The GTMnow Podcast

Inside Merge: The $75M Bet on Open Source AI (Powering OpenAI, Netflix & Uber) | Shensi Ding, CEO

• GTMnow

Share episode

Use Left/Right to seek, Home/End to jump to start or end. Hold shift to jump forward or backward.

0:00 | 47:18

The Co-Founder & CEO of Merge joins GTMnow to break down how she built a company that now powers integrations for OpenAI, Perplexity, Netflix, Uber, Mistral, and enterprise logos like Mastercard, JP Morgan, and Amex, backed by $75M from Accel and NEA.

Shensi gets tactical on the GTM decisions behind that growth: why she went enterprise-first into greenfield, why she's now layering a self-serve motion back on top (20,000+ self-serve orgs, 400+ enterprise customers), how forward-deployed engineers turned a 1% edge case into 5 to 10x usage growth, and why she rebuilt marketing to run like an engineering team. She also gets candid on founder-led sales, the mercenaries-versus-missionaries hiring trap, the open-source model shift and the China question, and why every founder who started pre-AI should build one product 0-to-1 themselves.

What you'll learn:
- Why she reversed the standard GTM playbook (enterprise first, then PLG) and how to run both motions at once
- How forward-deployed engineers close the 1% edge case that unlocks 5 to 10x enterprise usage
- Why marketing now runs like an engineering function at Merge, and how it took her from 1 to 2 dinners a quarter to 3 to 5 a month
- Why product got fast and go-to-market became the new bottleneck
- How to know when to hire your first salesperson (hint: you should be "dying" first)
- The mercenaries-vs-missionaries test, and the "anti-sell" she uses in hiring
- The 2026 shift away from "token-maxing," and how model routing cuts AI spend
- The truth about open-source (often Chinese) models: where they're built vs. where your data actually lives
- Why founders who started pre-AI should build one product 0-to-1 to understand what's now possible

Chapters:
00:00 Intro
01:14 What Merge does: products, workforce, and three core offerings
02:05 Agent Handler, Gateway, and the new Embedded Routing Stack
04:52 The 2026 shift away from "token-maxing"
07:29 Open-source models, the China question, and where data really lives
10:14 How Merge works with open-source model providers
12:00 What she's learned as a first-time founder: get over your ego
13:22 Reversing the GTM playbook: enterprise-first, then PLG
15:50 Running both motions: 20,000+ self-serve, 400+ enterprise
17:31 Why buyer behavior moved to Twitter and brand
19:06 Founder-led sales, and what a real sales process taught her
22:27 When to hire your first salesperson
23:49 Forward-deployed engineers and 5 to 10x usage growth
27:00 Keeping go-to-market up to speed with product
28:33 Rebuilding marketing as an engineering function
30:45 The dinner agent: 1 to 2 a quarter to 3 to 5 a month
31:24 How AI buying reshaped the marketing org
33:54 Maintaining agents: prompts over legacy builders
35:02 Mercenaries vs. missionaries, and the "anti-sell"
40:25 The market view: from hostility to collaboration
42:06 Is a valuation reckoning coming?
43:07 Why she stayed hands-on building Agent Handler and Gateway
45:18 What's next: training your own models off open-source
46:52 Where to find Shensi and Merge

Connect with Shensi Ding:
LinkedIn: https://www.linkedin.com/in/shensiding/
X: https://x.com/shensi
Merge: https://merge.dev

Host: Sophie Buonassisi, SVP at GTMnow
LinkedIn: https://www.linkedin.com/in/sophiebuonassisi/
X: https://x.com/sophiebuona

About the guest: Shensi Ding is the co-founder and CEO of Merge, a unified API platform that now spans product integrations, agent tooling, and LLM routing/governance. Merge powers companies including OpenAI, Perplexity, Netflix, Uber, and Mistral, and enterprise customers like Mastercard, JP Morgan, and Amex. The company has raised $75M from Accel and NEA.

Thank you to our sponsor:

Primer unlocks B2C ad channels for B2B marketers with precision targeting and next-gen measurement. GTMnow Network exclusive offer: Extended 45 day trial and 15% off with code GTM15 - just apply the code when you go to upgrade from the free trial.

See who’s really in your funnel: sayprimer.com



About GTMnow: GTMnow is the media arm of GTMfund, sharing the strategies, tactics, and stories from the operators and investors building the next generation of go-to-market.

Visit us on: https://gtmnow.com
Follow us on LinkedIn: https://www.linkedin.com/company/gtmnow
Follow us on X (Twitter): https://x.com/GTMnow_
Follow us on YouTube: https://www.youtube.com/@GTM_now
Follow us on TikTok: https://www.tiktok.com/@gtmnow_
Follow us on Instagram: https://www.instagram.com/gtmnow_/

The GTMnow Podcast
The GTMnow Podcast is a weekly podcast featuring interviews with the top 1% GTM executives, VCs, and founders. Conversations reveal the unshared details behind how they have grown companies, and the go-to-market strategies responsible for shaping that growth.

Visit gtmnow.com for more episodes and other interesting content. 

SPEAKER_01

Our companies like MasterCard, Jim Morgan, Amex, like very, very large companies, which is really exciting and very cool to like now be at that phase of a company where these really large organizations are very dependent on you.

SPEAKER_00

Shen C Ding, co-founder and CEO of Merge. You power logos like OpenAI, Perplexity, Netflix, Uber, Nistral, just to name a few. And you've raised $75 million from XL NEA in addition. What have you really learned as a founder through this process growing in a first-time company?

SPEAKER_01

Sometimes we'll have this moment like, oh, this is kind of embarrassing, like doing this, but like you have to do it. Like it's one of those things where you just have to get over your own ego in order to make the company successful.

SPEAKER_00

So you went enterprise fairly early, and then now you're actually almost going the opposite way and shifting back to a PLG motion. And you've got about 20,000 or over 400 enterprise customers. How do you think about the split?

SPEAKER_01

I remember thinking moving enterprise was really hard, but this is also very hard.

SPEAKER_00

What's next with AI?

SPEAKER_01

I think people are going to start experimenting, training their own models off of these open source models, but for very, very specific things that will become like a part of the equation.

SPEAKER_00

Absolutely. It's great to have you. Super excited to dive in. And for anyone unfamiliar with Merge, maybe let's start with just a little bit of context and background on merge itself.

SPEAKER_01

So uh we have two offerings, merge for products and merge for workforce. Uh merge for products offers a lot of components that allow you to build production AI. So we do offer integrations and tools for your products and agents, and then also LLM routing as well. And then we also offer merge for workforce, which is governance for your internal AI.

SPEAKER_00

Amazing. I'm sure the governance part is blowing up and we'll we'll kind of get into that. But uh that's definitely a bit popular. Yeah, yeah. And so you got three core products. Give us a quick breakdown of each and the exciting new feature launch under Gateway today.

SPEAKER_01

Yes. So we started the company um with our integrations for products um around six years ago when we were noticing that there was a lot of fragmentation in the market, and integrations were increasingly becoming more and more important. I think we got very lucky where AI would really end up needing a lot of integrations. And we were noticing, especially as RAG got really popular two to three years ago, and then more and more people were talking about context and how uh being able to sync a lot of data would really help make AI search really easy. Um, so we adjusted our product quite a bit to make sure that as AI Search got more popular, we are really like the premier partner for a lot of those AI search products. Um, so that's how we ended up powering integrations for companies like OpenAI, Me Straw. Um, and it's been really exciting, kind of growing with that. And around a year and a half ago, we started noticing there was a new segment of the market that was growing, which was uh integrations for agents. Um, and our first product didn't quite fill in the gap for what, for what like that new need was. And we felt like it was our right to win and something that we knew a lot about. And so we ended up launching a new product called uh merge agent handler, where we are able to help provide tools for these agents to third parties. We handle the OAuth, the refresh logic. Um, we help make sure the quality is really good. We also make sure that um, especially when it comes to like the different edge cases that could occur, uh, we're able to maintain all of it as well, especially as APIs change and there's new permissions um that may come up, especially with different personas. We're really that right that partner uh that works with a lot of our customers as we're offering integrations to their customers. And then we launched Merge Gateway around two months ago, uh, which is Smart LLM Routic. That's been especially exciting, especially right now with the changing atmosphere, especially with which AI models are most popular. Open source is increasingly becoming more popular. There's a lot of really interesting political dynamics that are also going on right now with open source models. I think also now people are starting to care a lot about their tokens fed. Well, last year that was not a concern at all. Um, and so we're really finding ourselves kind of at the center of a lot of these really interesting discussions. And so we're launching a new product again this week called Embedded Routing Stack, which previously, if you wanted to offer uh smart routing inside your own product, you would have that logic for yourself and it would apply to all of your customers. But now what we're making it, what we're offering is an API that allows you to embed the routing logic in your product so your customers can select what models they want to use and what how they want to optimize their AI usage in your product. Uh so essentially allowing you to like white label our functionality inside of your product, which is really exciting. Uh, we've noticed a lot of customers obviously passing along token spends to their customers as our pricing, um, which has caused a lot of companies to want to bring their own tokens, especially sorry, bring their own API keys, especially as they may have some pre-negotiated contracts. Um, that will also be available too. So just overall giving a lot more power to the end user or our customer's customer with how they which models they want to use and how they're using it.

SPEAKER_00

Very cool. And was that a tough decision to make the white labeling?

SPEAKER_01

There's a lot of resource constraints, obviously, at a startup. Um, and so it's kind of like deciding like what's a target first. Like, do we end up, you know, doing this first or that first? I think what we were really noticing was that the market was really moving this direction where people, it is just becoming more top of mind. Um, obviously it was becoming top of mind for our customers, but for their customers too, is becoming really of importance. Like, how do we end up reducing our AI spend? Not only like our direct model uh cost, but also the token spend that we're using in different vendors. And so there aren't really great ways to do this right now. And so we really felt like this is a really big feature that not a lot of people were offering yet.

SPEAKER_00

Yeah, yeah, definitely. And that's a huge topic because I mean, token maxing was traditionally something that companies were really encouraging employees to do, and there were leaderboards and everything. And it really feels like in 2026, we peeled that back from 2025 and and tried to actually use it to do so intentionally. Like, what are you seeing on your side? How is the behavior around token usage adapted now?

SPEAKER_01

Especially last year, or as people were experimenting more with AI, like I think last year was where there was a lot of experimentation, and this year is where people are a lot more comfortable and they're really starting to uh figure out like how to optimize what they're doing more. Uh last year, when people were experimenting, they would always just default to the most like most modern model. Um like if I was using an AI assistant, I would always choose the most recent model that I had heard of, whether it was like Opus 4.6 or 7 or like GPT 5.4, I think, or 5.3. Um, I would always just use the latest model that people were talking about and were think we're thinking, like we're saying really good things about um just because I wanted to have the best experience possible. And it didn't really matter because the AI assistant was just eating the cost. And whenever I was using my own tokens, it didn't really matter because it was kind of getting subsidized. And just overall, there was so much benefit to just experimentation in general that we would just want everyone wanted to just use the most modern, newest thing. But I think especially now that we're seeing a lot of open source models really catching up. Um, and each incremental version of like a new closed source model is not like that different from like the last version. People are really starting to think through, like, wow, I'm spending a lot of money now. Um, my entire team is trained. It's not like an issue where only um X percentage of my team is trained in AI now and I need to get as close to 100% as possible. Now you're kind of like at that 100%, hopefully. Um, and now I'm thinking through, okay, now we're at that capacity level. Everyone's spending a lot of money. Like, now what? Like, how do we end up making this like a really clear ROI for the business? And so people I think are very curious, especially as like open source models are getting a lot of attention, they're becoming more and more accessible. There's a lot of stigma against open source models, especially especially given like current political dynamics with China. But there's a lot of platforms that make it quite easy to just test your curiosity out and just see how it could go. Definitely.

SPEAKER_00

Yeah. And while we're on the topic, I mean, open source is a huge one. You see it everywhere right now. We ourselves, I mean, have invested in open source companies that are doing phenomenally too with GTM fund. And so you you see the momentum behind open source. And actually, separate aside, but I find their go-to-market fascinating. So I want to do a separate breakdown of open source go-to market. But when we talk about like the momentum that open source is having right now, for anyone unfamiliar, I guess explain why and then the political tension just around where your data is housed and these uh geographic tensions too.

SPEAKER_01

So a lot of these research labs are based in China, and um, I think DeepSeak was probably the most popular one that really caused a lot of awareness for open source and how good these open source models could be. There's a lot of muse around them, like, oh, China had like ended up doing a lot of distillation on the using like our US models, um, and it was like cheating and it was a the people felt like it was like very sketchy. And so I there I think it's just because there was so much conversation around it and uh like a big lack of trust around Deep Seek, um, that it created just like more stigma around like any models that came from China. They were probably sketchily done, like maybe they were good, but like I didn't want to be associated with it. And I remember having a lot of conversations with our enterprise customers where they would be like deep seek, like ew, like and um and I think that continued up until around like early this year as other Chinese models started getting more popular. And people as they were experimenting with it, they were like, Oh, like what's going on? There's always that dynamic, like yes, they are coming from China, but there's now a lot of different Chinese open source models. And just because a model was created in China doesn't mean that it's hosted in the US. It also doesn't mean that any of the data is go leaving the US or going to any regions that are outside of your any uh regions where like you don't feel comfortable. You can always control that and like the type of hosts that you use, especially if you use like US hosts that are um, you know, that are hosting those like open source models. So I think there's a lot of confusion and like a lack of understanding for what how open source works. And people just assume if it is Chinese, there's some kind of danger associated with it, but that's just not the case. But it is very interesting to see like a lot of the founders of like anthro, you know, the CEO and leaders of like anthropic and open AI, they're going to have a lot of thoughts about open source, especially because it does end up hurting a lot of the economics for what they can charge their customers. And so it'll be really interesting to see how it plays out.

SPEAKER_00

Yeah, and I always think, you know, the model might be built overseas, but if it's running in, say, the US, like where the model comes from and where your data goes are fundamentally two separate things, but we often misconstrue them. So it's interesting to hear your perspective there. And I mean, you are not, merge is not just for clarity for everyone too, open source yourself, but you are actually working with a lot of um open sourced kind of models. How are you working with them?

SPEAKER_01

Yes, in a couple different ways. We do have direct relationships with some of these partners. Um, so for example, like we work with like the ByteDance team, um, Alibaba, Quen. Uh so we have really, oh, sorry, Alibaba, um, and then also uh Kimmy as well. So we've had we've had really great relationships with a lot of these different model providers. And we will pre-negotiate their data retention contracts with them as well. Um, a lot of them do store uh have servers in China. And so we do have the ability for um our customers to make sure that if they want to block routing to China, they can do so as well. And then what ends up happening if you end up choosing an open source, a Chinese open source model that then routes to a US host potentially that is hosting the open source model with US servers. Um so we also work with like, you know, base 10, Amazon Bedrock, Fireworks, uh, a lot of like US hosts that will end up like also offering like these models as like offerings. Uh there's a lot of different fallback options here too. Um so it's a combination of like direct partnerships where they're hosting it themselves. And then also we're working with like different um like host providers in the US.

SPEAKER_00

Yeah, I kind of think of it as like merge. I mean it's closed source in general, but you help companies actually adopt open source AI models.

SPEAKER_01

Yes, for sure. And we even if you're mostly focused on like using open AI or anthropic or um like these more popular models, we can also help you optimize your spend within the within those providers as well. I think right now, like a lot of people are just routing all prompts to Opus 4.8. And sometimes, like if someone's asking like hello or like one plus one or like thank you, it could probably go to haiku. It doesn't necessarily need to go to Opus 4.8. Um, so even within like within that, we can still help a lot too.

SPEAKER_00

Super helpful for token usage.

SPEAKER_01

Yes, exactly.

SPEAKER_00

I love it. Well, Shansi, let's talk a little bit about your growth. So early, early days, let's start there. You were uh perhaps nervous or a little bit scared to onboard a customer like ramp when you're when there are about a hundred employees uh because the product was early. It was early days, but now you power logos like OpenAI, Perplexity, Netflix, Uber, Mistral, just to name a few. Um, and you've raised $75 million from Excel, NEA in addition. This is like tremendous growth. You as yourself, as a founder, like what have you really learned as a founder through this process growing in a first-time founder?

SPEAKER_01

I think, yeah, you just have to be willing to be embarrassed a lot and you have to really just shoot your shot. Like, I think that's something that's is just very uncomfortable, but you have to just do like the outreach, you have to just put yourself out there and you just are going to get rejected a lot. And I think that's something at this point where now, like sometimes we'll have this moment, like, oh, like this is kind of embarrassing, like doing this, but like you have to do it. Like it's it's one of those things where you just have to get over your own ego in order to make the company successful.

SPEAKER_00

Yeah. And I mean, we'll get into kind of how you're you're very hands-on in the process itself. But before we do, from that growth, you know, a big part we talk about with growth is go-to-market motions. And you yourself have a really interesting one because most companies will go from PLG to enterprise and kind of go with that upmarket motion from there. But your arc looks a little bit different. Like, walk us through how the motion actually evolved for you and what that did and why.

SPEAKER_01

Yeah, it's definitely been a journey and it's been a lot. Um, so we first got started, of as you mentioned, like we mostly focused on really small startups since we were infrastructure and people were really nervous about using our product in case you know we shut down or we didn't end up like lasting, or we just had a really terrible product. Um, and over time we ended up moving to bigger and bigger companies. Um, and around three years ago, we made a very concerted effort to move up market because uh we felt like it was completely greenfield. There wasn't really a player there. And it would allow us to really invest in our product quality and security in a way that would be really hard to replace. And so we we did a lot of changes there. We ended up like changing what our team looked like, doing a lot of um segmentation, uh, obviously investing a lot in our product and security posture. And to now we are actually mostly enterprise business, like most of our larger contracts, obviously there are larger contracts and a majority of our revenue are companies like MasterCard, JB Morgan, Amex, like very, very large companies, which is really exciting and very cool to like now be at that phase of a company where uh these really large organizations are very dependent on you. But what I noticed around like last year was especially with AI, a lot of marketing and also a lot of buying habits have changed quite a bit. Um, it's becoming a lot more consumer-y. And a lot of these like AI startups are starting to focus a lot, obviously, on like PLG sales motions uh or PLG like growth strategy. They end up being just more PLG. Um, and so therefore their marketing is a lot more like hype, launch videos, um, making sure that like people are just very aware of them and focusing a lot on brand awareness. And we were just so far away from that. Like we were very focused on like conferences, like ABM, and especially as we launched our two new products that were very, very AI and definitely felt a little bit more PLG. It just felt like our old marketing motion was not quite relevant for where the company was going and where we we needed to be. Um, and so we've actually actively been trying to diversify what our marketing and our customer base looks like by now also doing a lot more like brand marketing for startups, trying to make sure that um a lot more like developers and like ICs are more aware of us rather than it just becoming like a top-down awareness play. And so it's been really interesting seeing like how to do that in reverse. It's definitely hard. I remember reading, like I remember thinking moving enterprise was really hard, but this is also very hard. Um, and it requires us to have so different. And again, also just requiring us to hire like a different type of persona for someone who's like very like in the know and like understands like hype and like socials and the Zelda skill set that we don't currently like have. And so we've been excited to hire a few new people who'll be able to help us with that.

SPEAKER_00

Very, very exciting. So you went enterprise fairly early, and then now you're actually almost going the opposite way and and shifting back to a PLG motion, or shall I say, having both uh motions? Yeah, it would be more accurate. And you've got about 20,000, maybe maybe more than 20,000 self-serve orgs on it now, which is more of a PLG motion, about over 400 enterprise customers. Like, how do you think about the split now that you're running both? Like, where are you putting the investment when you're balancing both motions?

SPEAKER_01

Yeah, I think it's hard, um, especially since there's there's a lot of opportunity cost, especially with how easy it is to build things now with AI versus before, where you really had a lot more time to like fewer resources, but a lot more time to really like think through like, is this the right direction? Um, versus just like building it really quickly. I I think what we've been pretty good at is trying to make sure like as we lock new products that it's kind of like self-sustaining and it's paying for itself and we're not just like over-resourcing it before there's any data. Um, so as we've been, you know, growing revenue in each of these two new products, uh, we've continued to invest a lot more and we've been really careful to make sure that for our first product, we're not taking resources away. What's really great though is we do have this base of amazing customers that we're then able to now show how we've evolved as a company and they've been very interested in our newer products as well. And of course, like as we're expanding into like these new products and selling into different personas and companies, um, they're they're very happy to see that we are supporting uh logos that they recognize and trust um and that we are able to power companies of that caliber. Um so it has extended, it has helped us quite a bit, like that brand recognition of our existing customers, uh, which we are very lucky to have. But yeah, I would say like it's definitely not perfect, but we're trying to make sure that each business line is kind of like a self-sustaining company.

SPEAKER_00

And you you mentioned behavior is changing, like buying behavior is changing. Now, when we talk about going from enterprise to more PLG motion, is that actually for smaller companies? Is there a linear relationship there or has buyer behavior changed and more enterprises are a little bit more consumer-y too?

SPEAKER_01

I think everyone is changing a little bit because so much of the conversation around AI, especially for engineers, is on Twitter. I think before a lot of the conversations about like best practices for business building and like what different what software could help you with certain functions was really spread out across like word of mouth, um, different articles, conferences, Gartner. But now it's really not there at to the same degree, especially when it comes to like different AI tooling. All the conversation about different AI tooling us is on Twitter. Like whenever I'm hearing about like a different like coding app or like um some kind of like new AI design feature or like the new models, it's always released on socials. That's really primarily where it comes out. Um, obviously like Reddit too, but like, yeah, Twitter is really like a centerfold for AI discussions. And so because so much of it has moved there, you have to near fire where they're at.

SPEAKER_00

Yeah. And I kind of feel like anthropic is is a big momentum pusher on that front too. Just how they've been releasing so heavily through these videos on platforms like X. I'm sure other LLMs were too. I'm sure they were previously also, but just the momentum that these different model releases like 4.7 and 4.8 have created on social, like inevitably to stay up to date, you kind of have to be on Twitter now or at 100%. Yeah.

SPEAKER_01

They're real they're really good at it. Like all the all the model providers. Yeah.

SPEAKER_00

Yeah. Well, let's talk a little bit deeper on your go-to-market motion and dig into let's start with sales, because you've kept FounderLed sales going for quite a long time. Tell us about this a little bit. A quick pause because B2B has been notoriously slower than B2C on adopting strategies. But what we've all come to realize is surprise, humans are humans, and you can reach your B2B buyers effectively across B2C channels. Primer unlocks B2C ad channels for B2B marketers with precision targeting and next gen measurement. What happens is there are two ways you're wasting ad budget. One, you pay to reach people who will never buy, and two, the buyers who do show up walk away unidentified. Primer fixes both. This includes excluding the wrong people across Meta and Google before you pay for the impression. So it's great news. No more burning budget on students, competitors, and tire pairs. You get unlimited website reveals, every account researching you for identified, no caps, no per reveal pricing. You can see who's actually in market and feed it straight back into targeting. And primer is giving a super generous offer exclusive for the GTM Now Network, an extended 45-day trial, and 15% off any paid plan with the code GTM15. You can go and start your trial at sayprimer.com. That's S-A-Y-P-R-I-N-E-R dot com. It'll also be in the show notes. You will then go into the product, start your trial. There's zero friction to get started. And when you apply that code is when you want to extend your trial and or get 15% off your paid plan. Just enter that code GTM15. Back to the episode.

SPEAKER_01

Yeah. So back in the day, I would send a lot of outbounds. Um, I still send a lot of outbounds, but I would go through LinkedIn, I would like outreach people with messages. I would also use like Apollo and I would send messages out through sequences. And whenever someone responded, I would send them my calendar link, I would take their first call, or my co-founder would take the first call based on whoever was available. Um and then we would demo the product and then we would, you know, run the sales process until someone purchased. We definitely weren't perfect at it. And like when we ended up hiring like a real sales leader, we we learned a lot about like what a good process looked like. But even after that, like when as we moved to enterprise, like things changed a lot. And um, our current CRO, he had a lot of process where like we had we started having a deck. And like I didn't realize what a big difference a deck made, but it really shows like a level of respect to the customer and like you prepared for them, that you uh have thought about them ahead of time that I think the customer appreciates more than I really expected when I when we had first started the company. And yeah, now we we have like a it's kind of like the customer's like, okay, how do I buy you? Like I want to understand like what is the best way to test you and like you know, feel like I trust you and then then spend money if the if it goes well. And so we have to come in being very opinionated versus when we first started. I was like, oh, just try it out. If you like it, like let me know. Like, give us my money. Uh, which obvious which worked a few times, but like not not the it's not the same. Same way. And it really is up to the salesperson to really drive the deal through. And I yeah, I've gotten a lot of respect for how hard sales is. Like it's really an art. It's not, it's not easy. Um, and so yeah, our process has changed quite a bit. And then now, again, it's changing again since we have now this like PLG product layer on top. So we're trying to understand like, how do we end up selling this to like our enterprise customers that are used to a more structured process, even though it is very PLG? What does this relationship look like? So a lot of those things we're trying to figure out too.

SPEAKER_00

Yeah, and it really does feel like buyer behavior has shifted now where there's so much context, so much information that people are really looking for guidance on like how to think about something, how to think about a solution, the context behind it, not just try it out. So crazy how things have changed. But you know, you successfully scaled your sales org. It's one of the hardest things for companies to do. Like the amount of pain, I don't know if you felt this personally, but the amount of pain that companies often feel hiring the first sales rep is just off the charts. Like, how did you know it was time to hire that person? And how did you get it right?

SPEAKER_01

I think we were dying. And so that's why we knew it was the right time. I just like don't think you should hire a salesperson unless you and your founder are so bad at sales. But then if you are bad at sales, then the company's probably not going to make it because I think sales is like an inherent part of being a founder. Um, like you have to sit do sales in sales, obviously, but sales and recruiting, sales and fundraising, sales on like pro like your team so that they want to stay with you. And there's so there's sales in everything. So if you outsource like your initial sales to a salesperson, like you don't really understand, like you're not really understanding the feedback that you're getting from the field and why it's not hitting in the beginning because it won't hit, like it just won't resonate immediately unless you are very, very lucky. So yeah, I think you need to be at the point where you are like, wow, like I I think we are product market fit, like I'm dying. I really need more help. And someone said really professionalize us, and I think that that's the right time.

SPEAKER_00

And you guys run um a forward-deployed engineer team, correct? Yes. Yeah. So I mean, we recently uh published a piece actually about how companies are growing so quickly with direct sales. Cause in the past, I think the the perception was that PLG equals speed, and that is not wrong by any means. But the reality is when done right, direct selling can have just as much speed, if not more, too. And for deployed engineer is a key part of that. So, you know, around that is the idea that like a huge percentage, maybe even as steep as 95% of enterprise AI pilots actually kind of are unsuccessful at the integration stage and wall, not the model. And that's kind of that fix that the forward deployed engineer shipping integration comes in at. You at merge sell that layer and you run FDE teams yourself. Like, where's your perspective around forward-deployed engineers and how they come into the picture for other companies that are also trying to grow quickly, like merge?

SPEAKER_01

It has made such a big difference for our enterprise customers. And I I do think if we'd had an affordable deployed engineering team when we were mostly focusing on startups, it just would not resonate. Um, so you do have to get to the point where you are selling to enterprise companies where they can really use it, or you go straight to enterprise and then you can you can really utilize it. But yeah, I just think you're mostly selling to startups where they have really good engineers, like they're not probably not going to use your resources. But for us, like one thing that we were noticing was it was we had this product and it was like it would fit like 99% of the use case that they needed. But there was that 1% edge case that was really custom to them. And we could adjust our product to fit them, but it would only fit them. It wasn't great for everyone to just have that. But having a forward deployed engineering team allowed us to like fill in that really small gap. And then it would allow us to increase the usage and satisfaction with our product really, really significantly. And so we did that with two customers, and we just ended up seeing usage grow like five to 10x immediately after we ended up um use like having our forward deployed engineers like assist that account. Um, and so we were able to see like, okay, so these two use cases were very successful. Let's repeat for all these other customers that are having that same gap because now we've learned from that like little, like now we've learned like what to do here. Um, and so it's really helped a lot in situations where like our customer keeps being like, oh, like we really want to do this resource, like we really want to do this, we just don't have resources, or it's like, oh, like this product isn't quite right because of blah, blah, blah. Like a lot of those things are solvable, especially now with AI, it's like even easier now. And it allows you to remove like any tech debt from your product where it's like not, you know, like a perfect fit for everyone. So yeah, it really is like such a game changer for those large enterprise customers. And I think you you almost have to have them for that. But um, for if you're mostly selling to startups, like it's definitely not it's not your LPU and it will save you either. Yeah.

SPEAKER_00

Yes, exactly. Yeah. And you you can probably pattern match across too and take more of a proactive approach of going to customers with suggestions, which is incredibly valuable at the enterprise stage.

SPEAKER_01

Yeah, 100%. And like I think also just like saying, like, oh, we've done this before with someone else, like people immediately just trust the team a lot more and they really would uh value that resource a lot more than if it's like your first time. And so, you know, take that into account as you're doing sales.

SPEAKER_00

That is great advice. And now you got the product moving incredibly quickly at merge. What did that do for your go-to-market? How did you keep up to speed on the go-to-market side of the house?

SPEAKER_01

Yeah, um, I will say it's like definitely still hard. Um, I think historically what we really were lucky with was we had a lot of people who had been with the company for a long time. And so as we were adding new parts of the product, it wasn't too hard to just like add a little bit more. Obviously, that shifts a lot, especially as you add like new people. You have to keep like training with new things. Um, so we've been adjusting what our onboarding and like training looks like every month or so to make sure that we're continuing to keep up, keep up with the times and also like using AI as much as possible to make it so our team isn't like lost when they're like out doing like a meeting and they feel like they're not supported. But yeah, I mean, honestly, it is hard. Like it's just a lot to remember. And so I think we're trying to figure out like what are the sound bites for people to really remember and how do we loop in like SEs and specialists, like when it when we need to go into the deep dive, because it is like a we do have a lot of offerings and a lot of products. So uh it yeah, it will get hard.

SPEAKER_00

Yeah, and I mean a lot of launches, obviously, you know, we included, but when that happens, you know, there's this shift that we've seen at least uh where product used to be more of the bottleneck. And then as velocity has sped up, thanks to AI, uh, notably on the product side, now the bottleneck is actually your go-to-market. And companies are working on solutions for that and implementing really, really kind of smart, interesting, innovative solutions. I think you've done that incredibly well on the marketing side. How did you keep marketing up to speed with your product launches and other announcements?

SPEAKER_01

So one thing I agree with you, like I noticed at some point, especially like earlier this year, there was a period where like our product was outputting significantly faster than our marketing. But just because AI was not really ready for marketing yet. And I think that's why we're seeing so many like AI design companies and startups forming right now, is because everyone was really starting to feel this in the market. And so we started trying to figure out like, okay, how do we end up making like our marketing more of like an AI, but more of like an engineering function? Because it seems like it's probably moving in that direction. And if we want our marketing output to like keep up and pace, and we need to be able to like productize this a lot more. Um, and so we started thinking through like what were repeatable parts of the product, like what were repeatable actions that we do for marketing? It's like changelog. Okay, let's pull straight from the GitHub, let's like output this immediately and like add like a voice to it. Our socials and like our different um like new feature releases, it just pulls straight from our GitHub and like obviously someone does like a quick review, but now a lot of like the images and like the videos are all like auto gener and partnership images are all auto-generated. Um, and it has like our voice. Like we've continued to like add more and more. And so it's gotten better and better. Um, but we've been able to create like a lot of tooling that makes it much better for us to share with our customers like what new things are coming out. And then yeah, we've also like made like our even like our field events uh more AI too. So like our dinners, we used to be really, really hard to coordinate them because we'd have to have like everyone at the company look on LinkedIn, like who is in Austin? Like, let's invite everyone we know in Austin, and we would try like so hard to fill a dinner and it would just be a lot of time for like me, like the team, and like we don't have to do the CD chart coordination. It was just a lot. And so one day I was just like, okay, I'm just gonna like make an agent of that. And so we're putting everything that we end up doing for every single dinner into like one workflow, and we're just gonna productize this. And so we went from having like one or two dinners a quarter to now having like three to five a month because everything that we ended up doing for like outreach was just like who have we met, you know, in this city in the past like three to five years. Okay, let's like invite them. Do we have any closed or lost opportunities? Let's keep that open. Let's keep that into account for whether or not we reach out to them or not. Um, do we have any like similar customers to this company that we're trying to like outreach? Okay, then let's like account for that too. Or is the dinner full? Okay, then let's not invite any more people. Uh, we should sit next to each other, probably startups with startups and like bigger companies with bigger companies, partnerships with partnerships in the scene chart too. And so all of these things that we just kind of like mu intrinsically from like organizing these events, we just productize it. And so it's made our marketing like so much better. And so we're there's so much more though. Like marketing is obviously like a huge function, but we're continuing to try to figure out like what are areas that we can continue to like automate as much as possible. And it's helped us a lot.

SPEAKER_00

Yeah, that's that's fantastic because I mean, anyone that's done field marketing or or organized a dinner even in itself understands the pain that it can be. So that's fantastic. How how did that change your org chart at all? Like as you've implemented all these changes to make marketing function more as an engineering team, has it changed the shape of your function and team?

SPEAKER_01

I think we've just mostly focused on uh making like the team really AI native. So investing a lot in like training. Um, I think the biggest change in our org is just like how AI buying has changed things. So we've had to layer in like more brand designer, more content marketers, and like really focus more on like brands, a lot more than we ever did. And like how to look cool. That's so I think that's really like the I think that's like the weirdest part of like this big transition is it really is like an important part of like how people are buying things. Like you want to just buy things that you feel like are most like advanced, like everyone's talking about. You don't want to buy some like crusty thing. And so versus like four years ago, like you wanted to buy the most trustworthy, secure thing possible. And so, yeah, we're just we're really having to adjust based off of that more so than just like AI.

SPEAKER_00

Yeah. Why why do you think that's happening? It's such an interesting phenomenon.

SPEAKER_01

I I guess the IC is getting a lot more power for buying things versus like the leaders. Yeah. I think also there, to be to be totally frank, there haven't been enough security incidences for it to become really locked down. And so I think people are okay with more experimentation from like an IC level where people are trying all these different tools and like learning things and like getting a lot of productivity and then showing everyone all their employers like coworkers, like how they're you can be more productive. But the problem with a lot of these like really small startups is are founded by like pretty like young people who like might not have a security background. And so, like a lot of there might be some security vulnerabilities that they may not be aware of. And so there are some security incidences that are pretty pop like famous um that occurred because like one IC or like one sales rep like purchased a vendor that had a security incident and like it can have a really big impact. And we're so where it makes it so like the entire company then has to get locked out. But yeah, I I think more of those incidences are going to happen and then the buying power will go back to being top down. And it obviously depends on like how big the company is. Like if it's a big company, it will always be top down, but like things are shifting quite a bit every day.

SPEAKER_00

Yeah. I mean, when the playing field's kind of level set and everybody's trying to learn, there's so many different solutions. It does make sense to kind of go almost bottoms up or a little bit more evil because there's typically more ICs and like more people can also do that research. You're so limited in terms of like how many hours are in a day if it's top down. So that that does make sense. It'll be interesting to see how that pattern evolves. I know an interesting thing when we talk about shifting all these functions to be more greatly agentic is obviously the maintenance of it, can't get away from it. As you shift your marketing team and it's functioning like an engineering team, and no doubt have shifted other functions also. Like, how are you maintaining all these agents?

SPEAKER_01

I think it's just if you're using them every day, you start noticing if things are wrong, and then you just end up having to just be like, okay, change this. We haven't found it to be that hard. I think mostly because a lot of our agents are not in like traditional workflow builders. I think if there's like a component of like an agent historical, not historical, but you know, like um, yeah, like an older agent builder, it becomes a lot harder to maintain it versus now a lot of agents are just like prompts. Like you can just give like a list of prompts and it just occurs on like a loop. Um, and that's just as powerful as having like a historical like agent builder. Um, so I think because a lot of our workflows are just like in that prompt, it's so easy for us to be like, oh, you messed up this, just like change it a little bit and then it'll it'll it'll adjust. So the maintenance is a lot lighter versus like the with the old school like UI, that that's where it becomes really hard.

SPEAKER_00

Got it. So it's kind of up to what I'm hearing is it's up to the individual running that workflow to maintain their agent.

SPEAKER_01

Yeah, like you should see if it messes up. Yeah.

SPEAKER_00

That's good. I mean, in a way, it's like that is the biggest learning leak too. I think it's valuable to have dedicated teams from a maintenance perspective. And there's certainly a ton of pros, but also people don't learn, it's harder to create agents when you're not actually seeing the other side too.

SPEAKER_01

Yeah, totally. I agree.

SPEAKER_00

Shensi, to be able to hire people that are, of course, maintaining their own agents, but doing all this cutting edge work and staying abreast of all the different shifts in AI, you have to hire like A plus people. And there's a line from Christina Cordova who came on the podcast. She's uh CEO of Long Year now. Yeah, I I heard you reference her. She talked about it on our podcast on GTM now. And she's talked about mercenaries and missionaries. And I'll maybe quickly read a quote from her A common mistake leaders make is hiring mercenaries and being surprised when they act like mercenaries. If you're recruit by overpaying and selling off hype, you're just outbidding the competition for people who prioritize being outbid. When the tide turns, and it very often does, mercenaries don't suddenly become missionaries. They just find a new ship that's still rising. How has this impacted the way that you hire a merch?

SPEAKER_01

I think it's harder than ever to avoid hiring mercenaries. I think there were a lot of missionaries that became mercenaries, especially in this current environment where there's just so much fundraising, people are just throwing money at everyone. And I think the way to avoid it is just by being really upfront about like how you are different and what they can expect. And if someone is optimizing for compensation, you just can't hire them. So for example, if someone's coming, someone, it's always been true where like if you are working in big tech, you'd make more money like upfront than working at a small startup. Of course, like if startup hits, then yeah, you can make a lot more, but like that's not guaranteed. And like a big part of like life's journey is like having a mission and like working towards something and like having like a lot of really great stories, and like you're way more likely to have that at a startup versus a big company. And there is a trade-off for that. And I that's no different from now. Like anthropic and open AI are not small startups. Like obviously they can pay a lot more, you know? And so um it's so if someone wants to choose to work there, like they show that they should work there, but they're probably not looking to work at a small startup. It's just a very, very different experience. And so I I think we're just very, very upfront, and we continue to be very strict. Like, what kind of companies are you looking at? Like, are you looking at companies where in our size, bigger, smaller? Um, because also like we are um like series B, we're not like seed too. And some people also want like seed. Um, so really trying to understand like why does this person want this stage? Like, what are they looking for? Can we give it to them? And if they're looking all over and then also really looking at big companies, it's very unlikely that they're going to come. And so we just have to be a lot more aggressive about filtering.

SPEAKER_00

Totally makes sense. And uh we see it in our hiring processes also, whether we're helping portfolio companies or hiring at the firm itself. But it's crazy how upfront and honest people are. Like I find just by asking them initially what they're looking for in their career next before you even get into the job stuff. It's like people will just tell you. And often that will help differentiate whether they are truly a missionary or not.

SPEAKER_01

100%. And I think a lot of these companies, they were okay with just having mercenaries because mercenaries will have really good resumes. Like mercenaries will have worked at the hottest companies, they will speak of really great and they are good. Like mercenaries are usually good at their jobs too, but like they are not going to stay somewhere where it's going to be really hard. As she mentioned, like if there is a speed bump, they're going to start looking for like the next easy place. And so I think one thing that I noticed is like usually if you raise a really big ground, that's when a lot of people will come because they think it will become really easy. And it won't be always like that. Yeah, it won't always be like that. And so that's something that I'm like very hyper aware of because you need to figure out like what the incentives of those people are.

SPEAKER_00

So it sounds like be super upfront, figure out their incentives, and that's like how to actually dispel whether they're a mercenary or missionary.

SPEAKER_01

Yeah, and almost like anti-cell, like be like, this place sucks. Like you have to work a lot, like you know, it's gonna be like rocky sometimes. Like, do you are you sure you want to be here? And so I think really like that anti-cell sometimes will end up like sharing a lot of information about like, are they really into this? Or is it just like, I don't know, they just want like a job offer. Yeah.

SPEAKER_00

Yeah. And that might uncover some underestimated talent that ends up being the best possible talent you could have.

SPEAKER_01

Yeah, and I think it's becoming even it's kind of like it's exactly that. Like it's it's becoming even more important now to find underestimated talent versus like before. Um, and that's it's always been true. Like you could find underestimated talent, like you needed to find underestimated talent, but like now, like the obvious talent is very much gone. Like the very obvious talent, like will be aggressively recruited by like these large AI labs because like they are looking for a lot of people, obviously. Um, and so it's you need to find the people where like they would probably pass on them because they don't know these companies, these people are really young, might not have the experience that they need, or you give them like a much bigger role than they would be able to get at like a bigger company. So yeah, it's definitely like a more of a search, but like, you know, if you do it right, you can find really great people. Definitely.

SPEAKER_00

Yeah, I think from talking to people like Christina or interviewing people like Christina or Christinian or Chris Lee, you know, there's certain patterns emerge. And I love Chris Lee. He was the head of sales at Deal, one million to 1.5 billion. His example is, you know, he was at a previous company and such a painful kind of pain point or such an acute pain point around hiring and the timeline that it took. I think it took something like nine months instead of two months that he was in love with the problem. And so he's like been so passionate about the problem. And obviously that bleeds into the sales process into everything you do, but it also like holds you in this missionary state. Um, you're not just kind of looking for the next ship to jump on. Totally. That's also yeah. Well, Shenzi, I want to zoom out. We've been in the weeds of your go-to-market. Let's take a bit of a zoomed-out market view because you sit in the middle of pretty much everyone's stack. And so you see roadmaps and partnership plans really before they're announced or if they're even ever announced. What are some of the biggest shifts that you're seeing across your customers right now?

SPEAKER_01

I think there was a lot of fear last year around the large AI labs taking over like the entire application layer. And a lot of these traditional SaaS companies were starting to lock down their partnerships because they were nervous about all their data getting pulled out and becoming like obsolete. I think it be has become more clear that that's not really what's going to happen. And a lot of these companies are becoming more comfortable developing partnerships now and like opening back up and also developing their own AI products. I think overall, long term, everyone's going to compete with everyone. It's just the way it is. And now like AI has made it to the barrier to entry for a new like product or like new market is so low that everyone's gonna compete everywhere. So you also need to figure out how to collaborate everywhere as well. Um, so I think honestly, like last year was a lot of hostility when it came to like the ecosystem and like how people could collaborate. This year, a lot more people are realizing like, okay, we're gonna have to work together because like no one's going anywhere. So how do we end up doing that in like a more productive way? Which I think is good. Last year was looking a little bit scary for like where, you know, integrations were coming, but like obviously this year, not the case.

SPEAKER_00

Yeah, that that's great. That totally makes sense. And I mean, as part of that, you get to you have the privilege of seeing numbers that that kind of are are behind the scenes. Some companies they're raising at a hundred X, a thousand X. No name to name names, but do you see that there's a reckoning coming? Or how do you feel about kind of the valuation stages that we're at right now?

SPEAKER_01

Yeah, I think it does remind me. So we started the company in 2020 and we had a lot of friends who had started companies around the same time. And in 2021, um, we knew like what some of their revenue was, and they were raising at like huge, huge valuations, like one to two billion with like not a lot of revenue. And they never caught up to that valuation. Um, and so I think just seeing now like a lot of these companies and I know what their revenue is and like what they're raising at, like it's going to be tough. And I think it's it's again like a lot of mercenaries will join those companies because they're like, oh my gosh, like I'm gonna be rich. Like it's gonna be so easy for me to make a lot of money, but there has to be an exit. And like right after a value like fund big fundraise, it's like probably the worst time to join. Oh, so yeah, it'll be interesting to see what ends up happening with a lot of that.

SPEAKER_00

Yeah, super interesting. And Shensey, you know, you yourself, you're a busy, busy founder, but you were directly involved in the actual build of agent handler and gateway. And of course, you have a fantastic CTO. You guys at you have known each other since you were 18 years old, but like, why were you so hands-on in the process? What's what's part of that strategy?

SPEAKER_01

So the first reason why we were so hands-on was because we didn't have enough engineers, and so my co-founder and I needed to help out with that. But I think it ended up becoming an extremely valuable experience because we, to be honest, did not fully understand the power of AI with coding back then when we were using AI on our original code base. Um, that was human build. Um, it was much scarier for us to use AI on a product that had so many really large customers. Uh, we didn't want to break anything. We were just really nervous. And so we could only see like the power of AI for like really small changes versus like a large structural way. And so when we were building zero to one, it really, really, really transformed how we felt about AI and like what we how how we knew what was possible. And because Gil and my like co-founder and I could understand that, we were able to really implement it top down. Um, I think if the founders don't really understand like how AI can make a big dip impact on the business, then it's impossible for the team members to really evangelize it across the company without support that way. Um, and so it took us six months to start building that product um end to end. Uh, we probably did like three months of research while also coding a little, like building a prototype and presenting it. Um, but for Gateway, it took us three months, um, like start to finish. And it was really AI just shorter because AI kept getting better. AI was even better than back then. And so it just like that zero to one really taught us like where things are now. Like uh, we had built a lot of things back then that we felt like were really advanced, and we were able to just rip it out and use a new model to do things that back then and it just allowed us to move so much more quickly. We were also able to experiment with like new new things that were coming out. Like we were able to try skills and like all these like different things that were like, you know, and like building like new agents and like work trees um where it would have been a little bit harder on our older code base. Um, and so I think because like we were able to learn in that way, we we could encourage our team to learn and we could also teach them. And so I really, really, really encourage founders who started products before like AI got really good to build something zero to one because it really is such like a it really transforms like how you think about your own company and like AI training and AI understanding that will make your company better overall.

SPEAKER_00

That is fantastic advice. And you're staying close to the code and and to AI to really stay abreast of everything. What's next with AI?

SPEAKER_01

We have obviously a lot more to build. Um, so we're continuing to bulls a lot more. Um with this really exciting crowded launch this week uh that we're going to be working a lot with different customers for. And yeah, I think what increasingly is going to happen too is right now a lot of people are starting to realize that models or that there's like a there's no like perfect model for all use cases. And people are starting to think through like which model is the best for each task that I'm trying to perform. And that is a really difficult operational challenge. So, like, how do I end up like choosing the model? And then how do I end up making sure it's selected upon the prompt or like upon the task? But increasingly, I think what people are going to start noticing is like, okay, well, everyone is starting to use these models for like these different tasks. Like, how do I have my own edge? So I think people are going to start experimenting with uh probably training their own models off of these open source models with some data that they have internally, not for everything, but for very, very specific things. Um, and that will become like a part of the equation for what is the best model for each task and what might be like your own homegrown model. I think we're very early still. It's probably not this year, probably next year. Um, but it'll be interesting to see how everything develops, especially as things hopefully get a little bit cheaper.

SPEAKER_00

Yeah, super, super fast. I mean, that's like taking your data as a moat to a next new level of unlock.

SPEAKER_01

Yeah, for sure. And this will also be a way for a lot of these like traditional SaaS players to have some kind of edge, um, especially since the market has been so aggressive towards them more recently.

SPEAKER_00

Yeah, yeah, definitely. And where can people follow along with you and merge?

SPEAKER_01

Yes, obviously, you can find us at merge.dev. Um, you can also find me at Twitter at Shenzi.

SPEAKER_00

Um, and you can also uh find me on LinkedIn as well, Shensi Dang Merge. Amazing. Well, Shenzi, thank you so much for the time and conversation. This has been fantastic. Appreciate it. Thanks for having me.