Listen on YouTube // Apple // Spotify
Thank you to our sponsor partner:
Primer unlocks B2C ad channels for B2B marketers with precision targeting and next-gen measurement.
There are two ways you’re wasting ad budget: (a) you pay to reach people who will never buy, and (b) 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. No more burning budget on students, competitors, and tire-kickers.
You get unlimited website reveals: every account researching you, identified – no caps, no per-reveal pricing. See who is actually in-market and feed it straight back into targeting.
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
Who you’ll learn from
Most companies start self-serve and grind their way up to enterprise. Shensi Ding did the opposite, then reversed course again. The co-founder and CEO of Merge joins Sophie Buonassisi on 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
Episode highlights
1:14 – What Merge does: products, workforce, and three core offerings
2:05 – Agent Handler, Gateway, and the new Embedded Routing Stack
4:52 – The 2026 shift away from “token-maxing”
7:29 – 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
Key takeaways
1. Merge ran the GTM playbook in reverse: enterprise first, PLG second.
Three years ago, Merge made the go up-market decision to go up-market while enterprise was greenfield and build the security and depth that make you hard to replace. It worked. Most revenue now comes from the 400+ enterprise logos like Mastercard, JP Morgan, and Amex. Only now is Merge adding a self-serve layer back on top, because AI buying got consumerist and developers need to know you, not just the C-suite.
2. Tokenmaxxing is so 2025 – 2026 is about intentional spend.
Last year everyone defaulted to the newest, biggest model because the cost was subsidized and experimentation was the point. Now teams are trained, spend adds up real quick, and open source has caught up. The unlock is smart routing: send more simple tasks to Haiku, not Opus 4.8, and save the expensive models for the tasks that need them.
3. Where a model is built and where your data lives are two different things.
The stigma around Chinese open source models conflates origin with hosting. A model built in China can run on a US host (Baseten, Bedrock, Fireworks) with US servers, and customers can block routing to China entirely. Shensi works directly with ByteDance, Alibaba (Qwen), and Kimi, with data retention terms and US fallback hosts built in.
4. Forward deployed engineers are the enterprise cheat code.
Merge’s product fit 99% of enterprise use cases, but the last 1% blocked adoption. Forward deployed engineers closing that gap grew usage 5 to 10x almost immediately, then the pattern repeated across accounts.
5. Treat marketing like an engineering function and productize it.
When product started shipping faster than marketing could keep up, Merge broke marketing into repeatable, automated actions: changelogs pulled from GitHub, launch images and videos auto-generated in the brand voice. The standout is a dinner agent handling outreach lists, closed-lost accounts, and seating charts. Field dinners went from 1-2 a quarter to 3-5 a month!
Shensi Ding
- LinkedIn: https://www.linkedin.com/in/shensiding
- X: https://x.com/shensi
- Merge: https://merge.dev
Sophie Buonassisi (Host)
- LinkedIn: https://www.linkedin.com/in/sophiebuonassisi
- X (Twitter): https://x.com/sophiebuona
- Newsletter:
GTMnow
- LinkedIn: https://www.linkedin.com/company/gtmnow
- X / Twitter: https://x.com/GTMnow_
- YouTube: https://www.youtube.com/@GTM_now
- Instagram: https://www.instagram.com/gtmnow_
- TikTok: https://www.tiktok.com/@gtmnow_
The GTMnow Podcast – subscribe to support
It would mean the world to us if you subscribe to the show on your platform of choice. This helps support our growth, which in turn helps us continuously bring on bigger guests and deliver the content that you want to hear most.
The GTMnow Podcast shares how the best in tech build, scale and invest.
Visit gtmnow.com for more episodes, The GTMnow Newsletter editions, and other content.
GTMnow is run by GTMfund, an early-stage venture firm made up of 350+ go-to-market executives from the fastest-growing companies.
GTM 199 Episode Transcript
00:00 – 00:11
Shensi Ding: Are companies like Mastercard even more yet. 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. Shinsei ding.
00:11 – 00:22
Sophie Buonassisi: Co-founder and CEO of Merge You power logos like OpenAI, perplexity, Netflix, Uber, Mistral, just to name a few. And you’ve raised $75 million from Accel.
00:22 – 00:23
Sophie Buonassisi: NEA. In addition.
00:23 – 00:27
Sophie Buonassisi: What have you really learned as a founder through this process growing in a first time founder?
00:27 – 00:36
Shensi Ding: 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.
00:36 – 00:49
Sophie Buonassisi: The UN enterprise fairly early, and then now you’re actually almost going the opposite way and shifting back to appeal motion. And you’ve got about 20,000 Feltre or over 400 enterprise customers. How do you think about the split?
00:49 – 00:52
Shensi Ding: I remember thinking movie prize was really hard, but this is also very hard.
00:52 – 00:54
Sophie Buonassisi: What’s next with AI?
00:54 – 01:14
Shensi Ding: 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.
01:14 – 01:16
Sophie Buonassisi: Welcome to YouTube. Now.
01:16 – 01:18
Shensi Ding: Thanks for having me.
01:18 – 01:28
Sophie Buonassisi: Absolutely. It’s great to have you here. Super excited to dive in. And for anyone unfamiliar with merch, maybe let’s start with just a little bit of context and background on merge itself.
01:28 – 01:29
Shensi Ding: So we.
01:29 – 01:30
Sophie Buonassisi: Have two offerings.
01:30 – 01:48
Shensi Ding: Merge for products and merge for workforce. 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.
01:48 – 01:52
Sophie Buonassisi: Amazing. I’m sure the governance part is blowing up and we’ll kind of get into that.
01:52 – 01:55
Shensi Ding: But that’s definitely popular.
01:55 – 02:03
Sophie Buonassisi: 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. Yes.
02:03 – 02:30
Shensi Ding: So we started the company with our integrations for products around six years ago when we were noticing that there was a lot of fragmentation in the market, 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 Brad got really popular in 2 to 3 years ago, and then more and more people were talking about contacts and how being able to sink a lot of data would really help make AI search really easy.
02:30 – 02:51
Shensi Ding: So we adjust our product quite a bit to make sure that as I search got more popular, we are really like the premier partner for a lot of those AI search products. So that’s how we end up powering integrations for companies like OpenAI, Mistral, and it’s been really exciting, kind of growing with that. And around year and a half ago, we started noticing there was a new segment of the market that was growing, which was integrations for agents.
02:51 – 03:08
Shensi Ding: And our first product didn’t quite fill in the gap for what? For what like that you need was. And we felt like it was our right to win in something that we knew a lot about. And so we ended up launching a new product called Merge Agent Handler, where we are able to help provide tools for these agents to third parties.
03:08 – 03:29
Shensi Ding: We handled the OAuth refresh logic. We helped make sure the quality is really good. We also make sure that especially when it comes to like the different edge cases that could occur, we’re able to maintain all of it as well, especially as APIs change. And there’s new permissions that may come up, especially with different personas or really that right, that partner that works with a lot of our customers as we’re offering integrations to their customers.
03:29 – 03:53
Shensi Ding: And then we launched Merge Gateway around two months ago, which is Smart Limerick. 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, while last year that was not a concern at all.
03:53 – 04:10
Shensi Ding: 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 smart routing inside of your own product, you would have that logic for yourself and it would apply to all of your customers.
04:10 – 04:26
Shensi Ding: 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 your AI usage in your product. So essentially allowing you to like white label our functionality inside of your product, which is really exciting.
04:26 – 04:48
Shensi Ding: We’ve noticed a lot of customers obviously passing along tokens, their customers as our pricing, which has caused a lot of companies to want to bring their own tokens, especially bring their own API keys, especially as he may have some contracts that will also be available to. So just overall giving a lot more power to the end user or our customers customer with how they which models they want to use and how they’re using it.
04:48 – 04:52
Sophie Buonassisi: Very cool. And was it a tough decision to make the white labeling?
04:52 – 05:12
Shensi Ding: There’s a lot of resource constraints obviously at a startup. And so it’s kind of like deciding like what’s a target first? Like, do we end up doing this first or that first? I think what we were really noticing was that the market was really moving this direction, where people is just becoming more top of mind. Obviously, it was becoming top of mind for our customers, but for their customers to is becoming really of important.
05:12 – 05:27
Shensi Ding: Like how do we end up reducing our AI spend not only like our direct model 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 are offering yet.
05:27 – 05:47
Sophie Buonassisi: 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 the leader boards and everything, it really feels like a 2026. We peeled that back from 2025 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?
05:47 – 05:48
Sophie Buonassisi: Adapt it now.
05:48 – 06:06
Shensi Ding: Especially last year or as people were experiencing more with AI like I think last year was where there’s a lot of experimentation and this year is where people are a lot more comfortable and they’re really starting to figure out how to optimize what they’re doing more. Last year, when people were experimenting, they would always just default to the most, like most modern model.
06:06 – 06:23
Shensi Ding: Like if I was using an AI assistant, I would always choose the most recent model that I heard of, whether it was like 4.6 or 7 or like GPT 5.4, I think, or 5.3. I would always just use the latest model that people were talking about. And we’re we’re thinking like we’re saying really good things about just because I wanted to have the best experience possible.
06:23 – 06:38
Shensi Ding: And it didn’t really matter because the AI system was just eating the cost and whatever. 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 just want everyone wanted to just use them. It was modern US thing.
06:38 – 07:00
Shensi Ding: But I think especially now that we’re seeing a lot of open source models really catching up as 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. My entire team is trained. It’s not like an issue where only X percentage of my team is trained in AI now, and I need to get as close to 100% as possible.
07:00 – 07:16
Shensi Ding: Now you’re kind of like at that 100%, hopefully. 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.
07:16 – 07:29
Shensi Ding: They’re becoming more and more accessible. There’s a lot of stigma against open source models, 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.
07:29 – 07:49
Sophie Buonassisi: Definitely, yeah. And while we’re on the topic of open source, a huge one, you see it everywhere right now. We ourselves have invested in open source companies that are doing phenomenally too at GTM fund. And so 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.
07:49 – 08:04
Sophie Buonassisi: 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 geographic tensions too. So a lot.
08:04 – 08:21
Shensi Ding: Of these research labs are based in China, and I think Deep Six 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 was a lot of news around them, like, oh, China had like ended up doing a lot of distillation on the using like our US models.
08:21 – 08:40
Shensi Ding: And it was like cheating and it was a people felt like it was like very sketchy. And so I think just because there was so much conversation around it and like a big lack of trust around deep seek that it created just like more stigma around like any models that came from China, they were probably done like maybe they were good, but like I didn’t want to be associated with it.
08:40 – 09:01
Shensi Ding: And I remember having a lot of conversations with our enterprise customers where they would be like, deep seek like you like. And 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.
09:01 – 09:21
Shensi Ding: 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 leaving the US, or going to any regions that are outside of any regions where you don’t feel comfortable. You can always control that. And like the type of host that you use, especially if you use like us hosts that are, you know, that are hosting those models.
09:21 – 09:47
Shensi Ding: So I think there’s a lot of confusion, 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, you know, the CEO leaders of like anthropic and, and OpenAI, 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 as well.
09:47 – 09:49
Shensi Ding: Be really interesting to see how it plays out.
09:49 – 10:13
Sophie Buonassisi: Yeah. And we think, you know, the model might be built overseas, but if it’s running and say the US, like where the model comes from and where your data goes or 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, Reverend, to open source yourself, but you are actually working with a lot of open sourced kind of models.
10:13 – 10:15
Sophie Buonassisi: How are you working with them?
10:15 – 10:37
Shensi Ding: Yes, in a couple of different ways. We do have direct relationships with some of these partners. So for example, like we work with like the Bidens team Alibaba Quinn. So we have really oh sorry Alibaba. And then also Kimi as well. So we’ve had we’ve had really great relationships with a lot of these different model providers. And we will print their data retention contracts with them as well.
10:37 – 10:56
Shensi Ding: A lot of them do have servers in China. And so we do have the ability for 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 it then routes to a US host, potentially, that is hosting the open source model with US servers.
10:57 – 11:16
Shensi Ding: So we also work with like, you know, base ten, Amazon, Bedrock fireworks. A lot of like us hosts that will end up like also offering like these these models as like offerings. So there’s a lot of different fallback options here too. So it’s a combination of like direct partnerships where they’re hosting it themselves. And then also we’re working with like different like host providers in the US.
11:16 – 11:24
Sophie Buonassisi: 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.
11:24 – 11:43
Shensi Ding: Yes, for sure. And even if you’re mostly focused on like using OpenAI or anthropic or 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 open 4.8. And sometimes, like if someone’s asking like hello or like OnePlus one or like thank you.
11:43 – 11:51
Shensi Ding: It could probably go to haiku. It doesn’t necessarily need to go to opus 4.8. So even within like within that we can still help a lot.
11:51 – 11:53
Sophie Buonassisi: To super helpful for token usage.
11:53 – 11:55
Shensi Ding: Yes, exactly.
11:55 – 12:15
Sophie Buonassisi: I love it. Well, let’s talk a little bit about your growth so early early days. Let’s start there. You were perhaps nervous or a little bit scared to onboard a customer like ramp when you’re when there are about 100 employees because the product was really it was early days. But now you power logos like OpenAI, perplexity, Netflix, Uber, mistrial, just to name a few.
12:15 – 12:27
Sophie Buonassisi: 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.
12:27 – 12:54
Shensi Ding: 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 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, 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.
12:54 – 13:15
Sophie Buonassisi: 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 emotions. And you yourself have a really interesting one, because most companies will go from plug to enterprise and kind of go with that upmarket motion from there.
13:15 – 13:22
Sophie Buonassisi: But Europe looks a little bit different, like walk us through how the motion actually evolved for you and what that did and why.
13:22 – 13:37
Shensi Ding: Yeah, it’s definitely been a journey and it’s been a lot. So we first got started, as you mentioned, like we mostly focus 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 saying it or we just had a really terrible product.
13:38 – 13:55
Shensi Ding: And over time we moving to bigger and bigger companies. And about three years ago, we made a very concerted effort to move out market because 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.
13:55 – 14:18
Shensi Ding: And so we we did a lot of changes there. We ended up like changing where our team looked like doing a lot of segmentation, obviously investing a lot in our product and security posture. And so now we are actually mostly enterprise, a business like most of our larger contract, obviously our larger contracts and a majority of our revenue are companies like Mastercard’s would be Morgan Amex, like very, very large companies, which is really exciting and very cool to like.
14:18 – 14:36
Shensi Ding: Now via that phase of a company where 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. It’s becoming a lot more consumer, and a lot of these, like AI startups, are starting to focus a lot.
14:36 – 15:03
Shensi Ding: Obviously, like plug sales motions or plug like growth strategy. Are they they are being just more plug. And so therefore their marketing is a lot more like hype launch videos, making sure that people are just very aware of them and focusing 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.
15:03 – 15:24
Shensi Ding: It just felt like our old marketing motion was not quite relevant for where the company was going and where we needed to be. 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 I’m a lot more like developers and like ICS are more aware of us, rather than it just becoming like a top down awareness play.
15:24 – 15:47
Shensi Ding: 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 was really hard, but this is also very hard 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 these skill set that we don’t currently like, have.
15:47 – 15:50
Shensi Ding: And so we’ve been excited to hire a few new people who’ll be able to help us with that.
15:51 – 16:12
Sophie Buonassisi: Very, very exciting. So you win enterprise fairly early and then now you’re actually almost going the opposite way and shifting back to a plug motion or shall I say having both motion. Yeah, it would be more accurate. And you’ve got about 20,000, maybe more than 20,000 self-serve orgs on it now, which is more of a plug motion, about over 400 enterprise customers.
16:12 – 16:19
Sophie Buonassisi: 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?
16:19 – 16:36
Shensi Ding: Yeah, I think it’s hard, especially since there’s there’s a lot of opportunity costs, especially with how easy it is to build things now with AI versus before, where you really had a lot more time to like, if fewer resources for longer time to really, like, think through like, is this the right direction versus just like building it really quickly?
16:36 – 16:51
Shensi Ding: 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 over resourcing it before there’s any data. So as we’ve been, you know, growing revenue in each of these two new products, we’ve continued to invest a lot more.
16:51 – 17:18
Shensi Ding: 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 were then able to now show how we’ve evolved as a company, and they’ve been very interested in our new our products as well. And of course, like as we’re expanding into like these new products and selling into different personas and companies, they’re they’re very happy to see that we are supporting logos that they recognize and trust and that we are able to power companies of that caliber.
17:18 – 17:31
Shensi Ding: So it has extended. It has helped us quite a bit like that brand recognition of our existing customers, 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.
17:31 – 17:48
Sophie Buonassisi: And you mentioned behaviors changing like buying behavior is changing. Now when we talk about going from enterprise to for 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 consumerist too?
17:48 – 18:08
Shensi Ding: 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 difference what software could help you with certain functions was really spread out across like word of mouth, different articles, conferences, Gartner.
18:08 – 18:28
Shensi Ding: But now it’s really not there to the same degree, especially when it comes to like different AI tooling. All the conversation about different AI tooling is on Twitter. Like whenever I’m hearing about like a different like coding app or like some kind of like AI design feature or like the new models, it’s always released on socials. That’s really primarily where it comes out, obviously.
18:28 – 18:38
Shensi Ding: Like read it 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 your buyer where they’re at.
18:38 – 19:00
Sophie Buonassisi: Yeah. And I kind of feel like anthropic is is a big momentum pusher on that front to just how they’ve been releasing so heavily through these videos on platforms like X or other Llms were to I’m sure they were previously also, but just the momentum that these different model releases like one seven and 4.8 have created on social like inevitably to stay up to date, kind of have to be on Twitter now.
19:00 – 19:06
Shensi Ding: We’re at 100%. Yeah, they’re they’re really good at it. Like all the all model providers. Yeah. Yeah.
19:06 – 19:17
Sophie Buonassisi: 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 founder led sales going for quite a long time. Tell us about this a little bit.
19:17 – 19:18
Shensi Ding: A quick pause, because.
19:18 – 19:42
Sophie Buonassisi: 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.
19:43 – 20:22
Sophie Buonassisi: Walkway. 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 kickers. You get unlimited website reveals. Every account researching you identified no caps, no per reveal pricing. You can see who’s actually in market and feed it straight back into targeting and primaries, 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 GTM 15, you can go and start your trial at, say, primer.
20:22 – 20:43
Sophie Buonassisi: That’s a primer. 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. GTM 15 back to the episode.
20:43 – 21:06
Shensi Ding: Yeah, so back in the day, I would send a lot of outbound. I still send a lot of outbound, 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, and then we would download the product and then we would, you know, run the sales process until someone purchased.
21:06 – 21:35
Shensi Ding: We definitely weren’t perfect at it. And like when we ended up hiring like a real sales leader, we learned a lot about like what a good process looked like. But even even after that, like when we moved to enterprise, like things changed a lot. And our current Crow, he had a lot of process where like we had we started having a deck and like I didn’t realize what it made the deck made, but it really shows like a level of respect for the customer and like you prepared for them, that you have thought about them ahead of time, that I think the customer appreciates more than I really expected when I when we first started
21:35 – 21:52
Shensi Ding: the company. And yeah, now we have like a it’s kind of like the customer is 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, 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.
21:52 – 22:10
Shensi Ding: If you like it, like, let me know. It was my money, which obviously which works a few times, but like, not, not, not the same way. And it really is up to the salesperson to really drive the deal through. And 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.
22:10 – 22:25
Shensi Ding: And so yeah, our process has changed quite a bit. And then now again, it’s changing again. Since we have now this like 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 plug, what is this relationship look like?
22:25 – 22:28
Shensi Ding: So a lot of those things we’re trying to figure out to. Yeah. And it really does.
22:28 – 22:28
Sophie Buonassisi: Feel like.
22:28 – 22:32
Shensi Ding: Buyer behavior is shifted now where there’s so much contact, so much information that people are really looking.
22:32 – 22:33
Sophie Buonassisi: For guidance.
22:33 – 22:34
Shensi Ding: On.
22:34 – 22:34
Sophie Buonassisi: Like how.
22:34 – 22:55
Shensi Ding: 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 or it’s one of the hardest things for companies to do. I 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.
22:55 – 23:15
Shensi Ding: Like, how did you know it was time to hire that person and how did you get it right? I think we were dying. And so that’s why we knew it was the right time. I just 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 and the company is probably not going to make it, because I think sales is like inherent part of being a founder.
23:15 – 23:39
Shensi Ding: Like you have to do sales in sales, obviously, but sales and recruiting sales and fundraising sales and like, like your team so that they want to stay with you and there’s other sales and everything. So if you outsource, like your initial sales to a sales person, 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.
23:39 – 23:49
Shensi Ding: So yeah, I think you need to be at the point where you are like, wow. Like I think we have product market fit, like I’m dying. I really need more help. And someone said, really professionalize us. And I think that’s the right time.
23:49 – 23:54
Sophie Buonassisi: And you guys run a forward deployed engineer team, correct?
23:54 – 23:55
Shensi Ding: Yes. Yeah.
23:55 – 24:16
Sophie Buonassisi: So I mean, we recently published a piece actually about how companies are growing so quickly with direct sales because in the past, I think the the perception was that plug 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 forward deployed engineers, a key part of that.
24:16 – 24:46
Sophie Buonassisi: So you know, around that it’s the idea that 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 PhD teams yourself, like where’s your perspective around forward deployed engineers and how they come into the picture.
24:46 – 24:49
Sophie Buonassisi: For other companies that are also trying to grow quickly, like merge.
24:49 – 25:07
Shensi Ding: It has made such a big difference for our enterprise customers. And I do think if we had an affair deployed engineering team when we were mostly focusing on startups, it just would not resonate. So you do have to get to the point where you are selling to your price companies, where they could really use it, or you go straight to enterprise and then you can you can really utilize it.
25:07 – 25:27
Shensi Ding: But yeah, I just think you’re mostly selling 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 are noticing was it was we had this product and it was like it fit like 99% of the use case that they needed. But there was that 1% edge case that was really accustomed to them, and we could adjust our product to.
25:27 – 25:28
Sophie Buonassisi: Fit.
25:28 – 25:53
Shensi Ding: 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 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 5 to 10 x immediately after we ended up using like having our forward to put engineers like assess that account.
25:53 – 26:09
Shensi Ding: And so we were able to see like okay, so these two use cases were very successful as repeat for all these other customers that are having that same gap. Because now we’ve learned from that like little like now you’ve learned like what to do here. And so it’s really helped a lot in situations where like our customer accused me like, oh, like we really want to do this.
26:09 – 26:23
Shensi Ding: Like we really want to do this. We just don’t have resources. Or it’s like, oh, like this product isn’t quite bright 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.
26:23 – 26:37
Shensi Ding: So yeah, it really is like such a game changer for those large enterprise customers. And I think you almost have to have them for that. But if you’re mostly something to startups like it, definitely it’s not for you and it will save you either. Yeah. Exactly.
26:38 – 26:45
Sophie Buonassisi: Yeah. And you can probably pattern match across too, and take more of a proactive approach of going to customers as suggestions. It’s incredibly valuable at the enterprise stage.
26:45 – 27:00
Shensi Ding: 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 value that resource a lot more than if it’s like your first time. And so, you know, take that into account. Drink sales.
27:00 – 27:11
Sophie Buonassisi: That is great advice. And now you’ve 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 Dakota market side of the house?
27:11 – 27:29
Shensi Ding: Yeah, I will say it’s like definitely still hard. I think historically what we really were lucky with, 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 had a little bit more. Obviously, that shifts a lot, especially if you have like new people, you have to keep like training with new things.
27:29 – 27:45
Shensi Ding: So we’ve been adjusting what our onboarding training looks like every month or so. So make sure that we’re continuing to keep up with the times. And also like using AI as much as possible to make it to our team isn’t 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.
27:45 – 27:58
Shensi Ding: Like it’s just a lot to remember. And so I think we’re trying to figure out like what are the soundbites for people to really remember and how do we loop in like S’s and specialists like when, when we need to go into the deep dive because it is like we do have a lot of offerings and a lot of products.
27:58 – 28:01
Shensi Ding: So yeah, it will get hard.
28:01 – 28:25
Sophie Buonassisi: Yeah. And I mean a lot of launches obviously, you know, very concluded. But when that happens, you know, there’s this shift that we’ve seen at least we’re product used to be more of the bottleneck. And then as velocity has sped up things to AI, notably on the right side, now the bottleneck is actually your go to market. And companies are working on solutions for that and implementing really kind of smart, interesting, innovative solutions.
28:25 – 28:33
Sophie Buonassisi: 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?
28:33 – 28:53
Shensi Ding: 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. It’s 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 is really starting to feel this in the market.
28:53 – 29:10
Shensi Ding: And so we started trying to figure out like, okay, how do we end up making like more marketing, more of like an AI, 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 keep up in pace and we need to be able to like productize this a lot more.
29:10 – 29:24
Shensi Ding: And so we started thinking through like what were repeatable parts of the product, like what we’re repeatable actions that we do for marketing. It’s like changelog okay, let’s pull it straight from the GitHub. Let’s like output this immediately and like add like a voice to it, our socials and like our different like new feature releases. It just pull straight from our GitHub.
29:24 – 29:43
Shensi Ding: And like obviously if someone does like a quick review. But now a lot of like the images, like the videos are all like auto and partnership images are all auto generated and it has like our voice, like we’ve continued to like add more and more. It’s it’s gotten better and better, 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.
29:43 – 30:02
Shensi Ding: And then, yeah, we’ve also like made like our even like our field events. More AI too. So like our dinners, we used to be really, really hard to coordinate them because we 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 so hard to fill a dinner, and it would just be a lot of time for, like me, like a team.
30:02 – 30:25
Shensi Ding: Like we don’t have to the city and charter coordination. It was just a lot. And so one day I was just like, okay, I’m just gonna like make an agent that. And so we’re putting everything that we ended up doing for every single dinner into like one workflow, and we’re just going to productize this. And so we went from having like 1 or 2 generous reporter to now having like 3 to 5 a month, because everything that we ended up doing for like outreach, which is like, who have we met, you know, in this city in the past, like 3 to 5 years.
30:25 – 30:39
Shensi Ding: Okay. Let’s like invite them. Do we have any clothes or lost opportunities. So let’s keep that open. Let’s keep that into account for whether or not we reach out to them or not. 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 to or is the dinner full?
30:39 – 30:55
Shensi Ding: Okay then let’s not by any more people who should sit next to each other, probably startups to startups and like bigger companies with bigger companies. Partnerships with partnerships in this evening chart too. And so all of these things that we just kind of like new intrinsically from like organizing these events, we just productize it. And so it’s made our marketing, like so much better.
30:55 – 31:05
Shensi Ding: And so there’s so much more, though, like marketing is obviously like a huge function. But we’re continuing to try to figure out what are areas that we can continue to automate as much as possible. And it’s helped us a lot.
31:05 – 31:24
Sophie Buonassisi: Yeah, that’s that’s fantastic because I mean, anyone that’s done field marketing or, or organized it in or even in itself understands the pain that it can be. So that’s fantastic. 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?
31:24 – 31:45
Shensi Ding: I think we’ve just mostly focused on making like the team really AI native, so investing a lot and like training. 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 or content marketers and like really focus more on like brands a lot more than we ever did and like how to look cool.
31:45 – 31:58
Shensi Ding: So I think that’s 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.
31:59 – 32:09
Shensi Ding: And so of course, like four years ago, like you wanted to buy the most trustworthy, secure thing possible. And so yeah, we were just we’re really having to adjust based off of that, more so than just like AI.
32:09 – 32:13
Sophie Buonassisi: Yeah. Why do you think that’s happening? It’s such an interesting phenomenon.
32:13 – 32:38
Shensi Ding: 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 icy 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 like coworkers, like how you can be more productive.
32:38 – 32:59
Shensi Ding: But the problem with a lot of these, like really small startups that 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 like, famous that occurred because like one IC or like one sales rep, like purchase a vendor that had a security incident and like, it can have a really big impact.
32:59 – 33:15
Shensi Ding: And so where it makes it like the entire company that has to get locked out. But yeah, I think more of those incidences are going to happen. And then the buying power will go back to being top down and all these like how big 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.
33:15 – 33:34
Sophie Buonassisi: Yeah. I mean when the playing field is 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 effort, because there’s typically more ICS than like more people can also do that research. You’re so limited in terms of like how many hours in a day if it’s top down.
33:34 – 33:52
Sophie Buonassisi: So that doesn’t make sense. It’ll be interesting to see how that pattern evolves I know it interesting thing, when we talk about shifting all these functions to be more greatly agents 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.
33:52 – 33:54
Sophie Buonassisi: Also, like how are you maintaining all these agents?
33:54 – 34:21
Shensi Ding: 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 a, 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.
34:22 – 34:40
Shensi Ding: And that’s just as powerful as having like a historical like agent builder. So I think because a lot of our workflows are just like in that prompts, 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 adjust. So the maintenance is a lot lighter versus like with the old school like UI that that’s where it becomes really hard.
34:40 – 34:45
Sophie Buonassisi: 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.
34:45 – 34:49
Shensi Ding: Yeah. Like you should see if it messes up. Yeah.
34:49 – 35:03
Sophie Buonassisi: That’s good. I mean, in a way that is the biggest learning like to 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 to that.
35:03 – 35:04
Shensi Ding: Totally I agree.
35:04 – 35:21
Sophie Buonassisi: And so 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+ people. And there’s a line from Christina Cordova, who came on the podcast she’s CEO of.
35:21 – 35:21
Shensi Ding: Yeah.
35:22 – 35:44
Sophie Buonassisi: I heard you referenced her. She talked about it on her podcast on GTM. Now she’s talked about Mercers 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 on hype, you’re just outbidding the competition for people who prioritize being outbid.
35:44 – 35:55
Sophie Buonassisi: 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 is this impacted the way that you hire a merch?
35:55 – 36:13
Shensi Ding: 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 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.
36:13 – 36:17
Shensi Ding: And if someone is optimizing for compensation, you just can’t.
36:17 – 36:17
Sophie Buonassisi: Hire them.
36:17 – 36:39
Shensi Ding: So for example, if someone’s coming someone, it’s always been true. Or like if you are working in big tech, you would make more money, like up front 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. 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.
36:39 – 36:53
Shensi Ding: And there is a trade off for that. And I that’s no different from now. Like anthropic and OpenAI are not small startups. Like obviously they can pay a lot more, you know. And so it’s so if someone wants to choose to work there, like they show that they should work there.
36:53 – 36:53
Sophie Buonassisi: But.
36:53 – 37:09
Shensi Ding: They’re probably not looking to work at a small startup. It’s just a very, very different experience. And so I think which is 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? Because also like we are like series B, we’re not like C to.
37:09 – 37:24
Shensi Ding: And some people also want like seed. So really trying to understand 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.
37:24 – 37:47
Sophie Buonassisi: Totally makes sense. And we see it in our hiring processes also whether we’re helping portfolio companies or hiring it 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. So if it’s like people will just tell you and often that will help differentiate whether they are truly a missionary or not.
37:47 – 38:11
Shensi Ding: 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 works at the hottest companies. They will speak really great and they are good, like mercenaries are usually good at their jobs too. But like they are not going to say 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.
38:11 – 38:31
Shensi Ding: 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 will always be like that. And so that’s something that I’m very hyper aware of, because you need to figure out what the incentives that those people are.
38:31 – 38:38
Sophie Buonassisi: 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.
38:38 – 38:56
Shensi Ding: Yeah. It almost like anti sell like be like, this place sucks. Like you have to work a lot like you know it’s going to be like Rocky sometimes. Like are you sure you want to be here. And so I think really like that anti sell 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 if they just want like a job offer.
38:56 – 38:57
Shensi Ding: Yeah.
38:57 – 39:02
Sophie Buonassisi: Yeah. And I might uncover some underestimated talent that ends up being the best possible towns you could hire.
39:02 – 39:26
Shensi Ding: Yeah. And I think it’s become even. It’s kind of like that. Like it’s become even more important now to find underestimated talent versus like, before. And that’s always been true. Like, you could find underestimated. Like you needed to find underestimated help. But like now, like the obvious talent is very much gone. Like the very obvious tile like will be aggressively recruited by like, these large AI labs because like they are looking for a lot of people.
39:26 – 39:41
Shensi Ding: And so it’s you need to find the people where like they would probably pass on them because they don’t know what 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, like a bigger company. So yeah, it’s definitely like a more of a search.
39:41 – 39:44
Shensi Ding: But like, you know, if you do it right, you can find really great people.
39:44 – 40:08
Sophie Buonassisi: Definitely. Yeah. I think from talking to people like Christina or people like Christina or Christine or Chris Li, you know, there’s certain patterns emerge. And I love Chris Lee. He was the head of sales a deal one 1 million to 1.5 billion. His example is, you know, he was at a previous company in such a painful kind of pain point or such an acute pain point around hiring in the timeline that it took.
40:08 – 40:25
Sophie Buonassisi: And 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 honestly, that bleeds into the sales process, into everything you do, but also like holds you in this missionary state. You’re not just kind of looking for the next ship to jump on.
40:25 – 40:26
Shensi Ding: Totally. That’s also.
40:26 – 40:45
Sophie Buonassisi: Yeah. Well, 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 stack. And so you see roadmaps and partnership plans really before they’re announced or if they’re even ever announced, where some of the biggest shifts that you’re seeing across your customers right now.
40:45 – 41:04
Shensi Ding: I think there was a lot of fear last year around the large AI labs taking over, like the tire 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 obsolete. I think it has become more clear that that’s not really what’s going to happen.
41:04 – 41:21
Shensi Ding: 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 so the barrier to entry for a new product or like new market is so low that everyone’s going to be everywhere.
41:21 – 41:23
Shensi Ding: So you also need to figure out how to collaborate everywhere.
41:23 – 41:24
Sophie Buonassisi: As well.
41:24 – 41:39
Shensi Ding: 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 going to have to work together because like, no one’s going anywhere. So how do we end up doing that in a more productive way, which I think is good.
41:40 – 41:46
Shensi Ding: Lawsuit was looking a little bit scary for like where, you know, integration is requiring. But like obviously this year. Not the case.
41:46 – 42:06
Sophie Buonassisi: Yeah. That’s great. That totally makes sense. I mean, as part of that, you get to you have the privilege of seeing numbers that that kind of are behind the scenes. Some companies are raising at 100 bucks, 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.
42:06 – 42:22
Shensi Ding: Right now? Yeah, I think it does remind me. So we started the company 2020, and we had a lot of friends who had started companies around the same time. And in 2021, we knew like what some of their revenue was and they were raising it like huge, huge valuations, like 1 to 2 billion with like not a lot of revenue.
42:22 – 42:37
Shensi Ding: And they never caught up to that valuation. 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.
42:37 – 42:37
Sophie Buonassisi: They’re like, oh my gosh.
42:37 – 42:50
Shensi Ding: Like I’m going to be rich. Like it’s going to be so easy for me to make a lot of money, but there has to be an exit. And like right after that, like big fundraisers, like probably the worst hacked of. So yeah, it’ll be interesting to see what ends up happening with a.
42:50 – 43:07
Sophie Buonassisi: Lot of that. Yeah. Super interesting. And see you know yourself you’re busy, busy founder, but you were directly involved in the actual build of Agent Handler at gateway. And of course you have a fantastic CTO, you guys that you have known each other since 18 years old, but like, why were you so hands on in the process? What’s what’s part of that strategy?
43:07 – 43:35
Shensi Ding: 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 that was human build, it was much scarier for us to use AI on a product that has so many really large customers.
43:35 – 43:51
Shensi Ding: 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 0 to 1, a really, really, really transforms how we felt about AI and like what we how how we knew what was possible.
43:51 – 44:11
Shensi Ding: And because Gil and my like, co-founder and I could understand that we were able to really implement it top down. I think if the founders don’t really understand, like how I can make a big impact on the business, then it’s impossible for the team members to really evangelize it across the company without support that way. And so it took us six months to start building that product.
44:11 – 44:29
Shensi Ding: And to end, we probably did like three months of research while also coding a little like a building a prototype and presenting it. But for gateway, it took three months, like start to finish it. I just shorter because I kept getting better. I was even better than back then. And so it just like that 0 to 1 really taught us like where things are now.
44:29 – 44:43
Shensi Ding: Like 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 not just so much more quickly. We were also able to experiment with like new, new things that were coming out.
44:43 – 44:58
Shensi Ding: Like we were able to try skills and like all these different things that were like, you know, and like building like new agents and like work trees, where it would have been a little bit harder on our older code base. And so I think because we were able to learn in that way, we could encourage our team to learn, we could also teach them.
44:58 – 45:18
Shensi Ding: And so I really, really, really encourage founders who started products before, like AI got really good to build something 0 to 1 because they 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.
45:18 – 45:26
Sophie Buonassisi: That is fantastic advice and you’re staying close to the code and to AI to really stay abreast of everything. What’s next with AI?
45:26 – 45:53
Shensi Ding: We have obviously a lot more to build, so we’re continuing to build a lot more. I was really excited to launch this week that we’re going to be working a lot with different customers for. And yeah, I think what increasingly is going to happen to 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.
45:53 – 46:11
Shensi Ding: 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 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?
46:11 – 46:28
Shensi Ding: So I think people are going to start experimenting with probably training their own models off of these open source models with some data they have internally, not for everything, but for very, very specific things that will become like a part of the equation for what is the best model for each task that might be like your own homegrown model?
46:28 – 46:36
Shensi Ding: I think we’re very early, so it’s probably not this year, probably next year, but it’ll be interesting to see how everything develops, especially as things hopefully get a little bit cheaper.
46:36 – 46:42
Sophie Buonassisi: Yeah, super super fascinating. I mean, that’s like taking your data as a true next new level of unlock.
46:42 – 46:52
Shensi Ding: Yeah for sure. And this will also be a way for a lot of these like traditional stars players to have some kind of edge, especially since the market has been so aggressive towards them more recently.
46:52 – 46:56
Sophie Buonassisi: Yeah. Yeah, definitely. And where can people follow along with you and merge.
46:56 – 47:05
Shensi Ding: Yes. So you can find us at Merge Dev. You can also find me on Twitter. And you can also find me on LinkedIn as well. Shenzhen merge.
47:05 – 47:11
Sophie Buonassisi: Amazing. Well, Chansey, thank you so much for the time and conversation. This has been fantastic. Appreciate it.
47:11 – 47:12
Shensi Ding: Thanks for having me.
