Cody knows, Amelia is better than me now, right? And she, she's smarter. She's more patient. She's more quantitative, but she's not more technical than me. She's less technical. I mean, she's, I mean, she knows a little more HTML than me, but she hasn't built what I built on the software side. But because of API, she can build anything.

Yeah.

So you can build anything with an API.

So that connectors in there, you can grab the API docs and just copy paste it in there.

So if I want a dashboard out of Salesforce data, I can build it in Replit now versus impossible before. I'd have to hire a developer. I don't know if I didn't do it right.

You mentioned—

so we will all build apps on top of APIs that— and use APIs in ways that no one thought when they built pre-AI APIs. Yep.

Hey, real fast, I gotta hop to go teach a Python class. This has been absolutely amazing. I'm such a huge fan of Saster, and thanks so much for all of your Replit journey with me.

It's been great. It's fun. Do you need me more, or we— are we done, or you need another?

You're welcome to stay as long as you want.

You I got another session I gotta do in a minute, but you have one more thing you need from me?

Yeah, last question.

I got one more.

Yeah.

All right. So you mentioned how the way you're using different SaaS offerings like Salesforce, it's changing. So I guess I'm wondering like what advice do you have for folks, businesses out there in terms of changing their services, their value proposition to adapt to this new world?

The simplest thing I would get, and I built this on Replit, if you go to Saastr.ai, and go to the nav down there. There is an Agentic API grader. I built this on Replit. It, it's cooler than people realize. Okay. Cuz I couldn't do this on my own. And the Agentic API grader will grade your API. Oh, how friendly it is to agents. Okay.

That's key.

Uh, Stripe got the only A+. As you would expect, Anthropic, OpenAI got A's, right? Gemini, I think got an A or an A-. And if you think about it, what's the simplest thing you can do to your software? Even if your software is dated, Okay. Even if you can't get anyone to fix any bugs, if you can just update your API, then you can turn up— and I'll give you an example. Now you can turn kind of moldy old software into pretty state-of-the-art software. If the API works with agents, you can bypass all the UI, UX, other issues, right? And so for example, running this event, there is a niche piece of software called Bizibo. It is a CRM for events. Okay. Until 2019, it was state-of-the-art. I don't think they've added a feature since 2019. Okay. And it only survives because it is an uninvested category of software. Okay. And we're at the, at the edge of quitting, at the edge of quitting, but we asked the Replit in our 10-K, how, you know, can you interface with the Visible API so this dated thing can be expressed in real time for it? And it said, well, good, bad news and good news. The 10-K said built in Replit. Bad news is the API as documented can't do everything. Good news is I found undocumented APIs that let you do it. So what it did, combined the agent found undocumented infinite APIs and calls that put it above the line. So now we're gonna renew Bizabo and we're going to invest in it even though the, the web software, it can't keep up. So my point is, if you're a founder or an entrepreneur in tech, maybe you can't fix everything on the front end, but you can make your headless Salesforce, your headless Bizabo amazing by making it the world's most infinite API calls, all the context I want. So just go to this, it's part of Rep— we build on Rep, but go to saster.com and find the API grader. You can— it'll create your own API for you.

I see Perplexity, the CTO, also has turned his back on MCP, and they were one of the originators on some of that, and they're basically saying they're going back to APIs.

Yeah, that— well, that's what I learned from the grader. Like, when I talked with the Replit agent about it and Claude and others, he's like, don't even give them any points for MCP. Being MCP— MCP may help humans connect to your app. MCP does nothing to create an agentic automated workflow. Agents are going to work at the— build a great API. MCP is a dalliance so that I can get things to show up in Claude that I may not even want to do from Claude, right? It doesn't, doesn't make my automated dashboard and Replit work any better, right? I need a hardened API that just works.

Performant, easy auth.

Yeah, so this will— great performance, right?

Easy auth, stable, scalable, like documentation, all the things that we all know we want already.

Documentation. And those are— if you go to the API grader that we build on that Replit built for us, it— that's exactly what it says. And it grades them all on a 1 to 10 score. It runs 2 passes. First it gets Claude's thoughts, then it gets Gemini's thoughts, then it gets any other thoughts we had, then it aggregates it. And it's just, it's really just those 4 criteria, right?

Well, and I think this is going to be—

but you can do, if you, if you have a competent team, I'm not talking about Replit quality team. If you have a competent engineering team, I believe you can at least do that in a week.

Yeah.

You can get those scores, documentation, calls, um, OAuth, you can fix, you can just bang those out. So that in one week you may not even be able to launch a major feature, but you can launch an agentic, uh, an agent-friendly API, right? It's a big deal.

Well, I think this is the, this is the evolution of all these companies. People talk about the SaaSpocalypse and it's like, yes, if you don't upgrade and go API first, agent first, and you don't offer your solution as a headless API-driven thing that the agents can play with, then yes, you may be end up replaced. But if, as you see, uh, Salesforce just came out. You know, there's all these different companies that are now pivoting, revamping the API, making it agentic, AI-first. Again, if I go to your doc site and I can't click with one click to grab the markdown or send the agent to it and they can easily read it, I think those are the companies that are going to have a real hard time moving forward. But like you're talking about, if your API is robust, if you've got great documentation, if you can feed and interact with the agent, then, you know, there's some opportunity there at least to kind of work and be symbiotic in the space.

Let me give you another example. So Bizabo was this, again, this 15-year-old platform that got good enough because its API was, did, did need some undocumented. The aha moment I had at the beginning of this journey, and then we can break, was, um, I built a game too on, on Replit called Founderscape. It's really fun. It's a founder simulation from the beginning days to building your team, to IPO, to acquisition, even kind of to the singularity. If you play the game long enough, it's got this end state where all that matters is gigawatts and, uh, GPUs. Uh, it gets pretty funky at the end if it, only a few people have gotten that far. It's really cool simulator. It's free. It's founderscape.ai. Anyhow, so I decided after, after about a couple weeks of building this game, I learned a lot about building a game, is I wanted your co-founder to talk to you interactively. So you could talk to your co-founder, who should we hire? What should we do? Like this, right? And it's pretty cool. And I wanted to integrate 11Labs, right? Which maybe it's native in Replit today. It certainly wasn't native back then.

Sure.

Okay. And I said, I've never used 11Labs. I, I, and, and, and what I should have started the story was 11Labs just crossed $500 million in revenue in less than 24 months. Okay. And I said to the Replit agent, help me integrate 11Labs. Of all the APIs I've integrated that aren't like built into Replit, this was the easiest. It took me minutes to get my key from 11Labs and instantly put it into the secret store and it just worked. And it's not even Replit's credit because it's not a native integration, right? I mean, it helped that the secret store is easy to use, but my point is 11Labs did everything to support making it as effortless as possible and it just effing worked, right? And I'm like, And I said, I even said it to Marty and others, like, this is the, this is the best thing I've ever used. And, um, if you can do that, even if your software is 20 years old, you will get more and more customers, even if you can't do all the rest. And that was like, and then I'm like, I, you know, I, I don't even know how you use the web interface and I don't care, right? So I can break, but if you can do that for your app, you can go from, from tired to wired. You truly honestly can, right? And the last example, I'll give you a break is Twilio. Yeah, I talked in the— Twilio went dead, dead, growing 4%, going on the way to zero 18 months ago. But is Twilio built into Replit as a default integration?

Yes, we have SendGrid.

The only problem with that is that you just can't one-click generate a phone number for the SMSs, so you have to kind of do that on the backend. But yeah, you can connect with their API, and once you then bring the phone number over, it's all great.

So then all of a sudden, Twilio went from this dated commodity product to being super agent-friendly. Yeah, not at the ElevenLabs level, but, but then their growth went from 4% to 20% in one year.

That's what I mean about some of these primitives.

Do that yourself.

That's what I mean about some of these primitives. I think people underestimate the infrastructure or the ground-level advancements that are needed for real invention to take place. And if you go back, a lot of people talk about like, oh, this is the printing press moment. And it's like, what they don't understand is that the printing press wasn't just a thing that was invented. You had like paper manufacturing at scale, ink technologies that were advancing that allowed you to like print on paper and not have it just smudge and smear. You had eyeglasses that were also just kind of starting to be prolific so people could read smaller print. And so you had all these things that kind of like stacked on top of each other that made the printing press viable. There were versions of it before, but you just couldn't scale it, right? And so now we're kind of in this moment where like, okay, the primitives are not go rebuild Airbnb or Salesforce or any of these large SaaS from the ground up. It's like, those are now the primitives to build on top of. The next layer of invention is connecting the dots between these things. It's building and filling the gaps between, you know, the use case that you have versus the use case that they're building for at millions and millions of users at scale doesn't match the thing you need specifically. And so those little solutions, those little things that only you can see at the ground layer is where the people who I think are having the most progress in vibe code or in this agentic engineering era are having success, right? And to me, that's the sweet spot right there. It's like, go get the primitives, build on top of that, make sure everything's safe and secure, and then start experimenting, and then it's off to the races.

Yeah, we can break. I gotta go do a session, but I really think to tie it to the end, I— my lay version of that is, um, uh, if, if, if you're not sure what to do, build an n=1 app, make a list, sit down, and just force rank. What are the— for simplicity and need, what are the simplest 5 things I can't get done today in software? Not the most complicated 5 things, because maybe someday— think about it. What are the top 5 headaches I have that my stack just can't get done? Okay. And then rank them by simplicity with top 5, and then do your top 5 and then rank them by simplicity. Then ask Replit to go build it and see what happens. You can't lose. Like the worst case, you're going to lose 20 minutes of your time, right? And a few— and, uh, $20, $20, small chance today, if you invest 20 minutes, 30 minutes, you might solve an n=1 problem that to you might be worth 20. Not just— it's not just the hard cost, it's the soft cost. The fucking soft cost that someone has to spend a week printing out parking passes for 5,000 people. And the RepliAgent did it for me in a couple hours. Takes a 5,000-piece PDF, slices it into each page, gets the right one, and then routes it exactly right to the right individual all day, all night long. That is— I know. I mean, Amjad's so good. I couldn't go deep enough on that story to really impress on him how crazy, but that in some ways is the most impressive thing we've built. Because you could— is such a— as a weeks of manual mind-numbing effort, right? That is any— only we can use parking passes at the San Mateo County Event Center for these 3 weeks and fucking magic.

Right. Totally.

And you could never do it. So if you start there on your journey and then stair-step up, you will like, you'll be in, you'll be in the positive mode, the surplus mode, the credit versus debit mode. And even if you have some bumps down the road, it won't stop you, right? It won't stop you. So thanks.

Amazing. Thank you so much for stopping by.

Appreciate it.

Thanks for putting this great event on, Jason.