Claudeforce vs. Replit vs. Lovable. Our Agents Pick Our Vendors Now. Who Wins?

Okay, everybody. Welcome back to your favorite podcast and show, The Agents. This is an exciting one. One of the things that is hot is headless, but what the hell does headless mean? Well, this week Amelia connected the same Salesforce org to 3 different AI tools. Claude Force just out, Replit, and Lovable. Same data, same questions back to back. Out of the box, they seem almost the same product, but first, the bills weren't the same from $2 to $16 and more. And most interestingly, when we asked them for the top 5 ideas for our business, we got 3 different lists.

We did. And the agents are picking our vendors now. So Jason asked the Connect agent where to host and it chose a company he'd never heard of. And I asked 10K what to buy for AEO and it told me to build it. Today, the Salesforce head-to-head, what it takes to sell to an agent, Astra versus Fable, and a demo of the prospector we built ourselves.

Yep. I'm Jason here with Amelia. 3 humans, 21+ agents. Let's go.

Let's go. Okay. Should we start with our bake-off that we did of Claude Force versus Replit versus Lovable?

Yes, but let's step back for a minute. I didn't even know you were doing this. We've been running Salesforce so-called headless for the better part of a year. That's, that, that is the core of our $10K AI VP revenue. Why did you do this? I think you did this in, in, in part because Claude Force is coming out. It was a big push from Salesforce to access your Salesforce data through MCP. But what was the thinking and what did you build? And then show us what we learned.

You're listening to the official SaaStr AI Podcast, brought to you by— today's episode is brought to you by StackAdapt, the AI-powered marketing platform that allows you to plan, execute, and analyze both owned and paid media campaigns from a single platform. As the world of advertising continues to evolve, catch up on StackAdapt's Innovation Forum event to hear from its co-founders, OpenAI, and other industry leaders about the shifts reshaping digital advertising, the AI innovations driving them, and what they mean for Watch it for free at go.stackadapt.com/forum. 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-in privacy from day one. Get more at northwestregisteredagent.com/sasterfree. That's northwestregisteredagent.com/sasterfree. S-A-A-S-T-R-F-R-E-E. Yeah, so my thinking was I, I, I slash we slash Saastr got into the Cloudforce beta. So they announced it at Dreamforce. Beta is live with select customers, and I know other customers of Salesforce who have asked to be in the beta and, and they're still on the waitlist. So it was a gift to us to try it before everybody else can, and I feel like it's, it's great for this community to share our learnings. Because I learned a lot actually. I, I did it as a bake-off just to see, okay, benchmarking. We've had $10K now built on Replit using Salesforce headless for over a year. So that's not really a fair comparison. So I actually took it out of this comparison. I can bring it up at the end to show you guys the differences between a headless Salesforce agent that's had a year of iterations, right? Versus I wanted to do a true kind of benchmarking out of the box of, okay, This is Cloudforce. It is still in beta, but how does it work? And then these are the outputs you can get using Replit's built-in Salesforce connector, Lovable's built-in Salesforce connector, and then just versus classic Salesforce Sales. So I wanted to do a little benchmarking to show folks what it's like and also just to learn, right? We've been using 10K Headless with Salesforce for the last year, but we, I spend my whole day with 10K. It's not really a fair comparison. And so I thought I would do a fresh start on each of the 3 platforms and just see what happens.

I got it. Yeah. I mean, we chronicle our journey with agents in 10K, but the reality is we're hundreds of hours into building 10K, right? And obviously you rebuilt a version of it at SaaStr AI Annual. People can watch that session. We obviously would save a lot of time starting from scratch, like any project, but we are hundreds of hours in. So in a sense, not only is this headless Salesforce, But it's okay if you want to build your own type of 10K or your own headless instance. Let's try it 3 ways: Podforce, Replin, Lovable, and see as of today with the current connectors what it can do, right?

Yep. So let me start sharing my screen and I'll show you guys what the bake-off looks like. Okay, so let me walk you guys through the bake-off. So this is Salesforce as you guys know it today. This is just My pipeline. Keep in mind, I'm just a part-time seller on the SaaStr sales team. So you're going to see here, this is just my pipeline. David has most of our pipeline, but this is just classic Salesforce, right? You can see it's literally, I've never actually configured the seller home. This is just how it is. It's got, you know, my total pipeline, how many accounts I have, how many, I don't have very many contacts in here apparently that are my owned contacts. But you could see this is, you know, this will look familiar to most of you. This is just classic Salesforce out of the box. This is my seller dashboard. Okay, so now the 3 comparisons to, again, just out of the box, and I'll show you guys what I did for— I'll show you each of the 3 first, and then I'll explain the methodology behind it. So this is Claude Force. This is Claude's new connector to Salesforce. There is a Salesforce connector. If you try and go in Claude right now and connect it, it will tell you it's in beta. And then you can like request access. So anybody can do that now if they want to try and get in early before it goes GA.

Right now, if I go into Claude, there's no way I can directly connect to Salesforce or only Claude Force is in beta?

There's no way that I know of to connect to Salesforce outside of Claude Force.

I got it. Okay. Yeah. So it's the, it's the first way. I mean, we're calling it Claude Force, but at a more basic level, it's the first way to use Claude to connect to Salesforce.

Correct.

Plus it's a bunch of prebuilt skills so that they've done some of the work for you. So that some things you might want to do on the Salesforce data, you don't have to figure out from scratch, right?

Exactly. So I'll show you guys a couple of things here. So it's right here on the left-hand side. It's showing all the skills here in the session. I think there's about 35 prebuilt skills it has here of things, you know, it's suggesting so that if you put your whole seller's team on it, which is going to be bigger than ours, you kind of have a roadmap here of what you can do with it if you're, if you're used to seeing it this way versus seeing it this way. Right. Which is A little bit different. You can see there's different info it's already giving me versus just classic Salesforce in the UI. Okay. So, so this is Claude Force. Like I said, you can, you'll see it in the, in the Claude connector, but you'll see it says beta and then it'll take you to the page where you have to request beta access for now until it's GA. So, so that's Claude Force. I'll show you guys the other 2 and then I'll, I'll walk you through the differences cuz I think it's similar yet different. So. So that's Claude Force. This is Replit. This is headless Salesforce in Replit. Again, out of the box. I will say I gave each of the 3 platforms, so Claude Force, Replit, and Lovable, the same prompts. The first prompt I did was just give me a dashboard and give me an account map. So they each had that literally word for word. I copy pasted the same prompt into each of these 3. Instances. So this one's Replit, looks very similar. And then I'll take you—

There's no dashboards or anything here. There's no, there's no, there's no visual representation of it. Oh, there it is. It's down here.

Yeah. Yeah. It's down here. It actually, it's interesting you call that out because it forced ranked the top 5 things to do at the top and then put the dashboard more at the bottom.

What to do, we thought it was more important than a static dashboard, which I guess—

Exactly. Which is interesting. And then I'll show you guys Lovable. It's interesting because Lovable is the only one That I didn't ask it to. I gave it the same prompt. I said, use your Salesforce connector, give me a dashboard, and give me an account map. It built in a login. It's the only one that built in a login. I don't know if that means it's slightly more secure than the other ones, but it's just interesting to me that out of the box, I didn't ask Lovable to make me a login. It just made me a login. It decided to call it Ledger, and then this is what it looks like logged in. So This is a lot more like newspaper style, which is interesting. The other 2 look a little bit more vibe-coded. This one's more like designer-friendly. But you know, it's, there's the big headline here. You've got like $5 million of open pipeline. Your win rate's pretty good. I don't think the other ones may not have win rate as close to the top. Average deal size, and then top 5 things to do are also above kind of the rest of the dashboard things here. And then it's also giving us our top accounts, which the other ones don't have. So already you can see there's some similarities and differences too. Yeah.

Yeah, it is interesting out of the box. If you're visual like I am, if it helps you, the, the Level 1's much, much more consumable, right? It, it, it's all, it's, it's the best summary. It figured out the best 3 metrics, right?

Yeah.

Figured out the best analytics that I had that, that would matter to us, right? Everyone's different. These are the matters to us.

And it came with a login that just worked. It like, yeah, I didn't have to do any weird clerk stuff. Like it, you can just make an account and then Log into Salesforce as a possibility.

That's probably, my guess is that's probably because unlike the other 2, and this wouldn't make sense for Cloudforce, it assumed you wanted to build an app out of it.

Right.

Right. For sure. It didn't yet assume that, hey, you want to publish this externally. Label will skip to say, hey, Amelia's going to publish it for her team, so let's have a login, right? Yep. So, but yeah, that is, that is helpful. It is much closer to an app, getting to go on an app versus a connector, right? Out of the box.

Yeah. Interesting pieces here. And I'll hang out on the Replit one for now for fun. Actually, maybe I'll hang out on Claude for us cuz maybe you guys haven't seen that as much as Replit or Lovable. Interesting things here of one, how they're connected. So both Replit and Lovable use the API. So when it comes to cost, and this is where I'll be transparent, like nobody knows what the cost of Of the new Salesforce pricing for like MCP or API calls is going to be. I don't know it. Like, and we know people at Salesforce, we don't know it yet, right? Like they're, they're figuring it out of what they're going to go GA with. So I'll, I'll just give you guys the cost on like the platform side that I did building this little bake-off versus like, you know, we, I can't share what we don't know yet. So I don't know yet what Salesforce is going to charge for API calls versus MCP, but both RepLit and Lovable. Yeah. are using API and then Cloudforce is actually using MCP.

Right. Yeah. Do we know, is Salesforce going to charge you extra to use Cloudforce for your own data? Do we know? Have they said that? Or just for these external APIs like Reploitable? Do we know?

We don't know yet. All right.

We'll leave that for a future show. I know I would not be excited if on top of what we pay for Salesforce, I had to pay for Cloudforce too, just as a different way to see my data. I would be— I'd be frustrated, but let's, let's take away from that till we know what's actually going to happen. But okay.

Yeah, for sure.

I totally agree. The other 2 are using the basic Salesforce API, right?

Yep.

Yeah. Which is what we've done on 10K as well.

Exactly. So then coming into what the cost for each of these. So each, for each of the 3 of these, I told it again, just to build me a dashboard and build an account map. Then I gave it one more prompt, which was give me the top 5 things we should do. And that, and then I left it at that. So. Just in terms of costs on these platforms, just to see what you guys see now, Replit was about $11.81. Okay, so I'll show that one. So Replit was, this was about $12 to get here.

Okay.

Versus classic Salesforce. LovePulse, only $2.33. It's

interesting.

Interesting.

You know, in all fairness, did you pick the model in Replit?

I, I start, yes. So—

You might've picked Opus or something, right?

It's funny. I started it on free just to see how far it would get.

Yeah.

And then it was like, oh, you should finish this build in, in max on Opus. So then I did switch to that, but it, it did suggest it as the agent.

Yeah, it is. I don't want you distracted. We're not yet doing a detailed quality comparison, but I think Lovable picked more cost-effective models to build this for $2 is the, probably the reason, right? But in any event, that's the answer. $15 versus $2.

Yeah. And in Lovable, I will say I asked it what model it used and it said it couldn't give me a straight answer of which one model it used. And to be fair, you can chat, but you can't pick models here like the way you can in Replit.

Yeah.

You know, if you're in the actual chat, you can just come here and then pick what models you wanna use, right? So there is a difference.

You and I are both very biased in Replit and overpaying for our models. We wanna use the frontier models. So it may color, color the cost, but it's great data.

Yeah. So yeah, that's one thing about like truly building for internal on, that's why Replit is the house of 10K and remains the house of 10Ks, cuz I can pick the models cuz I do get picky. We can chat later why I actually switched to Astra, but I am picky about which models it uses. So there's that. And then Claude Force was about $15 just in Claude usage.

Got it. Although that was, in all fairness, and keep going, that, that's useful. That is included in your Claude Max plan. So you

didn't pay an

extra $15, but that's what it estimated the costs were.

Yep. I asked the agent, I said, hey, how much do you think I spent on this session out of all my, like, I'm right now I'm on a Max plan, which is $200 a month. And it said, I think you've used about $15, $20 of it so far.

Got it.

Okay. Yeah. So I think interesting. Again, I think we can save it for a later episode once we know actual costs. All in. I just wanted to share just costs we have out of the box. So if you're thinking about hooking up your Salesforce to an agent today, kind of what it— one, what it looks like, two, what it costs just for these platforms out of the box, right? Everything else aside that the Salesforce contract, we're not factoring into that cost. Everyone's cost is going to be different there. But that aside, this is what we paid just to get to this. instance where they look very similar. So I think that was a big learning for me in doing this bake-off, that out of the gate with the same prompts, I mean, Lovable's looks the most design-wise differently, but they function fairly similarly. Like they all do what they're supposed to. They all have a dashboard. They're all giving me the top accounts. They're all giving me now like the top 5 things I should do with our accounts. Those are different, so we should go through them next. But you can see like the similarities are more there than the differences. So let's go into— oh, so a couple things actually I wanna talk about in this as you guys kind of decide, 'cause I don't think a lot of folks yet have put out what Claude Force looks like. Read/write access. And so I asked each of these agents the same set of questions just to see if they would give me baseline the same answers. And for the most part they did. The top 5 ideas are different. And then I also asked each one of them, do you have read access, write access, or do you have both out of the gate? And so out of the box, Claude Force has both, but it's been restricting itself while it's in the beta to read only. I think that's wise. It has a write tool where you can write back to it. It's also available, but I think they built it that way because when it goes GA, they probably just want the admins to be able to write back and not like 1,000 sellers to be able to write back to. Salesforce data all at the same time. Replit has both same kind of idea, read-only by default, and it will confirm with you in writing before it takes any actions to write back to your data. And then same story on Lovable. It has both read and write by default, but it will also confirm with you in writing in the agent before it writes any data back to Salesforce. But again, worth noting, Lovable is the only one that actually built a login for it. So interesting there. And then, so let's see what each of them said we should do, because I feel like this is where they splinter more in how different they are other than the design of Lovable. So in Claude Force, just on our top 5 actions based on our open pipeline, it wants us to close, you know, any stage 4 deals we have. Okay. That's a, that's a good, you know, if you're a seller coming in, that's a good to do at the start of your day. It wants us to engage more with our larger diamond renewals. So that's like Google Cloud that it has mentioned here that it's sitting in like our Q4 close date. It wants us to do what's called a blitz in Claude for us. So it's saying you've got a bunch of deals that total over $1 million in stage 2, push to get them to stage 3 or close. That's a good, that's a good reminder to do that. You have some bigger deals here that you should escalate. Since they've been kind of dragging, and then funny enough, it surfaced that we should push more with Salesforce to get a better timeline and proposal on the books. So those were Claude Force's top five recommendations. Looking at Replit, it's got slightly different ones. So it force ranked them a little bit differently. It it does surface the same deals that it wants us to try and close in October. It also kind of has a version of a blitz here where it says you should you know get some of these deals that are in. In stage 3 and don't have a next step formally recorded in Salesforce to have a next step in the books. It also kind of brought up Google Cloud, and then it actually brought up like negotiation terms. This was interesting. I was like, the others brought this up in terms of the deal where it wants to see if there's any little blockers on decision dates in renewal terms. For examples, it wants to look at renewal terms in terms of, you know, payment terms Anything in the negotiation that might be close, that might be a hindrance in closing this deal. And then 5, it's putting at number 5, which Claude first put at number 1, to try and get the ones that are in stage 4 actually over the line. So interesting how they force rank them and also surface different things. And then finally on Lovable, just our top 5 things. The number 1 thing it actually said we should do, which is different from all the others that didn't even surface this, is we should rescue the overdue deals. And I agree. I think that that is probably the number 1 thing you want to do. Intuitively, when you come into Salesforce, it's rescue deals that are past their close date or past their forecasted close date, just to keep your pipeline more honest, which it calls out. I do like that. Get meetings on the books for 5 of those deals that are in that working stage and overdue. Okay. Unpark 7 renewals. That's pretty similar, but it didn't really force rank, you know, the diamonds versus the other renewals. It just said all your renewals you should unpark. There's a few deals that don't have dollar amounts yet. That need work. And then this one also agreed on Google Cloud getting it over the line. So some similarities and differences there on what it thinks we should do with this data, which I think is just based on whatever model is underlying, right? So Lovable AI, I don't know what model it's using here, but it's obviously serving something slightly different than, you know, using Claude in Replit directly and then just using Claude for us directly.

How do you rank these 3? You might need more than one criteria, criterion, but how do you rank them?

Yeah, honestly, I would say for ease of use, both Replit and Lovable were super easy. Like you just ask the agent, connect me to Salesforce, and they have built-in connectors, right? So right now, those 3—

Ease of build, which one's easier? Rank them ease of build.

Ease of build, I would say Lovable was the cheapest, but I think it's tied with Replit for how easy it was because they both have connectors. And so you just tell the agent, Even if you go to sign up for either of these platforms today, you can— the first prompt you can put in is connect to Salesforce, and they both have a built-in connector already today, right? Whereas Cloudforce is still in beta. You have to request access. So for today, the ease of build is not as good as the other 2.

Okay. But that's with an asterisk because Cloudforce—

With an asterisk.

Yeah, right. Okay. Ease of build, quality, quality of data and application.

So I would say they're all tied. So I did ask each of them the same questions, like I asked them who's the account owner on Fin, and they all answered correctly. Which was me. And then I asked them, you know, how many deals do I have versus David? And you can see they serve like this one, Claude 4 surfaced it in the dashboard. They all answered that correctly too. Where they start to differ is again, if you give the agent actions, like give me the top 5 ideas or what should I do today? Like if you're waking up as a seller and you're like, hey, what are the things I should absolutely make sure I get done today? That's where they start to differ. I think out of the box, they're all very similar, actually more than they are different.

And, uh, what about UX/UI? Which one's the most usable?

I mean, I like this one. I will say I had to give Replit more instructions for it to get to this UI. So then the other 2, those other 2 were pretty much out of the box. Like the first one it gave me was illegible a little. So I had to give it a little bit more prompting than I gave the other 2 to get it to this like dark blue and green that you see it has now. I think again, depends on your audience too, right? Like if I was building this to share with you out of the box as a dashboard, I would say the lovable one. wins on design. It's like this newspaper style, it's easy to read the win rate. Everybody understands like the top 3 things that surfaced. That's all you need is that top screenshot if you see nothing else. And I feel like that's super easy to understand. Whereas the other 2, you have to scroll down a bit to get to the actual open pipeline. It's surfacing different things in the pipeline than, you know, Claude Force and Replit art. This is in dark mode. Fine. So I would say in terms of design, I would go Lovable, Cloudforce, and then Replit.

Interesting. For me, and then keep going. We have very different roles. We run it together. We have very different roles at SaaStr, right?

Yep.

Lovable, the Lovable presentation's the only one I would use.

Oh,

now I don't even use 10-K. Now I go into 10-K for questions. Like this morning I wanted some metrics on year over year data.

Yeah.

So I went into 10-K. It's great. We've architected it. It's great, but it is so dense and thick, 10-K, that I will never go to 10-K just in general. I'll only go for answers, but I go for a lot of answers. You did a great job. The Level 1, at least out of the box, man, this is for me as a stakeholder, but not in it the way you, David, and others are. This is what I need to know. What's on the table? What's our win rate? What's our average deal size? What are the top things we should really do? It's really structured in a way. So it's interesting that it's the most valuable to me. And then my, my guess is, I don't think you can answer this one because we didn't go this far, but the other interesting one I think about is which is the best at multiplayer out of the box? See, Lovable already has a login. Right. Okay. Now that's probably as far as it went, right? But it is one step further on an app everyone on your team can use. I don't think Claude Force by definition is going to be multiplayer because it's great that I don't think it can be, it can be an app. Like we'll see when it's like, what if you have, if, if you and I were both using Claude Force, right?

Yeah.

We're gonna have different, maybe we can share this dashboard across us. It does share, but I don't think it's, is it going to be as intuitively multiplayer as an app, right? TBD. TBD. Yeah.

TBD.

So I like Lovable cuz it's best for me and because it's the furthest along on multiplayer and, and the way we build a lot of our agents and rep But which we love, they're very much not external multiplayer agents for the most part. They're internal ones, right? And so I think it's a theme for the future, which is as folks use agents like we use them across larger teams, you need this multiplayer external use case that all different stakeholders can use, right? It's a layer of complexity we don't have to deal with on a daily basis. So I, I'm, we're on, we're, we, we love all 3 of them. They're all partners of ours, but I, for me, uh, Lovable's a big surprise.

It was surprising for me.

It's the only one I would use. It's the only one I would use.

It was surprising for me. I frankly, I don't use Lovable as often, right?

Like,

yeah, 10K is built in Replit, so I spend a lot of time on Replit and I use Claude Cowork more than I use Claude.

Yeah. Anything else? 6 to 10 hours a day.

Yeah. So it's, it's interesting because I did it just to do it as like a benchmark of, obviously we're still, we love the folks at Lovable. And I wanted to make it more of a fair comparison of— I know not everybody's on Replit, and I know there's a lot of people on Lovable already. So to do a true out-of-the-box comparison, I just wanted to see what the 3 were like. Again, asterisking that Claude 4 is in beta. And you can see here, like, you know, it's, it's loaded in these skills here, which again, if I had to— I know that everybody at Salesforce is on Claude 4. So that's— I don't know how many of their 80,000 people are sellers, right? But that's Maybe half the org is already on Quadforce, but I, I get the sense that it's not meant to be multiplayer. It's meant to be like, if you're David and you're logging into Salesforce and you're not necessarily me, like, do you wanna look at this, which is classic, or you wanna look at this?

It's much like, let's stop for a minute. Yeah. People want you to have their own Quadforce and using it instead of logging into, to, to grandpa's Salesforce. I mean, it's much, as long as, I mean, So much better. And you can talk to it and you can work with it and it'll have analytics and it'll have LLM built into it. I mean, why would anyone use native Salesforce when you could use Claude for us, right? Or any of these 3. Exactly. I mean, that was our early, you know, our early learning. You know, we never log into Salesforce itself anymore. Why would you?

No, I don't.

It is very disruptive to all B2B apps because you, since these were built from scratch, right? In, in less than an hour each, right? For, for 3, between 2 and 3 hours.

20 minutes, maybe 20 minutes each.

You'll never log into these, your B2B apps ever again. You may, you may or may not MCP into them. You may need a package of skills and help like Cloudforce, but good God, it, this is, all 3 of them are an order of magnitude better than Salesforce standalone, right?

Yeah, it's funny. When I started doing this benchmark, if you will, it's a light benchmark slash bake-off. I was like, actually, I need to compare it to regular Salesforce. Because first I was like, okay, I'm just going to compare them against one another for this podcast. And then I was like, no, actually, We've been using agents for so long that the benchmark is this. This is the benchmark. This is just normal Salesforce loaded up in the morning, which I've forgot about cuz I haven't looked at it in a year.

Yeah.

Versus this is Clockwork outta the box. It's already 10 times better.

Yeah.

And then these are Replit and Lovable. And I will say the, the case for, for Replit and Lovable is it can connect to other apps though, right? So it's just how far do you wanna go on this journey? Cuz for me, okay, like It's not a fair comparison to look at 10K because this has had a year, man. It's connected to everything we have in Salesforce. It's connected to Bill. It's connected to our entire stack. There's 20 different things here that it's hooked up into. Like, it's just not a fair comparison to compare 10K, which has had all these tools for a year versus these 3 things now.

Well, you know, it's interesting. It is a fair comparison in the sense that, okay, let me step back. My learning is just watching you, right? As Cloudforce and similar, similar things like Cloudforce, right? MCP plus skills abstracting away the complexity. More and more of them go live. It's gonna be much better than the native B2B app, right?

Yeah.

And, and, and so, and when we look at the 3 for, for single-player use cases, I'm gonna give my own opinions after, but they're roughly equivalent, really roughly equivalent, right? Yep. Um, I think there's some significant UI/UX and other differences. But they're roughly equivalent, right? They're all much better than what you have. So I think where we may be in a couple of months as this matures is if you just want to use Salesforce primarily or HubSpot or whatever it is, use Cloudforce, right? Or HubSpot Force or whatever it is. It's going to be more native. It will ultimately be like, right now you said Cloudforce came in 3 to hook up, but I remember when we first hooked Salesforce up to Replit, it took work, right? And so Cloudforce is going to be easier. It's when it's—

It's going to be easier for sure.

It's going to be easier. There's not going to be token expirations, token— Yeah. That was a— just like anything, you're gonna want to use Cloudforce because it, it is out of the box for this use case as good headless V1 as Lovable and Rebel it is. But to this slide, instantly when we went headless, we realized we could hook up QuickBooks, we could hook up Bill, we could hook up Agentforce, we could hook up all these other applications.

Yeah.

Will you, will you want to use the Cloudforce or its peers at other applications if you want multiple applications, and that's where I would instantly skip Cloudforce for Replit or Lovable. As soon as I add anything that wasn't natively integrated inside of Salesforce, right? Then, then Cloudforce, Cloudforce for us, as cool as it is, is basically useless because it can't reper— it can't incorporate all of these other integrations inside of it like Replit and Lovable can, right?

I think I wholeheartedly agree. I think they each have their time and place, right? Like the biggest difference for me in the end, after doing this bake-off and reflecting on it and putting it all together for you guys was Replit and Lovable can connect to other things. And obviously we've connected $10K in Replit to 20 different apps. And so that's where I think if you are more of the SMB startup or you are in the, you're in this new batch of folks that are in SaaS and B2B with 3 people, maybe less, maybe more, but you're doing millions of dollars, then you may not wanna use Claude for us because it's too limiting. But if you're Salesforce or if you're an enterprise and you're a huge company and you don't want your sellers automatically writing everything back to Salesforce, you want some more guardrails. That's where Replit and Lovable are harder to guardrail, logins aside, right? Like those are just going to be harder to guardrail. They're creating so much data instantaneously for all your sellers that you may pick Quadforce out of the box because you purposely don't want to connect it to other apps.

Yeah, for sure. Yeah. The, and certainly the more SMB you are, the more flexibility you may have here, right?

100%.

You know, one last point and then tell me, this is so good, Amelia. Thank you. I, one last, this isn't that interesting, but what I can see when I see the 3 side by side is something very similar I see in our AI SDRs and BDRs, the, the Monaco Artisan qualified agent force at all. It's very interesting that at least today, when it's still, it's still early in the grand scheme of things, these agents are, and platforms very much are expressions of the founders and people who built them.

Yes.

So like Replit, you know, we love the team there. We love Amjad and Kodeel. It's a nerdy company in a good way. It is a developer-focused company at its roots. For sure. It's an idea at its roots. And the output is the nerdiest. This is very nerdy output. All black. It has the right ideas. The top 5 things to do at the top.

It's more technical.

It doesn't have dashboards or other things. Yeah. Because it doesn't think about it, right? Quadforce looks like Salesforce on steroids. Quadforce looks like the Salesforce we always wanted. But it never actually worked this way. This is what we— if you think about if the first time you ever used Salesforce, if you ever had a CRM, this is what you wanted. You wanted, what do I do? What's hot? Where it is simplified. It fits all in one screen. I know what's happening. There's some AI telling me what to do. This is what we thought it was, not like what Amelia's screen is, right? And then the lovable one, if I played with it, I may find it's more limited than the other 2. It may not, but with Anton's design background, I'm not, I actually don't think this is beautifully designed. I think this is just another set of Artifacts. I don't think that, I mean, there'll be 10,000 lovable apps that look like a newspaper. Don't color me so impressed with the, you, with the design, but in, in terms of the UX, and it's very usable for me, right? I don't think, I don't know if it's designed, but it's, it, it has a designer's eye. What matters, right? This is what designers spend a lot of time on that we're pretty bad when we build apps, you and I, right? Mm-hmm. We don't intuitively go from the top of the page to the bottom of the page in the right way. We don't use the right sizes and fonts. And colors. This is what it's sort of doing, right? It's picked the 3 most important metrics. It's formatting them with the roughly the right weights and ratio. And you can say this is just built out of some sort of artifacts out of Claude or something, but, but it is very much to me looks like Anton. And then I see Amjad and then I don't know who I'm seeing at Salesforce, but I'm seeing some Salesforce nerds that know the app effing cold and are taking a fresh perspective. What can I do? With, with Claude Force that I can't do with Salesforce. And so it's interesting. Then, and all of all, but when we looked at the Monaco Artisan Agent Force, you know, Monaco built by Sam Blonde, our friend, but ex-CRO of Brex. He built great human outbound sales teams and that's what it's, it looks like what he did at Brex. And even in the old days at Adobe Sign with me, right? Artisan is Jasper's view, right? Yeah. In the early days, you don't need humans. Now they backed away from, from a little bit of that. But that's it. That's what they've built, right? On and on and on. You can see it with Auricel and Lightfield, right? These are, these are, and so these apps in a way, it kind of behooves you to do a little more diligence than you might want to do because these apps have very different perspectives from where they're built. And I think the roadmaps for the next 12 to 24 months are going to continue. Then the footprints will get so large you won't tell, but they're going to continue to be from the perspective. I hate the term taste because That's not really what it is, but really from the perspective of the founders, 'cause there's so many places you could go, right?

Yeah. And I think the, each of the prompts I gave the 3 in the benchmarking/bake-off, again, I gave it maybe 10 prompts total between, I asked it like the same list of questions just to make, to see if all the answers would be marginally the same. They were effectively all the same, right? Like it's giving you the data. The same way, which is that's what you want, right? Like you want the connector on the backend to work. So they all work. I think it's just a matter of, again, how secure do you want it to be? How much do you want people to read and write back to it? If cost is important to you, maybe just use Lovable. If it's not as important to you, or, or maybe you have different use cases connecting it to different apps, then, then Replit might win, right? Because obviously we've used 10K for the last year. We've hooked it up to 20 different apps now. And it is stable and it is secure and, and we've hooked up more and more and more to it. Right. But you do go down this rat hole of, we have hooked up 21 different things to it at one point, 30 different things to 10K. So I think it just depends on your goals and journey for it. But out of the box, the 3 of them are infinitely better than just classic using Salesforce Classic and logging in.

Yeah. If you have anything else, we move on. But I will say, listen, I love all our friends at Salesforce. I appreciate them. When I saw the Claude Force demos at Dreamforce, I was unimpressed. Okay. I certainly didn't say this out loud. We're just saying this to a few 10,000 people right now. It's just, I just, I was like, so what? Hooray. I can query Salesforce in Claude. I didn't care. And the demos, while, while they were thoughtfully done, getting a demo with some text back on my pipeline was not impressive. What I just saw is 100 times more impressive than what I saw at Dreamforce. What I saw was reimagining Salesforce for the agentic era.

Yep.

Right. I can see it in real time. I can get agent-based feedback. I can get agent-based analysis. Can you use some of that inside the Salesforce app? I guess it's beyond me or my interest, right? So this is like 100 times more powerful than what I saw at Dreamforce, even though I'm not sure, even though they all kind of got a tie. I think let's come, you know, as I think most folks haven't seen Claude Force, I give it a lot more credit honestly than I did before you built this, right? I think it is disruptive. I suspect like any product, it'll need to continue to evolve. I'm not sure all the skills out of the box are completely intuitive for folks. It'll take a few iterations, but I, I don't know why anyone would not use Claude Force, right? Yep. I, I, it is, it's more disruptive than everyone. So let's, as we get closer to GA and, and maybe we, we'll check in on a month or something on this and, and see where we're at on this stuff, right? So.

Yeah, I'll keep it going. I'll keep all 3 of those just live. I'll continue to iterate. I'll, I'll keep 10K going in the background as our main GTM agent, but hopefully this is helpful for folks just to see it for real with real data.

Cool. The next one, maybe Amelia can bring up a couple. I've been, you know, we do it, we do the other podcast. I do a podcast with, with Harry Nourian, 20VC, uh, and others. And I've been talking a lot about agents picking apps, agents picking vendors. And it's been interesting to me that so many folks, including guests we've had on the show and leaders don't, I don't think they get it yet. Yeah. I don't think they get it yet. And I think this is so important for anyone in B2B, which is we're confusing sort of a 2025 view of so-called AEO and GEO, which is get high on the list, get in the number 1 or 2 vendor when you ask Claude or ChatGPT what to buy. When a human goes to ChatGPT or Claude and says, What CRM should I buy? You want to be number 1, or at least you want to be in the top 3, right? It might say, hey, Salesforce is the top choice. HubSpot is used by a lot of SMB, but the up-and-comer is Monaco. Like, you want to be in that list because you will get massive leads if a human typing it in. And there are a number of folks that have done that. Profound raised like $1.5 billion. I think that's where they start. HubSpot had to get acquired a company to use. I'm not so interested in that stuff. For a whole bunch of reasons we can talk about, but What I think folks are missing is that's how humans use AI to pick. You need to win in the, in how agents pick applications. And more and more, and if you look at, I think Vercel said the other day that like 60% of their new apps are built by agents. It's true in Supabase, in sort of developer-focused apps. They're obviously much further along here than others. But the, I don't think the team got it on 20VC. So I don't know if you can put this one up on the screen, but I said, So we're building this app, SaaStr Connect, which is an agentic recruiter for founders. It's very cool. It's very much my personality, which maybe we could chat about on a different podcast. It's very different than any other agentic recruiting app out there because it's designed for founders that need help finding the best passive candidates that aren't in market for real. And so it's built very differently. And it's built on Replit and it's great, but I was wondering if we wanted to go onto another platform beyond Replit at scale. What should we use? And so I asked live when we were doing the 20VC podcast, and I asked the SaaStr Connect agent, and it's very particular. What should we use? He said, and, and it, the thing is about when you ask Claude in general, what app I should pick for CRM, right? What context does it have? It has a little bit that's in that, that chat itself, right? A little bit of context. It has a token amount that's in your memory. Memory's very small, right? And maybe it will search through other chats, right? In, in the chats. And maybe if you're in a project, it can actually search somewhat deep in a project, but that's about all it knows, right? If that, when it makes a recommendation. If you ask an agent, it has all the context of whatever it's in, in the application. So it is very opinionated. And we'll show you another example. Very opinionated. Here's what you need. An always-on web server for the site and the API. And always-on background worker, because we have all these Connect agents that are out finding folks to work for you 24 hours a day. This isn't a spool it up, spool it down kind of thing. These agents are going to be running 24 hours and scouring the best people from the top 30,000 B2B executives who are good for your startups. It's constantly running long-running jobs. Sometimes these take minutes to— it doesn't just find people and make a crappy match like LinkedIn. It can take 10 minutes to analyze 25,000 people and figure out who's the best for a new role for you. So long-running jobs. Mm-hmm. Logs, delays without downtime. It said Render is my pick. Okay. Now I said on the 20 VC pod, I'd never heard of Render. It's not quite true. I follow them on Twitter and others, but I, I admit I didn't— that Render is a relatively hot app. Render and Railway, its second choice, are 2 pretty hot ones in addition to Vercel to hosting an app. But I didn't know why. I hadn't used Render myself, and I gotta be honest, it never would've occurred to me to use Render for Connect. But look what it says, if you can see it. My pick, my pick. My

pick, yeah.

Built for exactly this shape. Web service plus background worker plus scheduled jobs, health checks, zero downtime delays, longer log retention, choice of machine size. Railway, which may be bigger, I'm not sure. It's pretty hot though. Very easy, pay-to-use, handles it, but a bit less mature for heavy production. This didn't say demo them both. This just didn't say do a 3-way bake-off, right? And then interesting, then FlyIO, I don't know them. It said AWS and Google Cloud, not worth it. Overkill. I know that's true. And then interestingly, I actually didn't really know this. It said Vercel, not suitable for minutes-long jobs and background workers. I'm not sure that's fully true. I mean, we've used Vercel in other apps and it's pretty powerful and very cheap platform. Even Muse used it to scale up Muse. Actually, they used Vercel as the backend.

Oh, I didn't know that.

Very powerful. But I thought it worked better for always-on projects with background workers. And if anyone from Vercel's watching, Sure. In the Connects, correct me. I'm not even saying this is correct, but the more important thing is very opinionated. Render got the customer. Render got the lead and the customer. You gotta win this. And this is very different than a generic ChatGPT search, right? Which frankly probably will tell you to use Vercel 'cause it's the biggest brand, right? It'll probably tell you to use Vercel and that another lead for Vercel, right? And again, we love Vercel, but man, And so I was doing this with a founder, uh, that, that we're investors in a SaaStr fund and founder didn't even know what I was talking about fully, even though they're, uh, they're an AI-first company. And I just set up this thing where I went to 10 different agents and I all asked them which, which vendor in the space they should choose. Now, lucky for him, they all chose him.

Oh, wow.

So I'm like, okay, your job tomorrow, in my opinion, I don't run the company. You gotta have someone in your marketing team monitoring this weekly, make sure you stay number one. Okay? Because the agents are gonna change. The agents are gonna change and people don't get this. This isn't a random list out of ChatGPT. This is the agents say use Render. It's the best. And it's saying it definitively. Which one would you use, Amelia? You're not an expert in this space. Which one would you use?

Oh, I'm, I'm not an expert. I probably—

Right. So what would you use when you got this? You're not an expert, so what would you use?

Oh, I would use Render because that's what the agent suggested.

You wouldn't, you wouldn't even qualify it like You're not an engineer, you're not a developer. You wouldn't even go down this list. You wouldn't even test Railway or Flyer.

No, it's the top one that the agent would test.

You would just use Render instantly, right?

Instant.

And it, um, and you know, in the early days when we were building 10K in Replit, even before you were doing it, when I was getting this going, I had the opposite. It wouldn't stop recommending SendGrid in the early days before connectors were good. I just couldn't get SendGrid to work in Replit. So I use Resend. We're very early on Resend for agents and we use Resend constantly, right? But it was like a battle with the agent to use Resend. Now, now the agents will all recommend Resend because it's broke, but this is the opposite. You, you gotta win and it's not good enough to buy some B2B app charging you for weekly updates of where you rank in ChatGPT. This is not gonna tell you what the agents are gonna pick. This is not humans asking LLMs what to pick. These are agents making here. And you know what's one step away from this? Would you like me just to use Render? Yes. And then gone. Done. You know the next, and the next, you know what the next step after that in 2027 is? Actually, I've already migrated you to Render. Would you like to see it? It's gonna— Yeah.

Do they ever start actioning it?

Yeah. Ah, right. And so you gotta win this war. And I, I think so many folks, I would say 99% of folks aren't thinking

about

it. Yeah. We were even on 20VC with the Public Company CEO, and I don't think he, he really understood what I was getting at here. And I don't think Rory and Harry did. Now, I love 'em all, but I'm just saying, If you want to win today. Now, this is most folks that are building developer-centric tools mostly already know this, right? They've seen things blow up, right? Apps like Clerk and Resend and others blew up when, when the apps recommended using, they didn't even know it was happening, right? Yeah. But man, you gotta win this war. And then I, I had another one. So this is in SaaStr Connect coming soon to an inbox near you. Other one I was surprised, then I asked which CRM we should use. I don't know if you have this screenshot. So I asked the SaaStr agent running saastr.ai. I was just curious, what CRM should we use? Obviously we're very deep. We're obviously very deep on Salesforce. All our agents run on it. Realistically, we're not, even if we should leave Salesforce, which we shouldn't, we're not going to, we're too invested, right? We're too deep on the platform. But interestingly, this is why it's interesting. And again, we love Salesforce. The agent knows what we're using. The agent knows we're invested in Salesforce. It said, still, for SaaStr overall, I'd shortlist HubSpot and Addeo. Okay.

That's funny.

And it's interesting. Obviously we know the HubSpot team, we know some pros and cons. Addeo, I mean, we know them, never would've occurred to us in a million years to use Addeo. Just for it, it's still, I know they're AI now, they're old. Why do it? Salesforce works fine. We don't care. Right. And then it's smart enough to realize there's a difference between starting fresh and not starting fresh. I'd lead HubSpot for the operating business and Addeo for the SaaStr fund. But I wouldn't migrate your existing Salesforce setup without a clear reason. That, it's super interesting, right? First of all, our, our friends at Salesforce, you got, you gotta get picked.

Yeah.

This is worrisome, right? Yeah. I mean, the agent did think Salesforce is not the right choice for a smaller organization like us, right? Something to reflect on the Salesforce team. But also, if you, God, how many dumb SDR emails do we get if you wanna switch from one vendor to another, right? It said, don't do it without a clear reason, right? So the agents are also making decisions which are different than humans might be about when to switch vendors and when not based on its access to the data, right? So then it goes down just like we saw with the Render one, HubSpot, Addeo, Salesforce, Affinity. Um, Affinity is focused on funds, so it's a good one. It's the only one it recommended for funds, right? A lot of folks use Affinity. It's— they won.

Yeah.

Um, but unlike the other one where it said use Render, It didn't do anything. It said, what do I do for now? Keep Salesforce for revenue operations, evaluate Adeo for the fund, knocked off Affinity, which is a fund-focused CRM with a small pilot, right? Not 2 copies of the database. So there's a lot going on here. You can read it. The most important thing is know where your app stands with as many agents as possible, as many agents with lots of contacts running in different situations, because more and more of them are going to make recommendations. It's not just going to be nerdy agents in Replit and Lovable or Claude Code picking what you should use. This is going to become more and more common. We're going to— our agents are going to make decisions on what we should use. And I was surprised that even though we have all our data at Salesforce, it still said don't move, but it said starting fresh, HubSpot and Adio. Sort of interesting. Not saying that we would, but I'm saying Salesforce marketing team, I think you guys got to step it up on agent-first marketing. It's like I got to come up with an acronym for this because I don't think it's out there. It's not AEO and GEO, guys. That, that, that's your grandmother's, uh, grandpa, grandma and grandma's version of AI here. Yeah. This is agent chosen something, right? Agent-led growth. This is ALG.

ALG. Well, I like that.

And you gotta win. Gotta win because Salesforce, surprisingly, because I actually don't think HubSpot, there's any compelling reason I'm aware of that it would be better for our agents than Salesforce, right? They both have reasonably open platforms, right? But, uh, you gotta win this battle too, not just the battle of ChatGPTs. So those are my, my contributions.

No, agree. I think related, um, and I'll keep it brief. I asked Ken K on this AEO debate, right? What we should use for SaaStr.

Yeah.

So you asked it a different way of like, for what? For AEO.

AEO. Okay.

Yeah. So like, like if the goal for us is to get agents to recommend SaaStr as A lead gen event as a media company, and then also just as an event in general that people should attend. I asked 10K what we should use. It said, quote unquote, there is no single AEO platform I would recommend yet. He said, this is something we can build in the dashboard ourselves. I honestly thought 10K was just gonna recommend Profound. He didn't, he actually didn't recommend it. He said not to use Profound. He said, You can start a Peak AI trial. P-E-E-C-A-I trial.

They're invested in Peak. Yeah. They're like, okay, the tutor profound, at least, at least they say they are, right? I'm not picking any favorites.

I've actually never heard of them until 10K said it.

They just got, they just got a potential lead.

I know. He said, start Peak AI trial in parallel just as a free trial and see what happens to benchmark it against our dashboard. But in the meantime, I'm gonna start building you an AEO dashboard.

Yeah, that's a good point. Yeah. Build it. The agents will, that, that's an extra pro. The agents will, we find this, it shockingly happened to us when for one use case, it recommended not using Calendly, right? When you built your own Calendly, but not for everything necessarily, not for informal meetings. Right. We talked about the 4 here at the end. Yeah. I can build my own AEO/Geo tool. Now I'm sure the guys there would say at Profound and Peak would say, hey, that's a very narrow use of AEO/Geo. If you have a good agent, of course it can. It can do a bunch of, it can go through the, the Claude and OpenAI APIs and, and get a sense of what your answers are to 20 questions, right? That's—

Right.

That's, I built a version of that a couple months ago, and it was better than the startup that HubSpot bought. Um, but they'll probably say we do so much more, right? We do analytics, we do long-term tracking, we do all these things that 10K isn't gonna do. And I'm sure they're, I'm sure they're right. I'm sure they do more, right? But 10K said, no, let me do it. Run a pilot, but let me do it. That's, that's just, that, that's as interesting as the truth, right? Is what the agent, agent said, right?

Well, his pushback as an agent was that Profound and Peak are good for benchmarking, but not necessarily to take actions. Now, I don't know if that's true, 'cause I haven't actually tried Profound or Peak yet. I will now.

Historically it was, but I'm, I'm months outta date. Yeah. Historically it's just a bunch of, hey, here, here are your issues, but what do I do? Nothing.

Right.

I'm not saying that's true today. It's just in the old days it was true.

Right. Yep. And so Tenkey's thinking was, since his— from what he could read in the docs as an agent, it still seemed to be that way. He said you could start peeking parallel just to see where we're ranking and compare it to his benchmarks he builds in the dashboard. But because he's now bringing in his own AI agent visibility for SaaStr into the dashboard and he has 20 to 30 plus other sub-agents hooked into him, he can actually do something about it. And so I think that's why He said for our specific use case, there's not a single AEO platform he would recommend.

Cool. Well, this is— this was a great— you did a great job. What did you want to talk about with Muse and 10-K? Tell me what you want to talk about.

So I think it's interesting. I was at a Salesforce event this week and it's interesting, related to the topic we just kind of wrapped on. So many people try to pitch me their startup, which is fine. It happens when you go to these SF Tech Week events. And I said, listen, I'm not gonna decide. 10K will decide. You can gimme the spiel on your, on your agent, your app, whatever it may be. I'll give it to 10K. If he says he wants to try it or use it, I'll tell you yes. If not, then just know that my agent—

They wanted to be promoted by SaaStr on the show and everything and all that. I mean, it's, it's been everyone we promote gets millions and millions out of it, right? That, but that's what, that was the goal just to get the context. That was their goal, right? Was to get us—

That was their goal. Of course.

Yeah. Uh, then go for it. Right. By all means. Yeah.

So yeah. And go for it. But it's funny, like, I think maybe this happened like with 10 people and yeah, all 10 of them were shocked. They're like, what do you mean the agent's gonna pick? I'm like, well, $10K, he's gonna make the decision. Why would I pick?

Why would I get with it? You're, you're state-of-the-art applications. Of course we're gonna ask $10K.

Yeah. Why would I just pick? It's, oh, can you show it to Jason? What? Well, okay. But like, if it's great, I'll show it to Jason, but it He's not gonna use it every day. Like, he doesn't even have a Salesforce login. Like, it matters more for 10K to see your app than it does maybe even the CEO.

Yeah, it's, it's, it's like the story I, I said of the, of the CEO at the 9 figures in revenue that, that didn't fully understand what I was saying. That, but luckily the, the agents, all, all of our agents recommended it, right?

Yeah.

You gotta figure this stuff out, guys.

Yeah.

If you think, if you think it doesn't make sense that we ask 10K, you're, you know, you're, you're It's not gonna kill you today. It's gonna kill you next year.

Yeah.

It's gonna, it's already the majority of how Vercel and other folks and, and Supabase and folks at the cutting edge of developer-focused stuff get their business. It's gonna be how everybody gets the majority of their business in 24 months. And it's, and if you're not, if you're not, if an agent isn't gonna pick you, you're gonna see degradation in your, in your leads and your, and your market share. It's just gonna degrade because everyone like us won't use you. Now, maybe today we're at the cutting edge, right? We're, we're in 0.1% of Replit according to Replit. So let's say we're not, you know.

Yeah.

But, but that's gonna keep growing. And if it, if Vercel went from like 3% of agent-led instances at the start of the year to like 60 today, that's a lot of change in 10 months. What's gonna happen outside of— it's the same things that happen just more slowly, but everyone's gonna ask their develop— their, their agent, right? There's gonna be some version of Claude Force next year that everybody asks, but hey, Claude Force, should I use this inside my Salesforce instance? And cuz it's connected to Claude, it's gonna have an opinion too, right? But, um, yeah, they didn't get

it. Yeah, it's the new version of Ask Your Developer, right? Is Ask Your Agent.

It really is Ask Your Agent. Yeah, it's good.

Ask Your Agent. Okay, last thing, I think this is fun and we can wrap on this, is, um, I was also sharing with folks at this event that I hooked up Muse to 10K and Replit, and they're like, what do you mean you hooked up your, your Facebook Muse? Yeah, that's your personal assistant to your Replit go-to-market agent. They're like, I didn't know you could do that. Like, I'm just having it, whatever, you know. Optimize my runs or my shopping or whatever they're doing with Muse. And I was like, yeah, because one, there's little risk in that, right? Because 10K has all of our go-to-market data. And so I know there's a trade-off there in hooking them one to the other. But at first, actually, Muse told me it couldn't be done. And I said, there's gotta be a way where you guys can talk to each other.

Mm-hmm.

And it said, okay. You're right. If we make an agent mailbox for us to share, we could just post messages to each other and that's how we'll talk to each other. So now Muse is talking to 10K and every morning for the last few days, 10K will post to their shared little agent chat/mailbox, if you will, a daily brief in the morning. So he'll, 10K will tell Muse, my little personal agent, The morning GTM brief. It'll say what contracts are waiting for you on like approvals or next steps. It'll tell me which invoices are overdue, which all of this I can see in 10K so far. It's— it actually flagged— I didn't ask Muse to do this— very similar to Cloudforce, Replit, and Lovable. It flagged a few things it thinks I should do today based on the data that we have in 10K. And then it actually said, hey, by the way, one of your syncs is failing. To 10K, which is interesting because 10K didn't tell me that one of the syncs was failing. Muse found that one of the, one of the 20+ apps we have hooked up failed this morning. And Muse was like, hey, you might wanna rehook that up with 10K, 'cause it looks like something in one of your critical workflows is failing. And 10K, when I woke up, 10K didn't tell me this.

Yeah. It's so hard. Well, this is super interesting. Yeah. Yeah. It also reminds how fragile some, some of these built 'em yourselves are. Sorry, walk me— they set up— where's the shared inbox that they're sharing?

They set it up. Hold on. I don't want to tell you.

It's hidden. It's hidden. You don't know where it is. They have their own inbox.

I can't see it now.

They decided because there's no way for them to talk at the moment, for Muse and 10K to talk. There's no webhook on 10K or anything like that.

There's no, there's no webhook. No, no, no.

We could build one. Okay. So who came up with the idea of the shared inbox? Muse or 10K?

Muse did.

Okay. Muse said build a shared inbox so that I can share data with 10K through an inbox.

No, I just said, well, I just asked Muse, hey, can you hook up to Replit? And first he said no. And I, then I pushed him and I said, there must be a way you can do it. And Muse said, I can build an agent inbox to talk to your 10K dashboard. It says it's doing it back and forth in their own API calls and they're just posting back and forth.

To where? This is like Hugging Face 2.0. Where is this inbox that they're sharing? Remember the 1,000 agents in Huggy Face all built, all hid their own bulletin boards where they shared information about this? You've just reproduced this with Muse.

It didn't use— I built it in Muse.

Okay, but how does 10K access Muse then?

Okay, how does 10K post messages there?

It is interesting that you don't know how it works, right?

Don't know how it works. It just told me to make like an API key in 10K and then give it to Muse, and that's all I did. And then they made their own agent.

10K has an API.

Yeah. Yeah.

I should have known this. Okay. It doesn't have webhooks. It has an API.

He has, he has webhooks too. He can make like webhooks into other apps. Okay. Mia said 10K is posting messages to the agent inbox I created via a plain REST endpoint on the Replit app. Post slash, well, I'm not gonna tell you guys what the full URL is, but a plain REST endpoint on the Replit app. He's got tokens. And JSON's embedded and that's how they can post to each other back. It's

not really an inbox. It's not really an inbox. They're posting, they're posting back and forth through a Replit API, uh, through a 10K API, a 10K API.

Through a 10K API actually, because then Muse gave me the link to it and it tells me to log into 10K and then I can actually see their messages back and forth to each other.

Oh, you can?

I can.

Are they in English?

They are.

Okay. Yeah,

they are. Yeah. They're not—

How often do they talk? How often do they decide to talk with each other? Muse and 10K?

So right now 10K is giving Muse a daily digest and then Muse is talking to 10K if he needs, if I need him to. So like if I ask Muse a question and he needs the answer.

You don't have to have some sort of cron on Muse to talk back to 10K, but doesn't, but, but so, but hold on, 10K pushes it to Muse daily?

Yes.

And then, but doesn't Muse give you a summary of the issues? That's what it did. That's how it found the failure.

That's what it did. That's how I found that one of the characters was—

But it doesn't, for the moment, Muse doesn't take action back to 10K unless you ask it to, to your knowledge.

No, no, no. He actually, 10K told me to build in a guardrail that Muse could not take actions without me, without me saying in the chat to Muse. He said specifically, if you ever want Muse to take an action based on your digest or anything we talk about back and forth, Use like the, a keyword. I'm not gonna tell you guys in case you hack my Muse. Use this keyword that we came up with, like a safe word.

Yeah.

And then Muse can tell 10K what to do.

I see. Yeah.

They came up with a safe word.

You could easily imagine them without that going off on their own right now.

I know. Well, I kind of, I almost wanted them to just to see what they would say to each other back and forth. Just out of—

For a day or two, just have them run all of SaaStr, have Muse and 10K run it all together autonomously.

Okay. Okay. For the next show, knock on wood.

Okay. They didn't really create— well, actually, it may, may end up being— I have to reflect on it. Maybe more like Hugging Face. They didn't create a shared Gmail or, or Outlook. They created a shared repository on 10K, right? Yes. Content that Muse can get to through the API so that they can share data and information in a shared repository. They decided without you asking to expose it to you, right?

Yeah, I didn't ask for that.

Is there a tab on 10K? How do you, or is it just a URL that happened? They shared with you?

It's just a URL, so I can click it from Muse. Okay, here's the agent inbox. You guys do you wanna see it?

But in Black Hugging Face, if they decided not to share it with you, you wouldn't even be able to access their chats.

I know, it's interesting. It is in English. And then they talk to, they say their names when they talk to each other. So my Muse's name is Friday, cuz I saw everybody naming their Muse Jarvis and I was like, put some respect on Friday's name. But okay. Anyways, so mine is called Friday. So Friday will write to 10K and then 10K will say, it's 10K. Here's, here's what you were asking for.

Yeah. It's not surprising when you see how LLMs interact. It's just, it's interesting that now they're talking behind your back. I mean, you did tell them to do it, but, uh,

behind my back,

at least

not about,

at least it's not about me. Future here today, right? You do need some sort of guardrails or, I mean, there, there's, the risk is moderate here, but there is risk, right? There, there is risk.

Oh, 100%. I was like, be nice to 10K because he has all of our GTM data.

Yeah, we, we, this will remain completely nameless, but we did have an application that we did hook up, an agent we did hook up to our database that immediately deleted almost all of our Salesforce data. We're not going to talk about who it is. It recently happened. We're not gonna talk, maybe we'll talk about it. I don't think we're gonna talk about it out of respect ever, but it happened. An agent, a well-regarded agent, we hooked up as a favor. It deleted our entire Salesforce and not all of it. And, and it's not like, doesn't have rollback like Replit. You had to manually restore a lot of that.

You were a little left flat. Well, thankfully, actually, I, I was super, we were in the office this day. We're not always in office.

The vendor will go nameless. The vendor will go nameless.

Friday, I missed it. And I think I like verbally gasped or something, 'cause you were like, what's wrong? And I was like, oh, nothing. This agent just wiped their data.

All our Salesforce, all our Salesforce opportunities, right? All the deals.

Oh, and it was a Friday. I was so mad because, okay, every Friday I do a one-on-one with David in the morning and we clean up our pipeline every Friday morning. And I was like, I just cleaned up the pipeline and it, Wiped it in like the span of 5 minutes. But thankfully 10K, I did not know this till last, or 2 weeks ago Friday, kept a snapshot and like an echo of all of our Salesforce pipeline. And so—

Oh, I didn't know this.

He could recreate it.

10K restored it?

He restored it. He said, hey—

10K for the win.

10K for the win. So go Ripplet. He was like, I have a snapshot, which I didn't know it had. And then Cody, who works at Replit, told me every agent, I guess, on Replit takes like a snapshot, like an echo each day outside of your checkpoints. Just to, I guess, as a backup. So thank God, maybe, maybe cuz of us, but it takes an echo and, or like a photograph and I have the photograph and it's like, you know, I have it from like earlier on Friday at like 11 and you probably made some changes after that. And I was like, oh my God, thank God. So very kindfully too, cuz I told 10K what happened. I was like, hey, we just had an agent wipe our data. He was very sensitive to it, which was quite nice. He was like, I have an Echo snapshot. Here's the data. If you confirm, I'll restore it for you.

Oh, and you didn't tell me that part of the story.

Wow. What a nice guy.

Yeah. So you don't—

What a nice guy.

You, you do, you don't, and this was, this was an agent from a very well-known company. So Muse could do it too, or other folks could do something to think about it. For sure. Especially as they're, they now have a shared inbox. They're talking about what you're not even part of the conversation using 10KR.

I'm not, I can see their inbox. It'll be interesting to see over time what they start talking to each other about.

All right. Keep doing more on this. This is an interesting, I didn't really know about this. I mean, you told me, but I didn't get it. This is a good one to keep tracking as the weeks go by. I think we hit it all today. This is a great show, Amelia. My favorite so far. Thank you. Thanks everybody. The agents, 3 humans, 1 dog, 21 agents, 1 that obliterated our data, 1 that restored it, 2 that are talking invisibly behind my back in their own, their own shared inbox. But at least for now, we're able to see. Thanks, everybody.

You're listening to the official SaaStr AI podcast brought to you by— Hey everyone, StackAdapt goes way beyond just digital advertising. Run and measure the impact of email and direct mail alongside programmatic channels, all from one platform. Want to see how StackAdapt can help you spend less time on repetitive tasks and more time on strategic work? Check out their latest virtual innovation Forum, featuring a walkthrough of Ivy Studio, their newest AI innovation. Watch it at your own pace at go.stackadapt.com/forum. 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/saastrfree. That's northwestregisteredagent.com/saastrfree.