The Invisible Layer in OpenAI, Mistral, JPMorgan, & Netflix

0:00 We power critical functions for a lot of businesses. Their core could be powered and heavily driven by merge. So these sort of things we cannot break. We cannot go down. Everything's four, five, six nines of uptime. And if you lose that, you can lose all your customers overnight. >> Some of your customers are names like OpenAI, Perplexity, Netflix, Uber, Mr. Dropbox, Freshworks and more. So we do serve some of the largest AI companies in the world powering their AI search functionality and also connectivity in their products. Consumer expectations are higher now. like you just expect significantly more automation and so the SAS platform cannot keep up with the expected automation. It's just hard to compete. What is the state of cyber security in this layer of AI?

0:36 >> They now speak English perfectly, they now write code perfectly, and they now have unlimited manpower, all driven by AI. The second you connect it to tools, which is what everyone is trying to do right now, that's where everything goes wrong. It's so hard to block them because they're coming from IPs all around the world and all it takes is one and it could mean the end of your company. All right, Shenzy Gil, welcome to Sorcery.

1:07 >> Thanks for having us. >> Thanks so much. >> This is a merge special. We have a CTO and a CEO on >> We're happy to be here. >> So, what's going on? >> We don't get cancelled. >> I don't think you're going to get can although we did have some spicy preod talk. >> Yeah, we won't talk about stables high school gossip. No. So, you guys are on a tear. You have a lot of big announcements that keep on rolling out. What's the latest? What's going on?

1:33 >> Yeah, honestly, it's been really exciting for us. Um, around like a year and a half ago, we made a pretty concerted effort to do like three efforts in order to really move ourselves into the AI space. One of them was using AI really aggressively internally, pivoting our existing product to sell or not pivoting but like adjusting our current comp product to sell to AI companies and then also launching our own AI products and those bets have really paid off especially in the past year and a half. So it's been really exciting seeing like all the hard work that we put in um finally coming to fruition. So what are those products today?

2:04 >> Yeah, so we have now a suite of three products. We have merge unified which is our original product. Uh it helps companies sync data. So basically you want to bring that in power something like enterprise search uh do rag you you take you take that you ingest all the data we normalize it and it's there for your customers helps build a really good search experience uh since then we've also launched merge agent handler which is essentially one uh sort of MCP server you can connect to to help your customers offer or to offer hundreds of integrations to your customers inside of agents and this can be used for internal agents within a company when you're building your own uh it can be used for external agents. So if you're building a customer support bot for example, um and then it also can be used to power just internal use cases. Your whole team is using cloud code, your team is using, you know, uh cloud or chatgpt or whatever it is, the the connectors all just appear in there and work with whatever AI tools you're using. Uh and then lastly, we launched Merge Gateway.

3:00 Uh and merge gateway is infrastructure that just makes it really easy for you to switch between hundreds of different AI models uh and route them based on a a ton of different policies. It has security layer built in uh and it has a lot of smart routing policies, cost-saving policies and dashboards that make it easy for your whole company to really optimize uh your AI spend and usage. One of the observations that I've had over the last two three years was uh you can see when companies don't really make a concerted effort to pivot or rebuild or or embrace AI and you can tell either by what they're posting about online, who they're engaging with, like how their products are developing, but you guys did make that concerted effort. What was that process for you internally, especially with with your customers too?

3:50 >> I mean, it was really hard. Um, there were a lot there was specifically one really, really big deal that we wanted to close and it was going to take up like 90% of our resources. Um, and it was really hard to figure out like, okay, how do we end up like taking how do we end up finding resources to start building for the future instead of just building for right now? And thankfully, even though it was honestly at the time really hard, the deal paused for a month. And during that time, we actually got a lot done. we we like finalized the idea that for our like second product which was the a merge agent handler um we also started really aggressively leaning into AI coding. So Gil and I like paused back on the keyboard like using cloud code really aggressively using winds really aggressively so we could also learn how to code um with AI and that that month actually really transformed the business because with us having firsthand knowledge of how powerful AI is from 0 to one it allowed us to see what was possible with our existing products as well. Did you also notice when you were doing that the vulnerability of your business before? I >> I mean I think absolutely. I think we realized in that moment it was the classic innovators dilemma where we had you know this product it was growing really fast. It continues to grow really fast but we also knew where the space was going and we wanted to put as many resources as possible in our newer products. Um and so during that month Shensei was talking about we also realized like we do not have this ability to just put unlimited resources on this new thing. How can we get more with less? And that was why we we knew we had to invest in AI. We knew we had to be able to build new products and and you know adapt our existing products with just a couple people. So honestly actually building our our next product agent handler. We had one engineer and me and Sheny and we were helping on nights, weekends, we did whatever we could. Half because we needed to contribute. we needed to help and the other half was if we are the leaders of this company we have to know everything there is to know about how AI works how you build with AI so that one we are more effective and two we're building for where the puck is going >> yeah back then like mostly our engineers were like oh like I'm going to input this code snippet into chat GBT and like ask a few questions about like how to fix something um but after that it was like we saw what was possible everything was different we were like you need to be building completely zero to one with AI >> how do you keep your team up to date on everything. Obviously like that is an example but um some founders have really pushed their teams to actually use and embrace AI and then also with your new hires do you test them what is the experience with >> it's cultural like our whole team is really encouraged to do it and if you don't do it it does it is a part of your performance um so everyone will share like tricks that they learned um we'll sh we'll also feature people who are really AI forward and it's not just R&D so like our accounting team is really AI forward same with our recruiting and finance teams um and our marketing team too. So everyone is in cloud code.

6:34 Everyone is generating um things as and making it automated as much as possible and we just really try to highlight it as like a positive part of like the merge experience. >> We also have yeah we have brown bag lunches, training sessions. We just started a recurring session on merge skills. So just skills that you've built and how to use them. Um we ask about it in the interview process. So what we're not looking for is I use the latest cutting edge, but we're looking for, you know, hey, I use it to code sometimes.

6:59 My company doesn't let me use it to do all these things, but I really want to. That's what we're looking for is just the desire. Um, and I think we've brought in the right people for it now. >> Yeah. But I think it has to just be cultural. >> Yeah. It's so interesting. more of the conversations I've had recently when I'm talking with uh either public company CEOs or really high growth companies when they're hiring out, they're trying to hire for founders or previous founders, people that went through YC, like that kind of thing that like high agency self um uh what's it called?

7:34 Self-development. It's an autoidac type of DNA within talent. And I've just been seeing this over and over again where there is like a clear bifurcation and people who just like think you can still just like apply to get a job. By the way, the the bar is always very low or high. I don't know what it is. It's one of it's one of those heights where it's as you've taken on and you've used adoption with AI more and more internally and the models are developing faster and they're more uh they're just more um efficient. Have you noticed the company build faster? Are you seeing the direct correlation there?

8:09 >> Yeah, I mean last year we didn't increase headcount that much, but our revenue accelerated pretty significantly. Um, and so it's had like meaningful leverage on our business. >> And yeah, I think one thing Shensey actually brings up a lot of the company is Keith Ro's barrels verse ammunition, you know, sort of sort of view where you have some employees that are ammunition. They're really good at specific tasks and just knocking that out. And then you have barrels who basically just knock down doors and will do anything to get things done. And it used to be that, you know, a team that had a barrel of a PM or a manager and a bunch of ammunition could get a lot done. But now that you kind of can have one person just go use, you know, codeex, cloud code, whatever to go build something, you really just need a lot of barrels to just go get things done. Um, so that also has shifted how we're hiring.

8:52 >> How do you do you like Codex? How's that going? >> We we love it. Yeah, we we like we like Codeex and Cloud Code. We use both. Um, we're big fans of them. >> Yeah, we we allow just like a budget so you can choose whatever tools you prefer. Some of our team members prefer codec, some of our team members prefer cloud. It's really just dependent on like and some people prefer like other things. It's really just up to their preference.

9:09 >> Yeah. The only we just have security reviews. Obviously, we can have people going wild on every tool, but otherwise Yeah. >> Yeah. >> So, let's talk about the growth. So, what are the current metrics of the company? How many customers do you have now? >> Yes. So, we have over I think over like 20,000 um self-served organizations on the platform. Might be 25 now. I probably need to check. Um, and over 400 enterprise customers on the platform, too.

9:33 >> I think Brex is one of your customers. >> They are. We love Brex. >> Yeah, Michael Tannon Bomb um actually helped bring us in like a long time ago. >> Oh my gosh, I know. >> I know. I love him. >> I really love him. Yeah, he's Yeah, he's like the goat. >> What are the main categories of customers that you bring in and what are their use cases? >> Yeah, so there's a couple different categories like traditional SAS platforms um which can include something like AI. So like our first customer ever was Drada and like we of course sell to a lot of the stock two platforms like Drada Vanta um >> Spreo yeah you name it and then also a lot >> you know sadly they didn't sadly they didn't on board onto merge next and then expense management platforms um so yeah like bramp and like a few a few others as well um and then also large financial services so we have customers like JP Morgan um US bank um yeah and that's been really awesome for us. And then of course like the large AI companies and that's been like a newer segment that we actively invested in early last year. Uh so we do serve some of the largest AI companies in the world powering their AI search functionality and also connectivity in their products.

10:38 >> Yeah, that would be large LLMs as well as some of the largest you know AI I'm not going to say rapper but you know AI sassish platforms. >> Yeah. What is the difference in selling to these different because that's a lot of different categories there and also stages of company. >> The types of products that they're probably purchasing defer a little bit. Um, and so we're able to we're able to have like a a guess on like what what products they'll be mostly interested in. Um, so for like large financial services, usually it's probably like our unified API for deterministic use cases.

11:07 Um, or if they're building some kind of like agentic product, then they'll need our merge agent handler um product as well and maybe merge gateway. Uh, for like large AI companies, it it's usually like the connectivity that we're really really a part of. Um, yeah. And then for SAS platforms to be honest, like it could be all three. I think one other interesting difference that we've started to see is when people are buying us for AI use cases, they actually don't really know what they're looking for as much as it was in the past, right? Like we would Yeah. You know, 2, three years ago, we go to sell our unified platform to, you know, a classic SAS company.

11:39 They have seven people who are API experts on the call asking detailed questions around rate limits and how it handles this and that. And now we go on the call and they're like, "Sorry, we need a chart of how this works because we don't know the MCP protocol. So like we just need to understand here's our software. just like tell us is this is this going to work with it or not. Um so we're I a big part of that is like it's forcing us to evolve now adapt our product to what people think that they want when no one really even knows what they want right now.

12:04 >> That's so interesting. So what is I mean I guess to distill that down further what is the sales experience in the AI world like people just don't like they don't really >> we have to say like oh we've seen this from other customers like this these are like best practices are um a lot of times they don't have experience with like partnerships for these different integrations. they're not sure like what the best end user experience is. So we'll just we have to use our experience to kind of guide them through the best way to build. Um but also a lot of these AI companies they purchase much faster.

12:31 Uh like a large financial services like the deal cycles are definitely just longer. Um for SAS SAS platforms shorter because sometimes we're selling to an existing product that already has product market fit. If it's a newer product they're also like a little bit unsure of what adoption might look like. And for these large AI companies it's it's really fast um because the competition is just so fierce. Well, even a lot of times they come into calls and we're like, "Okay, this is what a PC would look like."

12:55 They're like, "Oh, no. We already had our agent built into it. We know it works. We just want to under make sure you know." >> Yeah. Really? >> Yeah. >> They do the proactive approach there. >> A lot of times, yeah. >> And so, what are their biggest use cases for that? What are they dying to have ASAP? >> A lot of times it's like specific uh deployment options. So, they'll have like specific security requirements. Um they might want something custom. They might want specific connectors that we don't currently offer. Um, but they've already done the diligence where we are the best partner. But yeah, usually the the deployment options are like the big the big part that we end up having to work with them on >> a lot of our product and adding new connectors and really supporting what our customers need for connectivity and for security and all of that like that's that's great. Um, but a lot of times now I think I think we kind of had this almost >> move back in time with trusting cloud and trusting other services and and you saw it when AI first when Gen AI started to take off. Everyone was like, "No, I don't want you to train on my data and I don't want this and that." And all these custom clauses got inserted into every contract that was like no use of AI, no doing this. Um, and I think there was just fear of uncertainty kind of like we have on these sales calls in general with people who just don't know what they're looking for. But now I think over the past year, you've seen people use the models, trust them more, and now everyone's backing away from that language. But there's still a lot of reservations now around using anything AI in the cloud. And so still demands for no, we want this running in our own infrastructure. we don't trust a multi-tenant. Uh but it is kind of again moving back towards okay we trust the cloud again.

14:21 >> Yeah. There's also especially with the players that we're working with there's already pre-existing partnerships or requirements where they can only use certain cloud providers and so then we have to like oblige by those requirements as well which is our pain but you know it's sometime it's worth it. >> Yeah. I'm curious, you you spoke a little bit about this uh in the explanation, but like cyber security has become a hot topic and we're seeing prevalent breaches over and over and over again and a lot of times it's through integrations and different APIs connecting things together whether it's like someone's using I don't know like Verscell was talking about theirs openly like we saw the merk war one we've seen so many okay so what is the state of cyber security in this layer of AI >> yeah I mean I One of the biggest problems so what caused you know Merkore and a bunch of others is the supply chain attacks where everyone's using these same open source packages but now the number of pull requests or code change requests that are being sent to GitHub is I I haven't seen the chart in a little bit but it's like massively soaring to the point where it's you can tell it's all agentic right agents are pushing a ton of code you don't have enough humans to read all that code and so things are slipping by things are getting in and one of them was a vulnerability that gets injected into an open source package that everybody relies on and uses and so all of a sudden that that little you know virus or that file goes into all the code bases and people are just getting really screwed over by that. So I think there is a need for for seriously like slowing down especially with these core packages being incredibly careful. Um yeah I think that's that's one of the the big ones we're seeing and we're only going to see more of that as more AI generated code continues to get pushed out.

15:59 >> So what's your solution? So that so that one is a tougher problem but I think I think the the problem that that we fall into and and what we're doing is is really this problem around uh integrations empowering data to be sent anywhere. So if you think about the world before genai writing code before agents actually you know sending data to places what you had was you had an engineer who explicitly said pull this data these exact fields from this platform and then take it transform it in this way and send it to this platform exactly in this way and then a second engineer had to go and approve that code and make sure that that looked right now instead you're saying hey AI agent that is non-deterministic and can do whatever you want whenever you want take the data from here and send it here and please don't send the social security across with it and then next thing you know that gets sent across. So that's where we are. We're blocking that type of thing from ever happening. We don't trust agents. We we say, "Hey, we can we can try we can try to set rules, but there need to be hard guard rails and blocks for things like sensitive data being sent across." Um and so we've doubled down there. We continue to double down around other things like jailbreak detection and all of that as well.

17:05 >> That's crazy. >> I know it is. >> I know. We've built a lot. It's really cool because I think honestly like a lot of the issues with agents is like data leaving the system. Um Yeah. And that's what would be really helpful. >> Yeah. And and you know I've written about this and we we'll write more about it but you know it's it's kind of like if you have this sort of evil genius who's a mass murderer but they're locked in a jail cell. Like they can't do anything, right? And it's the same it actually is the same thing as an agent though, right? Like what what do you care if the agent is so smart it could do any cyber attack and knows how to do everything but it has no tools and you're just it's sitting there just talking to you. Like who cares? it can't do anything. The second you connect it to tools, which is what everyone is trying to do right now, that's where everything goes wrong.

17:46 >> Well, I heard with Slack too, there's like major there's like a major leak there just as you deploy it, you can get all this sensitive data just through onboarding onto Slack. >> Yeah. >> Company information. And I would probably take that part out, but >> yeah. Yeah. I don't know about that one. >> I've talked to a couple people about this. This one of our interviews was with Gilly Ronin from Cyers >> and he had a crazy like doomer viewpoint but he's also being realistic. He's like look if you are multiplying the amount of instances that you could be hacked and there are breaches it's obviously going to happen.

18:22 >> So we were like talking about he thinks like really dark days are about >> also it's getting easier to find breaches too. I think you can just tell there's a lot more like automated like ways to try to find bugs and like security issues. Um, so yeah, it's both sides. It's easier to like create issues and also easier to find them. >> I I'll be Yeah, to be honest, like I I'm on call this week for our team. I like to go back into the on call rotations just to kind of see what's going on. Um, and yesterday I had to manually go in and intervene because we all of a sudden had over a thousand bot signups in in like an hour and we could see them actively scanning all the endpoints across our back end and it was called like scanner A, scanner B, scanner C.

18:59 And it's so hard to block them because they're coming from IPs all around the world. Fortunately, we have some good tooling with bot detection and we were able to to block them. Um, but those are just getting really prevalent and all it takes is one and it could we know this from some recent instances like it could mean the end of your company. Um, so >> yeah. >> What else are you seeing? I mean, you probably read and >> Oh, yeah. He's so good at that.

19:19 >> And research everything. >> Security reviews too for like different vendors that we end up onboarding on too. >> Oh, wow. Yeah. Yeah. I mean, I see a lot like we've just been seeing just repeated attacks. We've been seeing people just, you know, like uh I'm I'm sure you saw the GitHub one recently. I think that was what not that long ago where basically Whiz found that they were able to do a single git push of a file and gain access to every single repository hosted on the platform. And fortunately they responsibly disclosed it, but I'm sure Whiz did not find that manually. They're using AI. They're finding these things. Everyone's hearing about Mythos. Who knows how true the whole secrecy and whatever story is, but the point is like good models are going to be really good at finding these things. So we need to get ahead of that.

19:58 >> So what is your view on mythos? I mean my I guess I guess my take is like it's >> hypothetically or you know realistically I don't know which one it is. >> I mean I think it's realistic that if we have an incredibly smart model it's going to be able to find a lot of things that have existed in these packages that we've been using for 20 30 years. I think it's also good that we're letting it loose uh you know for for kind of like security and researchers up front to find those things and patch them because we saw how bad the Axios attack was where they just I'm sure they used an agent to find you know a sort of like some problems and and find patterns of how they could easily inject uh something into the package without getting caught. Um, but it worked and and so I I think it's it's also the other problem here is, you know, a lot of these attacks don't come from Americans or from from people who who live a really great quality of life and are not, you know, not like looking for ways to make a quick buck. It's it's people who who necessarily need to do other things to find to find money in the world. They now speak English perfectly. They now write code perfectly. And they now have unlimited manpower, all driven by AI. And so I I think we're just going to see more and more of it. Sorcery is brought to you by Brex, the financial stack trusted by more than 30,000 companies, including one in three ventureback startups in the US. Nearly 40% of startups fail because they run out of cash. Rex is literally built to help founders avoid that.

21:20 Unlike traditional banks that let your money sit idle, chipping away at it with fees, Rex is designed to help you spend smarter and move faster. Their all-in-one solution combines checking, treasury, and FDIC protection into one powerful account. You can send and receive money globally at lightning speeds. Get 20 times the standard FDIC coverage through their partner banks, and even high yield from day one with same day and even same hour liquidity. Access your funds anytime. Companies like Scale AI, Door Dash, Service Titan, HIMS, Anthropic, Flexport, Robin Hood, and Plaid. Trust and use Brex. Start today at brex.com/sourcy.

22:01 That's brx.com/sourcy. Touring is training the next generation of AI with tasks that require real expertise and realworld judgment. That's why companies like Nvidia, Anthropic, Salesforce, and Gemini partner with Turing. Turing builds realistic reinforcement learning environments and data systems based on real operational traces. The kind of infrastructure frontier labs need to train super intelligence. Visit turring.com/souc ry.

22:34 >> So as CTO what are the precautions that you take? You have a bot detector all these other things internally like what are the how do you go about it? >> Yeah so bot detection is one of them. Um, you also, you know, I think in in security in general, I think this is not commonly known, but your biggest threat is always internal, right? So, we're we're less afraid of someone scanning our ports and finding issues as we are someone compromising someone internally and fishing an account and then using that their account to do things or, you know, hiring the wrong person, hiring someone who's secretly a spy. Like, that stuff sounds far-fetched, but it happens. Uh, and that's why we we're crazy about background checks and doing reference calls and and making sure that we really are only bringing in trustworthy people. Um, so again, I'm actually more afraid of that, of us bringing in someone, you know, bad than I am of anything else.

23:20 >> Oh, that's it's a good point. >> Yeah. >> I'm so curious. This might be a little off topic, but when cuz most of these are organizations that try to hack into companies, breach, and then sell the data for millions of dollars. >> What do they do with that data? What happens? >> What >> when they sell it to someone? What What do they do with the data? >> I think it depends on what the data is.

23:42 >> Yeah. I mean, it depends on what the data is, right? So, imagine someone steals all of uh I'm not going to use I don't want to use merge as an example. It's bad. It's bad karma. Our data is our data is unstealable. But let's say like someone hacked plaid. I'm just going to pick them. Oh my god. They're finance data. Yeah. >> Make up a company. >> Yeah. All right. All right. Let's say someone hacked a uh financial data company that that has, you know, a bunch of I don't know. I need we need TO ROLL THIS BACK. OKAY. LET'S SAY there's a scenario. What would the scenario be?

24:17 >> Okay. So, let's say that there is a scenario where data data got got hacked. And this could be people's financial data and potentially their social security numbers, right? There's there's a few different things that they'll do. So, first they'll reach out to the company and say, "Hey, we have all this data. Pay us for it." Um, and the company has a lot to think about. They immediately engage crisis management and their thoughts like candidly going through their head is is, you know, one, if we pay for this data from this company, are they truly going to then wipe all the data and get rid of it? You know, we still have to notify everybody because technically once a second copy of that data is out there, it's out there. Um, so that's kind of the next piece. So, people are thinking about how much it's going to cost to get them back. They're thinking about damage control. They're thinking about their brand. Thinking about all of it, but the attackers know that. They know that actually this data might be useless to us. It might not be worth anything, but we know that to the company. It's their reputation on the line. And so that's who they're going to blackmail. They're going to go after that company and say, "If you do not pay us for this data, we're going to leak it." And we know that means that you're going to lose all of your customers. There's other instances where they're they genuinely hacked credit card numbers and the data itself can just be immediately sold on a black market. Um, so that's another instance. In either case, the company that got breached technically has to notify everyone about it, though they often tend not to do it. Um, but that's why I think companies really care about it. You know, morally, you should care about good security, but business is business. PE, companies are thinking about their own reputation and how much any breach is going to cost them at the end of the day.

25:43 >> Okay. So, what else is the state of the AI infrastructure market? What's going on? >> I think I think costs is another one that's been that's been, you know, really crazy to follow along. You know, you see everyone saying token max, token max. And that's really great in theory. It actually is great in practice, too. You're seeing a lot being built, but then the bill comes to the CFO and it's actually really brutal and way worse than they expected. And even if they got double the productivity, they might not have had double the budget for headcount and they they, you know, spent unreasonably. And so, we've invested really heavily in in cost savings um using our our gateway product, which has been been really cool to see as well.

26:19 >> Yep. Yeah, and I think a lot more margin focus is also starting to increase. Like obviously we saw cursor increase their margins which is very very hard to do. Um but it is really important and there are a lot of really great open source providers out there too that are very cheap and maybe that's like the best platform to route to if you're doing like what's oneplus 1 or like hello or like hi or like thanks. Um but it's really hard to automate that.

26:39 >> Yeah. >> Yeah. >> Yeah. >> What else are you guys seeing? >> I could share another this is not quite gateway. Um but another another problem that we've really seen actually is is internal use of AI. Okay. >> Um so when we talk about you know our our agent handler product which helps you add you know integrations to your products one of the things that we we heard in a lot of customer conversations was hey actually internally we want our employees using and automating everything. And you know, I I actually had a uh you know, the CTO of a really large financial firm say to me, uh, my employees have been coming to me and saying, "All my friends are using agents to automate this and do that and do that, and why aren't we?" And he was like, "Cuz we're regulated. Cuz we can't let you just connect all of our internal services to all of your external services." Um, but really, they should be able to do that, right? It just security is a big problem there. Um, so one of the big things that we've invested in and that we just see people with a lot more demand for is the the governance and control over the agents and full visibility into what your employees are actually doing with them.

27:38 Um, so let my employees actually go and connect Salesforce and read all the data out of it and send it to other places, but also do it in a way that I'm going to detect if if you know again like bank numbers are being sent places or um if if if proprietary data is being moved out of company boundaries. Um, so that's yeah, another big and and for that you see companies being like this has to tightly couple with our identity providers and keep up with all of our employees and if they ever leave they immediately get access revoked. Just been seeing a lot of demand for that sort of thing.

28:09 >> Yeah. Another thing is also granular access. So for example, if you hire a PR intern and you want them to have access to like your accounts. So you want them to be able to see the accounts in Salesforce. Um, right now it's either you give them full access to Salesforce or you can't give them access at all. Um, and there isn't really a great solution for the in between, but we're able to make it so you could your that like IT manager or that CTO can make it so that PR intern can only like see accounts and nothing else. Like as hard as they try to like write data to Salesforce and they try to delete any data, it's just not possible. Um, so we're able to help you do that.

28:37 >> Yeah. And the the one of the problems there too is like people aren't just using one platform for all of this, right? People at companies are constantly saying, I want to try this new platform, this new platform, this. And security can't govern that. They can't keep track of what's being connected where. Um and so part of what what we built was was basically that central hub. So you know you built you you connect to all this software in one place on merge agent handler and then merge agent handler connects to any platform you're using. So if you have some intern who's like I want to try this new AI tool you say okay you can use that but don't the only tool you're allowed to connect it to is merge agent handler that connects to all the tools in a more govern.

29:11 >> What's also cool about that is it's kind of like the employees like central node for all connectivity. So whether they're using codecs or claude or you know perplexi or whatever you're using all the connectivity and the actions through go through that node. So the CTO is able to see all the activity if there's any security violations and also if they want to revoke any access or add more access. >> Can I ask this what are the most popular connections? Yeah, I mean it's just like general like productivity tools like ticketing systems. Um like file storage systems like Google Drive, Box, Dropbox.

29:38 Um also people are really obviously like messaging systems are very common just for like automation of um communication. Um email is very popular also like code repositories are very popular but yeah just like general general like productivity tools. There are more specific um connectors that are very popular for specific functions. So like obviously for accounting it's like the like most common accounting systems like Quickbooks, Netswuite, um Zero marketing teams also are starting to use a lot of different platforms too like HubSpot um like a HR I don't even know how to say this one but yeah like all these like different like um GEO different platforms too. Um but yeah just overall like it >> it everything is getting connected and even if there isn't even if there isn't a public MCP server we'll we specifically create our own tools so that's not like a blocker for anyone.

30:23 Yeah, we've also seen a lot of like customs systems of record for specific industries like credit healthare and then actually just more of a fun one that we've we've done as well is you know we have we have a bunch of consumer connectors that we've built. So like Whoop and Aura for example, one fun use case was we saw an employee at a different company uh connect their Whoop to their ASA task tracker so they could see how their stress correlated with the tasks on their board.

30:49 >> Oh my god. Oh my god, that's hilarious. Wow. Do you guys watch these trends, keep track of them for marketing purposes and for sales purposes? What have you seen over time? >> So, I think we're just naturally really interested in it. So, we just follow a lot obviously on Twitter, but then also um >> No, internally from all like customer use like are you watching the trends of what people are shifting to because obviously a lot of these code platforms are new and some of the old SAS platforms are they're boring and people don't want them anymore. We get a lot of heads up on like roadmap plans like where people are planning on going um what they're where they think things are going and also honestly on the partner side because we have so many different partners and Gil and I are also meeting with them a lot. We hear a lot of industry gossip about like future plans for like what they're doing with their API partnerships that they're going to establish. Um so we hear a lot naturally just because we are in the middle of all this activity. Um so yeah I I don't know. Yeah, gossip just like naturally comes to us. Yeah, a lot of the giants that that seem stuck right now are planning big moves. We'll see how fast they move on them, but there's some exciting stuff coming.

31:54 >> Yeah. >> Speaking of one of those and your favorite person, uh Salesforce and their headless announcement. Yeah. >> Yeah. I love Benny off. You do? >> I do. I know. I love Benny off. >> She's read all his books. She loves Benny off >> his podcast. Yeah. I listen to all of it. Yeah. >> And you had a story that you guys were about to meet him and he didn't show up. >> Yes. One of our team members, her husband works at Salesforce and it was like, "Oh, like I can get you guys invited to Ben's holiday party. Like, would you guys want to go?" And we actually had like our board meeting. We were like, "Oh, like you know what, like we should go like we we like we've been work we've been trying to like work with Salesforce for a while. Let's let's try to go." And so Gil flew out um for one day on economy so that he could go to the holiday party in New York. Um and so we go to this holiday party. We're like ready. We're prepped. I've read all I've read his books. I've listened to every podcast he's been on. I deeply research all their products, things they were buying, new tech they have.

32:43 >> Yes. Um, and we're sitting there and then he just never shows up. >> And then like 4 hours later, DJ's cleaning up. Gil still like, "Let's just sit here and soap." He comes by. But but I think what was really helpful from that time actually was that we did I did do a lot of research on Salesforce and like as did Gil. And um, we talked a lot about it and one thing that's really remarkable is it's very hard to have a dominant product and company for 30 years through multiple tech shifts. And so I just would not count Benny off out.

33:11 Um I think like he's able to adjust really well to whatever market dynamics are coming. And I think it's partially because of his his philosophy of like the beginner's mind that he always talks about in his book. Like if I started the company today, what would it look like? And so I think that if if Salesforce can do a headless um can make like can do headless like anyone else can. Um it's it's really really impressive what they've been able to do because it probably is very hard for them to accomplish that.

33:35 >> Yeah. and Salesforce I I I especially think they'll succeed with headless just because throughout throughout the past like they've always had >> for people who don't know what headless is also explain that. >> Yeah. So essentially like you never need to go to Salesforce you can have your agent it can it can you know go sign up through a through the API or through a CLI tool like a command line tool. It can you know create accounts. It can it just kind of does everything without actually going into the platform.

33:57 Headless being almost UI list. Um, and in the past they they've already shown that they are open to this for for 20 years because they have an open API and a lot of people have built on top of them. They they've built a great ecosystem around them where it's their mo you're using other tools. They actually, you know, maybe they do, but like on the surface they appear to not care if you really go into the app. They just care that your whole stack is built on top of the data that's stored inside of there. So this is actually nothing new, right? It's just saying, okay, now agents are capable of using APIs to a whole additional level. we're going to facilitate so that they can do everything via API or MCP or whatever you want to refer to it as. Um, so I I think they've shown that that was a mo that was something that was very successful for them and this is only going to continue that.

34:40 >> Oh yeah, I we we've like taken a lot of inspiration from what like what he said about beginner's mind for our company too. And so yeah, I would I would not count him out >> every day. What what if we started the company today? What would we do? >> What would you do? >> We did it. >> We're doing it. >> We did it. We did it. >> We did it. It's a good answer. >> Wow. How'd you come up with that?

35:03 >> So, I guess like because you guys are so close to the metal and you understand everything that's been evolving with APIs, can you explain how that market has evolved and do you think more companies are going to take the headless approach? >> I think you have to because I think a lot of times um not for all software but for some software like you're not going to have time to make a decision for what vendor to use. We notice this with packages a lot also in cloud code scarily but like you know it'll just decide which is the best package to for your product. Um and so they you'll probably end up picking vendors based off of that as well. Um and so I think you have to do that in order to compete.

35:37 So you need to make it easy for people to sign up create accounts pay um as much as possible because not everyone's going to want to meet with someone. There are exceptions of course like infrastructure security products like someone will want to it's still a trustbased business where you want to meet with someone. you want to you want to gauge like how trustworthy someone is and if you can really rely on them. Um but for a lot of like very simple use cases yeah you don't you an agent an agent can make the choice >> and also headless really removes a lot of the the barrier of learning any new platform like even Salesforce which isn't you know necessarily a technical platform like you I can remember the first time going in and being like what is an opportunity I just don't grasp this concept makes no sense to me yeah but like technically let's say you want to have this this whole business that you're running you know you are the one person company that's the next you know unicorn or whatever you can't be knowing the intricacies of your host ing platform and of Salesforce and of everything, but you need all that stuff and you need those those functions going. Um, but you know, and and Lovable is a great example, right? Like you have your site, Lovable deploys it, it gets it running in the cloud. Maybe it's not like infinitely scalable, but it works.

36:38 There's no reason that you shouldn't be able to tell an agent, you know, locally like go build this thing and it deploys it to AWS in a very scalable way or to Cloudflare in a very scalable way. Like everything can just be built and done by a local agent without you needing to know how any of it functions. um just using using sort of headless uh integrations all from one central agent. >> Shenty, you mentioned like a really interesting thing when you were describing Salesforce and I I don't know maybe it changes with this wave of companies but in the AI world like everyone is so AI build but it is evolving so fast. you guys made a concerted effort again to like rebuild and pivot towards these new opportunities, but how do you see that playing out, especially because you've you've been founding companies for a while? Like, how do you see this next generation evolve? Like, do you think people will be able to adapt like that?

37:30 I I just think I think it'll be hard. Like I think right now it's very much easy mode for a lot of these companies and but every year from what we've seen like we started a company in 2020 and I hear like everyone says like the year they started is the hardest year but like when we started in 2020 it was like peak covid and people we fundraised when people weren't even used to doing zoom meetings. Um and that was like very hard and then the year after like then like everyone was fundraising crazy the year after everyone died and then after that like then it became just like a crazy again. Um, and so yeah, I just think you need to really learn how to adjust regardless of what happens with the market. And I think Venoff's really good at it. And I think a lot of these people are not going to be used to that.

38:05 >> Can you tell when you're looking at companies or you're like seeing them online like which ones are faking it or not? >> I I think I think the best way to succeed is to just do things. And I think if you overintellectualize your company building instead of actually doing anything, you're too high on Maslo's hierarchy and that means that you're not actually able to suffer later. >> So, we talked about this a little bit off camera, but we were talking about talent and recruiting and you were saying some of your best hires came like off cycle, not through like fundraisers and that kind of thing. What do you look for in talent and what are those kinds of traits?

38:41 >> I mean, you you have to actually be interested in what we're building. Like if you're only joining because like you know like you think we're hot it's going to be easy or like um you think you're just going to like only make a lot of money and you can just coast that's just not the company that we are that we are and for most companies um that's just not a great fit. So it's really just like hiring missionaries versus mercenaries and filtering for that and and it is hard like I think during like the per like the more upfront you are about how hard it is to do company building the more you're able to weed out the people that are just trying to coast and just like ride on your coattails and not do anything. Um, but yeah, I mean it is hard and like it's interesting to see like all these like people joining bouncing from company to company to company where they think it'll be really easy because once something gets hard, they're just going to abandon you.

39:24 >> It's a common >> I know Christina Cordova had a really great tweet about this, but I I totally agree. Like you just you see those people and like they have really great resumes, but they're just not on your they're not in the boat. >> What was her tweet? >> I forget what it exactly was, but it was basically just like it was after one really hot company was going through a tough time. Um, and like a lot of they were people were starting to leave and it was basically just like yeah when like when you like hire these people just because you're really hot like they're just I I forget the exact I don't know I don't remember what it was but I don't know if it really resonated with me. I don't know if you find it.

39:55 >> It's the shiny object syndrome. >> Yeah. You you find it you see it a lot too in interview processes where you're talking to someone who's just like I'm I you know I'm interviewing currently only with top hyperrowth companies and um I want to derisk with every single question I ask. I want to understand how I basically have this like high-risisk high or sorry lowrisk high reward situation which just doesn't come often. Um so it those those are the people that we try to weed out pretty quickly too.

40:20 >> Yeah. Also interestingly like some um some candidates will be like oh like what is your cash burn? Like you're not spending a lot but I also want like a real 1 percentile um salary. >> Oh my god. How are you to get that? >> Oh my god. What are the craziest asks that you've had? >> I mean >> we we've had Oh, so we had someone we were like >> we Okay. Yeah. So, like people want to just like join the exact team with like no experience >> really.

40:43 >> Yeah. >> They're like, "Oh, I'll just be COO or I'll be co-CEO." And you're like, "What?" >> I was like, "Who are you?" Yeah. No, that does that does happen, too. Um, but yeah, obviously. Or like someone will just be like, "Oh, yeah. Like, I I I want you to pay like public company salaries." And it's like, well, then why are you here? >> Like you like the way it's supposed to work when you go to a smaller startup is like you're taking a bet on the equity.

41:04 You can't make more when you're going to a smaller startup. And maybe you can. There are some companies where you can do that, but like I think the rule and like also like in order to find people who are really there for the company, yeah, you have to take a little bit >> the the risk in in in high-risisk high reward is the equity and lower cash. Like that's what you're doing. If you're getting the same amount of cash, there is no risk trade-off.

41:25 >> Okay. So, I want to go back into the SAS apocalypse a lot. You know, I don't think we we uncovered that enough. So, what are you seeing in the SAS world? I mean, the public markets, they they've become so volatile off of all of this, but whether it's like enterprise sales or other kinds of trends that are going on. >> Yeah. I mean, I think enterprise sales for these large companies is just much harder because the time to build the same exact product in-house is just significantly lower. Um, there's also just less leverage like when you're doing a negotiation against the customer for why they should renew. Um, yeah, like you can always just like, oh, I could just build this. And before that, that meant a very different cost. But now the cost for doing that is very very cheap.

42:03 >> Yeah. And I think people push back on that a lot at the beginning and they still do a bit but as models get better and better. Like we actively vibe code we we we do this every day and now it's not just like hey go build this thing and now let me correct it a lot and like no AI is coming back to you asking clarifying questions and it's going and building a pretty robust system. It gets to the point where English becomes your programming language. Like, yeah, at some point are you going to need to buy SAS or are you going to be able to say duplicate this best-in-class platform?

42:32 >> Also, I think like consumer expectations are higher now, like you just expect significantly more automation. And so, the SAS platform cannot keep up with the expected automation. It's just hard to compete. Like you don't want to have to buy a platform and then build your agents on top. You just want it to come out of the box. >> Does that make you nervous? I mean, I think we're we're good at this motion, you know, and and we also have a sort of like in some ways forward deployed team. And the idea is, you know, sort of if you can take off-the-shelf software that does most of what you need and then have someone customize it to exactly what you need at a relatively low cost, which is what AI facilitates, then I think there's still a lot of value in in sort of platforms powering or buying, you know, sort of ready-made software off the shelf because there's still a lot of nuance and a lot of detail that you avoid having to figure out um that that falls outside the time it takes to code in general. Today's episode is sponsored by VCX by Fundrise, the public ticker for private tech, allowing investors of all sizes to invest in venture capital.

43:29 Learn more at getvcx.com. Some of you may not have heard this yet, but our sponsor Public just launched something called Generated Assets, and it brings AI into investing in a way I've honestly never seen before. Here's how it works. You type in an idea like AI powered supply chain companies with positive free cash flow or defense tech companies growing revenue over 25% year-over-year. Publix AI then dispatches a swarm of agents that scan every single US stock, evaluates them, and instantly builds a custom index around your thesis. What really stands out is how clearly it explains why each stock is included. And before you invest, you can even back test your idea against the S&P 500, so you're making decisions with real context, not just guessing. And beyond generated assets, Public lets you invest in stocks, bonds, options, crypto, all in one place.

44:18 They'll even give you an uncapped 1% match when you transfer your investments over from another platform. If you want to build a portfolio that actually reflects your thesis, visit public.com/sourcy, paid for by public investing. Full disclosures in the description. Enterprise AI runs on Merge, the AI info platform for integrations, agent tooling, and model orchestration. So your teams ship product, not plumbing. Mistral, Dropbox, and Drada already trust Merge and production. Start building at merge.dev.

44:48 Founders scale faster on deal. Set up payroll for any country in minutes. Hire anyone anywhere. Get visas handled fast and get back to building. Visit deal.com/sourcy. That's deeal.com/sourcy. I think we've become like a little desensitized to the valuations for AI companies or AI enabled companies. >> Yeah. >> Do you think they're rational? What do you what do you see with Silicon Valley and maybe even the broader market could be upper market?

45:18 >> Yeah, I I think it's it's it's crazy because the public markets are has just such a different story versus the private markets obviously. So I think it'll be interesting to see when like opening Ianthropic go public because then it'll become like kind of like a merging of the two. Um, yeah. I I don't Yeah, I don't really want to like talk get hit, so >> Oh my god. >> But yeah, like I obviously have a lot of like private thoughts on >> Yeah. to not to not name any companies.

45:45 Like I remember when when we first started Merge, we were like, candidly, it was it was really frustrating to see some companies where we knew what their revenue was and we knew they were raising at like a 3,000x multiple on that revenue. we were seeing insane valuations and like you can't help but be a little jealous or a little just like come on like why is this happening and you have VCs telling you you don't want that that's not the thing but in the moment you do it's of course you want that um and we're really glad we didn't because obviously a lot of those companies ended up falling and some of them ended up doing really well but but most didn't and we're seeing it now a lot of the numbers that are coming out some of them are really really successful companies and are obviously going to continue to grow um but we we know specifically like we know the numbers we know sales figures for some of these companies raising at hundreds of millions to billions and again we're looking at thousandx to you know hundred to thousandx multipliers like a lot of these companies are going to fail um or they're going to be doing well but just not be able to raise their next round and and they're going to be forced to massively lay off and and slow down.

46:43 >> How do you stay focused through all of that? There's just so much noise. I think >> yeah, we just have to think long term and also like I think we just have you just have to focus on like your wins for like your customers and the type of customers who you're onboarding and like what your actual business metrics are. Um but yeah, like obviously like it's it's a it's competitive market with when it comes to recruiting talent. Like it definitely makes it a lot harder especially for people who are optimizing just for like what has like the biggest valuation, what has the flashiest like numbers. Um and so that's where you really filter for missionary versus mercenaries.

47:11 >> So I think you guys have been a little bit humble. I know and I think these are some of your customers. So some of your customers are names like OpenAI, Perplexity, Netflix, Uber, Mistral, Dropbox, Freshworks and more. So how did that happen? >> Yeah, a lot of work. Like it was very a lot of work. Yeah. Um when we first got started, especially because we were infrastructure, a lot of startups like were very scared to use us. I remember talking to RAM and they were I think like a 100 employees at the time. Um and we were really scared to onboard them because our product was so like early and um yeah it's it's we've obviously gone like a long way since then and obviously they've grown a lot on us as well and yeah we just we've had to adapt the company a lot like I think after our series B we made a really concerted effort to move up market segment our team um have a more mature sales motion also just make sure our product was really enterprise ready and that was really hard but like last year was really when it all started kicking in.

48:06 Yeah, it it really took climbing a logo ladder, having the riding off the reputations of of each successive company size growth to be able to close that that next level and prove that we could handle that. >> Yeah. And then once you like close of not up. >> Oh yeah. Yeah. >> Yeah. >> Was it intimidating getting bigger and bigger logos? >> Oh my god. Yeah, >> it is. It's intimidating every day. Like we we power critical functions for for a lot of these businesses. like we're talking their core. You log in that could be powered and heavily driven by merge. You go and you use some core AI model and it it looks up any data from anywhere. That's us as well. So these sort of things like we cannot break. We cannot go down. Everything's like four, five, six nines of uptime. Absolutely essential. And if you if you if you lose that, you can lose all your customers overnight. Um like some companies have recently.

48:54 >> Wow. >> Yeah. >> No pressure. >> No pressure. None. >> Yeah. >> None. You guys are so chill. Oh yeah. >> Speaking of that, so one of our sponsors is Brex and they're all about performance, spending smarter, we love faster. Love Brex. Um, and so this is a question I usually like to ask regards to performance, but like how do you think about the metrics you use to measure success? What are the next milestones that you want to go after as a company?

49:23 >> So yeah, revenue obviously matters the most, but also how much money we spent to get that revenue. That's really important. So we're always looking like our gross margins, our cash burn, um yeah, our cash burn multiple, what our runway looks like. Those are just really important for us and we're always looking at that every week. >> Yep. And then the the metrics that we, you know, believe obviously heavily lead to that what what our the quality of our product. We think we think our reputation and how people view us in the market is the driver of that. And so for us, it's never been okay to be in second place. We want to be the leading platform always. If we receive any negative feedback, we action it immediately. Um we it's it's critical.

49:55 We are the number one product on the market. We will not let that change. >> What are the biggest misconceptions that you think are happening in tech right now? >> Girl, what? >> I didn't even um >> like they told us it would be easy and then it's like >> I didn't realize this is the SAT. >> The biggest misconceptions. >> Biggest misconceptions. What about what was it about >> in tech right now? Yeah. Everybody's tied to their screens looking for the next like model release, but like what's the higher picture like >> Oh, yeah. Okay. So, I mean, one hot take one hot take that I have is that I've noticed a lot of companies like kind of overengineering their like ML usage.

50:35 Like they'll have like they'll build their own custom models. They'll try to um train their own models when really like you should you could probably just use the generic model and then focus more on making your product better. >> That's my I mean when should you build verse buy? I mean, yeah, there are some specific situations where you like obviously have to and maybe I don't know, but I I've noticed there are some companies that I was very surprised to hear how advanced they were when it came to training their own models when their product was lagging. And I and I don't think that their customers end up actually seeing the benefit of all of that >> work.

51:07 >> And sometimes you need to just focus on like what everyone could publicly see more. >> I actually want to second that. you see a lot of people trying to build like a custom harness or, you know, um use something like a a workflow builder to build agents that are that are repeatable and and all of that. And I just ultimately think none of it matters because we're we're almost at the point now where English is the language that you use to tell an agent. It will be deterministic very soon. We already see it. You just say, "Hey, go do X thing."

51:33 That hey, go do X thing is your artifact from then on out. That gets repeated by the agent infinitely. So, I think all these platforms that are around like build agents that are more reliable and that do things more repeatably and none of that really matters. All that's going to matter is just can you type it in English somewhere, have that run on a periodic cadence for your background agents. Um, and then do you have auditability for and observability for security?

51:56 >> Okay. All right. Well, this is an easy closing question. What are you most looking forward to this year? >> It's not easy. >> This is the hardest QUESTION I'VE EVER BEEN ASKED. OH MAN, WHAT'S YOUR FAVORITE COLOR? I don't know. >> Okay. >> White. >> Oh, yeah. This year. >> I mean, there's a lot. We put a lot of hard work into the past few years. Um, and like it's all really coming together. So, I'm this I I'm really excited for it to just like finally come through. I actually bought a spell earlier this year. I bought like an Etsy spell. Like, >> you bought I'm sorry. You bought a spell?

52:29 >> You never bought a spell? >> Is that why you're on sorcery? You know, like this started as a witch podcast. Really? Yeah. I love witches. You got a spell? >> No, I I'm just kidding. Oh, I have a only $10 for three wishes. >> I know. >> That's it. >> I know. And one of them already came true. >> And then they email you. >> Really? Yeah. The other one's like on its way. >> Are you sure you're not from LA?

52:53 >> Do you have crystals? >> No, I don't. >> No, I don't. >> I do ask everyone what their horoscope is, and I actually did not believe in horoscopes until when we started this company. Um, like 80% of our early team members were Tauruses and Libras because they can endure like abuse really well. Oh, I know. >> What are you? >> I'm an Aquarius. >> What are you? >> Taurus. >> Yeah, he's very like abusable. Yeah.

53:12 >> Damn. >> I know. >> Abusable. It's great. >> I'm a I'm a Leo. >> Oh, you know, you are a podcast host. >> What is What does it mean though? >> I just like I actually don't really know other than you're just like >> flashy. I think that's the only thing I don't really know anything about. But >> I'm quite an introvert. Oh, really? I don't. Yeah. My family was making fun of me at dinner last night cuz they're like, "Everything that you've been afraid of since you were a child, you're now doing this like facts."

53:39 >> Face your fears. >> Yeah. >> Um Yeah. I have I have uh one other one that I just forgot. So, give me a second. >> Um >> Oh, okay. So, yeah, >> this just in. >> This just in. What I'm really excited about for this year, uh no, what what I am really excited about for this year is that companies are going to actually start using AI agents. Um, so we we built agent handler and a lot of our products gateway because we saw all these problems with us trying to use AI and we were like, "Oh yeah, this is clearly a problem. Everyone's going to hit it." And then we start getting on calls with customers to sell it and they're like, "Yeah, bro. Like we just gave our our employees access to chat GPT in the browser. They're not allowed to use anything." And that was painful because we were like, "We have all this stuff ready and like no one is ready for it." And this year we're now starting to see companies being like, "We need this.

54:24 We need this." And we now not only have those products, but we've been able to run cycles on with a lot of the early adopters. So, they're built out and they are they are ready to sell as companies start to like kind of come online this year. >> It's a great way to end it. You're ready to go and thank you guys so much for having us. >> Awesome. >> Hey, it's Molly. If you enjoy our interviews, check out our newsletter, sorcery.bc, where we deliver a once a week top deals and tech headlines email, and also go deeper on our podcast interviews.

54:52 Subscribe to Sorcery today and don't forget to subscribe to the podcast on YouTube, Spotify, Apple or wherever you listen. Link in description to sign up.