Fireside Chat with Kady Srinivasan & Chang Chen — Solving the Last Mile: APIs, Accuracy, and the Agentic GTM Stack

1231
03:31:20.230 –> 03:31:33.159
Julia Nimchinski: And next up, a fireside chat with Katie Srinibasan, CMO of U.com. Super excited to host you, Katie, and she’s gonna be joined by Cheng Chen.

1232
03:31:33.160 –> 03:31:42.219
Julia Nimchinski: Growth Advisor and former leader at Otter.ai, HeyGen, Cartesia, and Microsoft. What a treat! Welcome to the show, how are you doing?

1233
03:31:43.210 –> 03:31:45.000
Kady Srinivasan: Good, thank you for having us.

1234
03:31:45.580 –> 03:31:46.980
Julia Nimchinski: Our pleasure.

1235
03:31:48.580 –> 03:31:54.500
Julia Nimchinski: Yeah, so the topic is gonna be solving the last mile of AI.

1236
03:31:55.370 –> 03:31:58.519
Julia Nimchinski: Super excited to dive in! Chang!

1237
03:31:59.000 –> 03:32:00.699
Chang Chen: Yeah, this is what you’re excited!

1238
03:32:01.240 –> 03:32:25.709
Chang Chen: Yeah, super excited, and I’m, it’s definitely an honor to join the city with Kelly, and I’ve been a huge fan and, super, super impressive background, how you had the three IPOs, and all the logos, like u.com, Lightspeed, Klaviyo, Dropbox, I’ve been looking forward to learn from you today.

1239
03:32:25.710 –> 03:32:35.129
Chang Chen: So, given your extensive successful experience, what one not-so-obvious advice would you offer to founders as seeking goal-to-mark success?

1240
03:32:36.270 –> 03:32:49.809
Kady Srinivasan: Yeah, it’s a great question, and really nice to chat with you, Chang. You have an impressive background, too, so we can both learn from each other in this conversation. I think the thing that I… so I advise a lot of startups.

1241
03:32:50.270 –> 03:33:03.789
Kady Srinivasan: Currently, and the one thing that I see over and over is there isn’t a lot of clarity in the beginning from a go-to-market perspective. So, when a founder… when the founder’s starting on this journey of going through their

1242
03:33:03.790 –> 03:33:10.479
Kady Srinivasan: pre-seed and seed and Series A and all that kind of stuff. The… you have to be very intentional about how you

1243
03:33:10.480 –> 03:33:14.220
Kady Srinivasan: create the right kind of go-to-market motion. And what I mean by that is.

1244
03:33:14.220 –> 03:33:38.599
Kady Srinivasan: You have to know, based on your ICP, whether you’re going to be predominantly an inbound go-to-market motion, or if it’s… you have to straight off, right off the bat, start creating an outbound motion, or you rely heavily on partners, and what that means to you. And you have to be very intentional about what it is that is going to get you the most leverage. So if, let’s say, you are a company that’s focused on

1245
03:33:39.090 –> 03:33:50.989
Kady Srinivasan: SMBs. Yes, you can go and start doing a bunch of demand gen, but also, is there an… is there an opportunity here to go and create the right kind of partnerships? And that becomes a big…

1246
03:33:50.990 –> 03:34:01.159
Kady Srinivasan: way for you to get leads. So, you have to start thinking about how does partnerships kind of become part of your go-to-market motion, not just running ads to drive inbound.

1247
03:34:01.160 –> 03:34:25.369
Kady Srinivasan: If you are an enterprise-focused company, how much of your revenues or your pipeline should come from true outbound or enterprise selling? I mean, what does that mean for hiring sellers, and what does that mean for hiring the right kind of DNA of sellers who know how to get into enterprise? What percentage of that should come from your inbound? So there’s a bunch of things that, as a go-to… as a founder, you need to really think about.

1248
03:34:25.520 –> 03:34:40.179
Kady Srinivasan: And you have to think a few steps ahead in terms of what does that landscape of go-to-market motions look like, and really leverage that. And just as a reminder, there’s inbound, which is high velocity, usually leads coming in the door.

1249
03:34:40.390 –> 03:35:03.949
Kady Srinivasan: usually smaller ACV type of deals. Then there is outbound, which is usually high touch and low volume, so you’re actually outbounding to people. And then there’s partner, right? Partner go-to-market motion. And within each of those, there are hybrid, too. There’s field sales, which is usually in the outbound bucket, and then there’s PLG, which is basically freemium to premium.

1250
03:35:04.000 –> 03:35:09.290
Kady Srinivasan: So that would be my advice, is really think very carefully about your go-to-market motion.

1251
03:35:10.050 –> 03:35:33.289
Chang Chen: Yeah, it almost sounds like, in addition for the company to find the product market fit, they also need to think about, based on their market and the… and based on the product, they also need to find the product channel fit, to find the right… to write… to find the right go-to-market motion, so that they can… they can… they can… they can properly reach out to their, potential clients.

1252
03:35:33.690 –> 03:35:40.919
Kady Srinivasan: Yeah, exactly. I like the… like what you said about product channel fit. I think it’s a product… it’s a market…

1253
03:35:41.050 –> 03:36:00.780
Kady Srinivasan: go-to-market fit, almost, that we are talking about. You won’t be able to, expect to go sell to CIOs through an inbound channel, for instance. No CIO is going to come to your website and fill out a demo request form. Similarly, you won’t be able to attract a whole bunch of SMBs by having an enterprise motion, so there’s some

1254
03:36:00.780 –> 03:36:02.429
Kady Srinivasan: Things there to be figured out.

1255
03:36:03.480 –> 03:36:14.360
Chang Chen: Gotcha, gotcha. So now that we are… we are… we are embraced with so many different AI products, AI agents, so what does it mean to be an AI native CMO nowadays?

1256
03:36:14.750 –> 03:36:39.750
Kady Srinivasan: Yeah, I think it’s a very interesting and evolving kind of a role. I don’t think any of us really know what it means, and what I say… what I mean by that is any role right now requires a lot of usage of AI tools in the right way to increase your own productivity and your team’s productivity. So that’s kind of a given, right? But on top of that, as

1257
03:36:39.750 –> 03:36:48.660
Kady Srinivasan: marketers, we also have to figure out how to speed things up by 10x to get to the market. You have to figure out, like.

1258
03:36:48.800 –> 03:36:53.620
Kady Srinivasan: if I had a thousand, 100,000 customer TAM,

1259
03:36:53.620 –> 03:37:17.720
Kady Srinivasan: And it took me… in the old days, in the SaaS world days, it took me 2 years to get to 10% of the TAM. Right now, that’s probably going to be… you get 3 months to go after that 10% of TAM, so you have to know how to get in front of those customers ASAP with a lot of urgency. So, that means that, as a CMO, you have to be really thinking about leveraging tools to really

1260
03:37:17.770 –> 03:37:32.360
Kady Srinivasan: ship things faster, campaigns, or ads, or landing pages, or website, whatever that is, that… that… and you have to have that 10x kind of speed built into everything that you’re doing. So that means, then, you have to change your

1261
03:37:32.660 –> 03:37:47.179
Kady Srinivasan: org structure, that means you have to change your processes, that means you have to change your go-to-market tech stack. So, that’s what I think it means to be an AI native CMO, is you have to really just embrace the idea that your job has to be

1262
03:37:47.270 –> 03:37:55.669
Kady Srinivasan: About moving 10x faster than you currently are, and what does that mean for you to… and how do you get to that outcome?

1263
03:37:56.770 –> 03:38:12.559
Chang Chen: Gotcha. It sounds like there’s so many things that we can leverage, can do campaigns, ads, landing pages, they can all be changed faster, but for a company that who are just thinking about getting started, do you have a suggestion on which area that they should prioritize over others?

1264
03:38:12.830 –> 03:38:34.280
Kady Srinivasan: Yeah, I would say the… you know, I think there are some very obvious things that are now very quick and easy to do, such as websites. You can spin up websites in hours rather than… or pages and websites in hours rather than days and months that it used to take us in the past. So, ensuring that you start there

1265
03:38:34.280 –> 03:38:37.279
Kady Srinivasan: Design is another one, you can…

1266
03:38:37.450 –> 03:38:59.009
Kady Srinivasan: leverage a lot of different tools, and you worked at HeyGen, you can leverage a lot of different tools to create designs at speed and scale that… landing… sorry, ads is another one. So start there, use a lot of these tools to really just get a lot of things out in the market, experiment and test it.

1267
03:38:59.220 –> 03:39:09.519
Kady Srinivasan: And that’s… I would call that, like, more of an application layer, so you have a ton of things you can do at the application layer. By the way, one personal favorite of mine right now is Gamma.

1268
03:39:09.970 –> 03:39:13.589
Kady Srinivasan: The presentation app, or,

1269
03:39:13.610 –> 03:39:27.090
Kady Srinivasan: app, agent, whatever they want to call it. It is unbelievable. I have been able to do so much with Gama that I would never have been able to do with Google Slides, and so I’d strongly recommend people look at it.

1270
03:39:27.090 –> 03:39:47.409
Kady Srinivasan: So there’s… there’s the application layer, then one level deeper is what I would call, the, the sort of the workflow engines. So, by… what I mean by that is, these are the tools that are helping send emails. These are the tools that are helping personalize messages. So, one tool that I’m really bullish about is LevinX, and…

1271
03:39:47.410 –> 03:39:49.820
Kady Srinivasan: That, basically, it’s an AI SDR

1272
03:39:49.820 –> 03:40:13.490
Kady Srinivasan: type of a tool, and so you start integrating those tools within your tech stack to say… to see how you can start to outreach faster to different micro-slivers of segments. Then there’s the true data layer, which is then, at some point, you have to start… work with your engineering team or your developer team to bring all of that data into your data.

1273
03:40:14.040 –> 03:40:36.299
Kady Srinivasan: stack, learn from it, improve, and then feed it back into that go-to-market stack. So that’s kind of how I would approach or break this problem down. Each of those layers will take some time to really absorb and get right. Now, that’s on… just on the technology piece. I also want to say there’s a whole conversation we can have about org

1274
03:40:36.410 –> 03:40:44.379
Kady Srinivasan: Organization and roles, about… and where people can start that maybe we can address, at a later time.

1275
03:40:45.590 –> 03:41:10.550
Chang Chen: That’s amazing, that’s definitely amazing to hear. Yeah, so, first of all, it’s really great to hear that you love Gamma. We’ve been working with Gamma for the last year, and, every single one of my team members really loves Gamma, and, it’s always a lot easier to talk to your partners for a product they are, they, that they are, that they actually love. So, and Gamma just… just a… just a stimulus ad.

1276
03:41:10.550 –> 03:41:15.440
Chang Chen: Gamma just released a 3.0, and we released the Gamma agent.

1277
03:41:15.570 –> 03:41:39.720
Chang Chen: Just two days ago, so if anyone is interested, then, we, would love for you to try Gamma. And back to the topic. So, it’s a, it’s a, it’s a, it’s a well summary to say that, you, you, you separated the whole agent application to applications, infrastructure, as well as the data layer. And in that sense, that there are also different.

1278
03:41:39.720 –> 03:41:46.860
Chang Chen: organization changes. How do you encourage the team to embrace the changes that when you feel some resistance?

1279
03:41:47.820 –> 03:42:06.649
Kady Srinivasan: Yeah, I think it’s, it’s a lot about the… assessing the skills versus the motivation part, right? Like, the… there are people who have the skills, and that has nothing to do with age, by the way. They have the… they want to learn, they want to be curious. For them, it’s just about exposing them to, here’s all the things that you can do, go

1280
03:42:06.650 –> 03:42:11.670
Kady Srinivasan: try Descript, go try these, tools, the… the Gamma…

1281
03:42:11.670 –> 03:42:32.339
Kady Srinivasan: thing, for instance. And then there’s the… the motivation aspect, and there are some people who get stuck in… it’s overwhelming. I don’t know where to start, I don’t know how to change, I like, like, doing things in a certain way. And so there, it’s more about training and, you know, the old world of change management, and just making sure that they’re

1282
03:42:32.340 –> 03:42:43.249
Kady Srinivasan: they understand where the… what’s the most important thing that you can do, to get this going. One very important thing that we didn’t address in the AI-native CMO

1283
03:42:44.010 –> 03:42:49.769
Kady Srinivasan: thing is, agents. So, I talked about application infrastructure and data layer.

1284
03:42:49.790 –> 03:43:04.759
Kady Srinivasan: What I should have also mentioned is then you kind of evolve to the point where you’re creating agents that do marketing-specific tasks, right? And that’s what my team is doing right now. I’ve hired a

1285
03:43:04.760 –> 03:43:21.880
Kady Srinivasan: I call them a… I call them a prompt marketer, but their job is to make agents that drive marketing use cases and workflows. So, in the org, going back to the question about the org stuff, you first get them to use tools.

1286
03:43:22.040 –> 03:43:29.790
Kady Srinivasan: Then you kind of help them train, and then you get them to the point where they’re able to create the right kind of agents that uses both

1287
03:43:29.790 –> 03:43:43.629
Kady Srinivasan: prompt engineering, and then as you get more and more sophisticated, you get towards context engineering, which is where the real holy grail of productivity is going to be. And that’s the place that I’m trying to get all of my team to.

1288
03:43:44.420 –> 03:43:56.139
Chang Chen: That’s so interesting. It’s interesting you mentioned agents, in addition to applications for infrastructure and the data layer. Do you think it’s, it’s kind of overarching,

1289
03:43:56.140 –> 03:44:07.929
Chang Chen: is a kind of overarching mechanism that connects the application and data together, or how do you think about the positioning of the AI agents within the different layers?

1290
03:44:08.210 –> 03:44:10.590
Kady Srinivasan: So, I kind of think of it as,

1291
03:44:11.200 –> 03:44:20.679
Kady Srinivasan: all of that, supercharged, is an agent. The application infrastructure and data. And what I mean by that is, let’s take a very simple use case, right, in marketing.

1292
03:44:20.720 –> 03:44:34.129
Kady Srinivasan: In the past, we had product marketers who were doing competitive analysis all the time. They were scanning the market, they were looking at what’s happening, competitive moves, what do we do in terms of positioning, etc.

1293
03:44:34.150 –> 03:44:53.450
Kady Srinivasan: What we have done now is we’ve built an agent. It’s called a competitive agent, and all it does is basically scans the market based on certain, you know, prompts and guardrails and guidelines about the kinds of information that’s happening. There’s no one person who could do that now anymore, right? It’s because there’s so much going on.

1294
03:44:53.820 –> 03:45:06.599
Kady Srinivasan: So that you take that agent, and then you start plugging it into… I mean, ideally, we haven’t done this yet, but ideally, you take that agent and you start plugging it into a product positioning agent, which then starts to refine and

1295
03:45:06.600 –> 03:45:18.660
Kady Srinivasan: create better value proposition and product. So, you can imagine this competitive agent now is… has an app layer, it has the workflow layer, it has the data layer, so it’s like a supercharged

1296
03:45:18.660 –> 03:45:32.390
Kady Srinivasan: set of things that you can do on top of what you were doing with deterministic, sort of, steps and rule-based steps. So, that’s the promise of the agent, is it literally takes your workflows.

1297
03:45:32.560 –> 03:45:33.460
Kady Srinivasan: And…

1298
03:45:33.960 –> 03:45:42.910
Kady Srinivasan: Automates it, or maybe that’s the wrong word, but takes it to a whole new level, where you’re unleashing 10x more productivity out of it.

1299
03:45:44.180 –> 03:46:06.799
Chang Chen: Gotcha, that’s really cool. One of my teammates, now that he actually gave himself a new title, his new title is AI Agent Manager, because he literally manages an AI agent to run different workflows for him, so it’s good to hear. And in terms of the enterprise adoption for agents, applications, and the AI influence, how would you say where we are?

1300
03:46:06.800 –> 03:46:11.440
Chang Chen: Bay, based on your observations, are we…

1301
03:46:11.630 –> 03:46:15.409
Chang Chen: Single-digit percentage, where we are halfway there, where.

1302
03:46:16.810 –> 03:46:33.810
Kady Srinivasan: It’s, I would say we are very early on, in the… in the whole, space. And the reason I say that is, I think at this point, we are still sort of bolting on all of these different things on top of our existing workflows, and kind of trying to…

1303
03:46:33.810 –> 03:46:44.180
Kady Srinivasan: accelerate those workflows using tools and platforms, but I think the… where it’s going to get to in maybe a few months or years’ time is

1304
03:46:44.200 –> 03:46:55.600
Kady Srinivasan: those workflows may not be relevant any longer, so maybe there are different and newer workflows that need to be, changed, you know, done differently. So, an example is,

1305
03:46:55.940 –> 03:47:01.509
Kady Srinivasan: if you look at GEO, or AEO, as people call it, right? The idea of

1306
03:47:01.790 –> 03:47:18.260
Kady Srinivasan: first of all, the old version of this was SEO. SEO was basically just creating the right kind of content with the right meta tags, with the right keywords, so that you could capture the attention of the user that’s looking for that keyword. So the whole workflow was built around

1307
03:47:18.260 –> 03:47:23.859
Kady Srinivasan: Let’s research what customers are looking for. Based on those keywords, go build the content. Based on that.

1308
03:47:23.860 –> 03:47:30.789
Kady Srinivasan: do some technical SEO on the website, and then hopefully you get crawled and indexed. But with the… with GEO,

1309
03:47:30.800 –> 03:47:51.729
Kady Srinivasan: that whole… the way you develop content is completely different. You don’t need to do all of that stuff, right? To some extent, the… to surface… to get into the 10 blue links. What you have to do instead is just generate a whole shit ton of content that is structured, that’s deep, that’s insightful, and just get it out there, almost like

1310
03:47:51.730 –> 03:47:54.430
Kady Srinivasan: Content as a factory, and…

1311
03:47:54.580 –> 03:48:06.170
Kady Srinivasan: that automatically starts to pick up the… or you automatically start to become more visible in LLMs, because you’re that much content out there. And plus.

1312
03:48:06.400 –> 03:48:19.310
Kady Srinivasan: It’s also very important to know that social media, like Reddit and LinkedIn, play a huge part in how you show up in LLMs. So now, that workflow never existed in the SEO world.

1313
03:48:19.550 –> 03:48:37.860
Kady Srinivasan: you never really automatically thought about, what do I need to do on LinkedIn to show up in Google, right? That never existed before. Now you have to think, what kind of content do I put on LinkedIn so that I show up in ChatGPT, in the LLM, right? So this whole workflow just got blown out of the water.

1314
03:48:37.860 –> 03:48:53.480
Kady Srinivasan: And that’s the direction I think we are all heading in, in the future, is… I called it, in one of my LinkedIn posts a while ago, I called it workflows to be deleted. What are the workflows to be deleted? Make sure you know that, because that’s going to accelerate you further.

1315
03:48:54.670 –> 03:49:04.729
Chang Chen: Gotcha, gotcha. When you say the workflows that need to be deleted, are you thinking about to switch more of your resources to GEO, and then deprioritize the SEO effort?

1316
03:49:05.010 –> 03:49:21.819
Kady Srinivasan: No, not necessarily deprioritize, because I think SEO is always going to be the foundation, it’s hygiene, if you will, but SEO is all about optimizing for humans that are searching. GEO is optimizing for LLMs that are trying to synthesize information. It’s a whole different ballgame. So.

1317
03:49:21.820 –> 03:49:38.889
Kady Srinivasan: The workflow to be deleted there is, don’t spend so much time trying to optimize for the 10 blue links. Make sure you’re optimizing for content production at scale, or I should say, high-quality content production at scale that lives on your website and a lot of other surfaces.

1318
03:49:40.200 –> 03:49:47.400
Chang Chen: Gotcha, gotcha, that makes a lot of sense. And in terms of AI agents, are you using any AI agents for this type of workflows?

1319
03:49:47.800 –> 03:49:50.739
Kady Srinivasan: Yeah, we, so,

1320
03:49:50.760 –> 03:50:04.769
Kady Srinivasan: my company, u.com, we have a agent builder workspace, so you can actually go and build a bunch of agents, which is what we have done. We use, we dogfood our own product, if you will. We also use,

1321
03:50:04.770 –> 03:50:19.020
Kady Srinivasan: So another, another one that we are planning to trial is, by GrowthX. this, I, I don’t know how many people know, but, Marcel, who’s the CEO of GrowthX, he has created this,

1322
03:50:19.020 –> 03:50:35.519
Kady Srinivasan: this agent plus service type of a… type of a framework for people to go and actually create a lot of content, so we are planning to think about using them. And then, there’s just a ton. Like, 11X is an agent, for instance, that we would use to do AI SDR stuff.

1323
03:50:35.700 –> 03:50:40.339
Kady Srinivasan: The trick here, though, is to make sure that when you’re using the agents.

1324
03:50:40.780 –> 03:50:53.370
Kady Srinivasan: you’re all… you’re chaining them all together, so it doesn’t become, like, you know, 15 different things that you’re using without any orchestration. So back to what you said earlier about the agent manager.

1325
03:50:53.550 –> 03:51:02.009
Kady Srinivasan: That is actually going to be a good role, because it needs to be able to orchestrate all of these agents coming together.

1326
03:51:02.970 –> 03:51:20.909
Chang Chen: Gotcha, makes sense, makes sense. And in order to chain those AI agents together, does the company require certain infrastructure, or… or… or you are all relying more on the human to be able to… to putting soul into the workflow and to… to… to provide the expertise judgment?

1327
03:51:21.320 –> 03:51:37.249
Kady Srinivasan: I mean, I think there are… if there are simple things you can do… you can use N8N, Zapier, or Zapier, sorry, N8N Zapier, or Make.com, you can do simple things with, with just using those kind of connectors, but,

1328
03:51:37.250 –> 03:51:45.629
Kady Srinivasan: my company, for instance, we do… we build RAG pipelines for customers, which basically means you are ensuring that

1329
03:51:45.630 –> 03:51:57.480
Kady Srinivasan: the right data is flowing into the right things with the right context, because context is super important to deliver the right results, right? And that’s not an easy, chaining to do, because you can’t just

1330
03:51:57.740 –> 03:52:14.880
Kady Srinivasan: plug and play something. You have to really build some of that technology and create the right kind of hygiene in the RAG pipeline, you have to do some evals and all that sort of stuff. So, it depends on your use case. If you’re… have a very thin use case, then start with something, which is, like, I love…

1331
03:52:14.880 –> 03:52:18.049
Kady Srinivasan: Zapier, it’s quite an amazing tool.

1332
03:52:18.050 –> 03:52:28.479
Kady Srinivasan: But if you have a more production-ready use case that you… where you need accuracy, you need specific types of data, then you need to come to

1333
03:52:28.560 –> 03:52:31.970
Kady Srinivasan: u.com, or, you know, companies like that.

1334
03:52:33.020 –> 03:52:41.839
Chang Chen: Gotcha. And for u.com’s platform, do you enable people to build agents, and you have infrastructure connecting together, or.

1335
03:52:41.840 –> 03:52:42.300
Kady Srinivasan: Yeah.

1336
03:52:42.300 –> 03:52:47.899
Chang Chen: You have the ready-to-use agent that the non-technical people can just plug and play.

1337
03:52:48.320 –> 03:53:13.240
Kady Srinivasan: Yeah, so this is my… now my marketing pitch. It’s, our sales pitch. We do have 100,000 agents that people can use, plug and play. They cover a wide range of things, like they do… there’s an SEO research agent, there’s a competitive agent, like I mentioned, there’s a content creator agent, I forget what… there’s a bunch of things. But we could also help you build the right kind of agents, because the…

1338
03:53:13.240 –> 03:53:16.949
Kady Srinivasan: The… what we have is a platform where you can actually create

1339
03:53:16.950 –> 03:53:33.059
Kady Srinivasan: the right agents based on the prompt, and the kind of data that you want to upload, and all that stuff. But the true magic, in my opinion, of our platform is the way we let people access all of these through APIs. So you can actually, if you’re building an agent.

1340
03:53:33.070 –> 03:53:44.719
Kady Srinivasan: and let’s say you want to get a deep research into X, Y, and Z, you don’t actually have to code that, you can, pull us in through an API, and that gives you the…

1341
03:53:44.720 –> 03:53:54.210
Kady Srinivasan: you know, the most accurate, kind of, freshest kind of results that you can embed right into your agent. So, it’s… it depends on what your,

1342
03:53:54.220 –> 03:53:57.010
Kady Srinivasan: Scopus, or what you want to do.

1343
03:53:58.400 –> 03:54:03.739
Chang Chen: Sure, gotcha. So the API enables people to do things very flexibly, and also scale with automation.

1344
03:54:03.740 –> 03:54:14.960
Kady Srinivasan: Exactly, yeah, yeah, yeah. The APIs are, I think… we help our customers really become intelligent about how to get the right data and…

1345
03:54:15.170 –> 03:54:26.969
Kady Srinivasan: access the right LLMs for the right, workflows and the right outcomes, because not every LLM is good for each workflow, and so we allow people to customize what they want and get to the right outcome.

1346
03:54:27.590 –> 03:54:37.740
Chang Chen: Gotcha, that sounds super cool. So it feels like the platform can almost connect the application layer with the company’s infrastructure and the Surface Agents to help people to automate the whole.

1347
03:54:37.740 –> 03:54:39.720
Kady Srinivasan: Absolutely. Yep.

1348
03:54:39.720 –> 03:54:50.550
Chang Chen: The biggest challenge for the team to adopt this might be the data layer, because I used to be a data analyst.

1349
03:54:51.050 –> 03:55:04.010
Chang Chen: 20 years ago, right? And then we had all saying that when you have bad data dean, then you have bad results, out. Absolutely. So, so, so, so, so now that in the, in the context that, the AI agents need

1350
03:55:04.010 –> 03:55:19.660
Chang Chen: good contacts and to do really good data, you know, in order to properly understand and to properly to, operate. What’s a suggestion to the enterprises who are about to, adopt the agent to help them to automate the workflows?

1351
03:55:20.050 –> 03:55:20.790
Kady Srinivasan: That’s okay.

1352
03:55:20.790 –> 03:55:21.680
Chang Chen: Yeah, later.

1353
03:55:21.850 –> 03:55:29.439
Kady Srinivasan: No, you’re right, absolutely right. That’s why we say the… it’s the accuracy, last mile accuracy problem. If you don’t…

1354
03:55:29.440 –> 03:55:49.509
Kady Srinivasan: have agents that are… sorry, if you don’t have the right data, then your whole accuracy is out of the picture. I think it depends so much on context, so what we do really well is we pipe the right kind of unstructured data really well. You have to know at a company what your structured data looks like and how that

1355
03:55:49.510 –> 03:55:55.560
Kady Srinivasan: That kind of marries with the unstructured data, so there’s a decent amount of setup that needs to happen.

1356
03:55:55.560 –> 03:56:04.179
Kady Srinivasan: But we… this is… what we do is we have forward-deployed engineers, we have, which I think is a new term, by the way.

1357
03:56:04.180 –> 03:56:18.670
Kady Srinivasan: and we have… or maybe it’s not as new, because Palantir introduced it. We have people who can help the company set it up in the right way, because the outcomes are always going to be only as good as the repository of information you’re accessing.

1358
03:56:19.570 –> 03:56:21.110
Chang Chen: Gotcha, gotcha.

1359
03:56:21.110 –> 03:56:41.049
Chang Chen: Cool, that’s so amazing. And for the unstructured data, I think that that may be something new comparing to the pre-Chat GPT time, because previously that we could only process the structured data. And with the unlock of the ability to understand the unstructured data, what would you think is the biggest application?

1360
03:56:41.560 –> 03:56:42.240
Kady Srinivasan: Yeah.

1361
03:56:42.240 –> 03:56:43.700
Chang Chen: You know, in the marketing world.

1362
03:56:44.170 –> 03:56:55.959
Kady Srinivasan: Yeah, no, I think it’s, definitely… in my opinion, it’s the marriage of not only the structured, unstructured data, but it’s a marriage of that with the… with web search data, right? That’s the…

1363
03:56:55.960 –> 03:57:18.799
Kady Srinivasan: that’s the beauty or the promise of all of this, is I need to be able to say, I’m running this particular campaign for this kind of a person, I want to have all of the stuff around how my salespeople are liking it, but I also want to be able to search the web and say, which other company is talking about the same kind of thing, and marry that together and

1364
03:57:18.800 –> 03:57:20.249
Kady Srinivasan: Give me that report.

1365
03:57:20.340 –> 03:57:30.530
Kady Srinivasan: That is kind of the holy grail of bringing web data and private data together in the right pipeline that helps us build some of these outcomes.

1366
03:57:31.360 –> 03:57:45.229
Chang Chen: Gotcha, gotcha. In order to marry all the data and the different, different… and the understanding of our users, and be able to ask the right question for agents to execute, it sounds like a cross-functional collaboration effort.

1367
03:57:45.230 –> 03:57:53.330
Chang Chen: And, how… how… how should CMOS rethink the streamlines across, like, product growth and marketing in the AI-native environment?

1368
03:57:53.500 –> 03:58:17.110
Kady Srinivasan: No, it’s a great question. I think I kind of always refer to this as a multi-threaded operator, right? I… again, like, I wrote a post about this called multi-threaded marketer, but it’s the same concept that applies to operators. You… I think in this new world, we have to be thinking more like GMs of certain products, rather than, like, specific swim lanes of, I’m in product or sales or marketing, because you have to think end-to-end customer journey

1369
03:58:17.110 –> 03:58:22.050
Kady Srinivasan: What does that look like? And then bring the right kind of people together to really shape the output.

1370
03:58:22.050 –> 03:58:23.590
Kady Srinivasan: So,

1371
03:58:23.590 –> 03:58:34.139
Kady Srinivasan: We can expound on this at some point, but keep in mind, this is the idea of a multi-threaded operator who can bring multiple disciplines together to really think about outcomes.

1372
03:58:36.200 –> 03:58:38.890
Chang Chen: That makes a lot of sense. And…

1373
03:58:39.170 –> 03:58:44.000
Chang Chen: And Julia, do you have a question? Or are we on time already?

1374
03:58:44.230 –> 03:58:58.830
Julia Nimchinski: Yeah, this just… the fireside chat just flew by, super endlessly fascinated, and, love being behind the wall here. Katie and Cheng, what’s the best way to support you? What are you excited about? And since this is,

1375
03:58:58.830 –> 03:59:07.000
Julia Nimchinski: Agenda Distribution Summit, what would be your advice to our Sea Excel community watching, in terms of distribution for this year?

1376
03:59:08.570 –> 03:59:12.319
Kady Srinivasan: My… my thing would be, I think…

1377
03:59:12.660 –> 03:59:24.099
Kady Srinivasan: Agents are here to stay, but not all agents are created equal. I think what, to Chang’s point, what goes into an agent is as important as which agent you use, so…

1378
03:59:24.300 –> 03:59:33.590
Kady Srinivasan: Make sure that you’re setting in the right inputs and outputs if you’re using agents. That… because that is going to determine the… what outcomes you’re going after.

1379
03:59:35.180 –> 03:59:37.120
Julia Nimchinski: Thank you so much, and how about yourself, Cheng?

1380
03:59:37.510 –> 04:00:01.389
Chang Chen: Yeah, we, so at least at our team that we are, aggressively adopting agents because we definitely see it helps us to accelerate the execution speed, and when we have a faster speed to test different, experiments, that means that we are, we are… we are accelerating our learnings as well. But we do want the, still, our expertise and our team members to be able to set up guardrails

1381
04:00:01.390 –> 04:00:07.330
Chang Chen: So that we… we will be able to, learn faster without breaking things too badly.

1382
04:00:08.810 –> 04:00:20.549
Julia Nimchinski: Thank you so much again, everyone. Please follow Katie and Ching, and we are transitioning to our next Fireside chat. Welcome to the show, Mark Stuz and Tuba Duras. Thank you again.

1383
04:00:21.300 –> 04:00:22.719
Chang Chen: Thank you, thank you.

1384
04:00:23.150 –> 04:00:24.080
Julia Nimchinski: Thanks!