HappyRobot: Automating the Work That Moves the World

1. 0:00 – 1:10

   Intro & HappyRobot’s $44M Series B

   1. DHDiana Hu0:00

      [upbeat music] So I'm excited here to welcome HappyRobot announcing their Series B for $44M, which is really impressive. You guys just raised your Series A not too long ago, less than a year. You grew 10X in revenue within the last year, less than a year, and you were in the batch not too long ago in summer '23.

   2. SPSpeaker0:28

      Yep.

   3. DHDiana Hu0:29

      So tell us what HappyRobot is.

   4. SPSpeaker0:31

      Yeah. Thank you so much, Diana. HappyRobot is building the digital workforce for real-world operations. Companies like, uh, DHL, uh, Uber Freight, uh, Flexport, they use our AI agents to automate a lot of their communications and workflows. So let's say, uh, on the context of, um, freight brokers, which is this intermediary between, uh, shipping companies, people that are shipping things, um, and carriers, companies that have assets that move the things, there's this intermediary called a freight broker, which basically receives a lot of phone calls, uh, from drivers asking for a load. So let's say I'm a driver calling in

2. 1:10 – 3:05

   What HappyRobot builds: AI agents for logistics

   1. SPSpeaker1:10

      on a load I see posted online.

   2. DHDiana Hu1:12

      Okay.

   3. SPSpeaker1:13

      An AI agent's gonna pick up as the freight broker here. HappyRobot. How can I help? Hey, what's up, man? Uh, looking at this load I saw online, wanna see if it's still available. Sure thing. Do you see a reference number on that posting? Yeah. I think it's gonna be 901347. I just made it up. I don't think it's gonna find anything. And your MC number? My MC number is 234567. And it's basically gonna find if I'm on that list. So you're a carrier. You're gonna see if I'm a registered carrier. You're with HappyRobot Transportation, right? Yep. Yeah, that's us. You found us. Thank you. Actually, you know what? I'm actually in the middle of something, so I'll call you back later, okay? All right. No problem. Feel free to call back when you're ready. Have a good one. Thank you, sir.

   4. DHDiana Hu1:53

      I love that it has a lot of the natural kinda speech and even a bit of background noise.

   5. SPSpeaker1:59

      Yeah. Yeah.

   6. DHDiana Hu1:59

      So it sounds real, because if it's too clean, there's this thing where it's like, ooh-

   7. SPSpeaker2:03

      100%

   8. DHDiana Hu2:03

      ... it's, like, too clinical and robotic.

   9. SPSpeaker2:06

      100%.

   10. DHDiana Hu2:06

       That's really cool.

   11. SPSpeaker2:07

       Yeah.

   12. DHDiana Hu2:07

       So take us to the beginning. So this sounds really cool. Now you are... You grew over 10X just within 12 months, which is incredible. I remember you, you had just done your Series A last year.

   13. SPSpeaker2:22

       Yeah.

   14. DHDiana Hu2:22

       And then you started closing these big contracts with big logos like DHL. Take us to the beginning, beginning. How did you all three start this company and met? Because you both were at your PhD programs in computer vision.

   15. SPSpeaker2:35

       You wanna tell the story how we met [laughs] ?

   16. SPSpeaker2:37

       Not entirely the PhD. I was, I was highly technical. So we met in, in Spain, in Madrid, like, 12 years ago, literally second day of university. We always tell this story, like, we sat together, and I asked for a pen and a, and, and pencil and, and paper. And then Pablo was here, so we, we got together and we started working on a million things like robotics association together on a, on a, on a few startups. And fast-forward, like, 2022, we got together again to, like, start something. We were tinkering. You were doing the PhD in Munich

3. 3:05 – 5:40

   From Spain to YC

   1. SPSpeaker3:05

      thinking about what's next for you.

   2. SPSpeaker3:07

      I was, uh, pretty much done with paper writing. As much as I was having a lot of fun during the PhD, I was learning a lot, a lot really, uh, during that time. Thanks to Javi really who actually put me on the path of, like, AI and, and w- research really when, when AI was not really a thing. I was gonna end up in, like, e- energy. Javi kind of saved me from that. [laughs]

   3. SPSpeaker3:28

      Yeah. I'm the older brother, uh, of Pablo, and, um, I did economics and law. Very boring. I don't recommend it. [laughs] During economics, I, uh, really dig, uh, very deep into econometrics, and that was very close to AI already, machine learning. And I told Pablo, like, "Hey, maybe it's late for me." It wasn't really, but [laughs] I was already in my career path, uh, to be a CFO and whatnot. I told him, like, "This is the next thing. Forget about energy." Which in hindsight, energy is gonna be [laughs] a very interesting thing. But, um, yeah.

   4. SPSpeaker3:58

      Long story short, I drop out of the PhD, tell Luis to come to Germany. We start working together on some computer vision stuff, uh, Javi supporting us from, from the US. He was already here in the US. And, uh, at some point we applied to YC, uh, get rejected on the interview the first time. We reapply. Uh, we had that one with, with you, Diana, and, uh, after a few, uh, chats, we get into YC. Uh, and that summer of sum- summer of '23, that batch of 2023, it was real transformational for us because we went into the batch with, with a little bit of revenue. We already had 70K in ARR, and every one of those peers, of o- of our peers were basically like, "Holy shit."

   5. SPSpeaker4:37

      "You guys are killing it."

   6. SPSpeaker4:38

      "You guys are killing it." We're like, "Oh, wow, like, I guess we are." [laughs] And like, at some point we realized after the batch we were not, and we hadn't really found anything, um, worth tackling. We had, like, a few customers, but nothing more. And then right on Demo Day, uh, we woke up that, that morning, obviously came from before, but that morning we wake up and we're like, "Okay, yeah, we, we're pivoting." And, like, I remember saying to you, like-

   7. DHDiana Hu5:00

      I remember that

   8. SPSpeaker5:01

      ... so that transition, uh, that took us to, to what we're building today, these, these AI agents for, uh, very messy industries and, and very compri- complex industries like supply chain and logistics, uh, we can dive deeper into it, but that was a bit of the story.

   9. DHDiana Hu5:15

      Tell us a bit about that. That's very jarring. So you went from feeling, uh, top of the world because you had some revenue-

   10. SPSpeaker5:21

       Yeah

   11. DHDiana Hu5:21

       ... because a lot of the companies we accept, over 60, 70% are pre-revenue, and you had revenue, and then you decided to give that up all away. And on Demo Day [laughs] when you're supposed to present, "We're giving up on this idea. Things aren't working." What,

4. 5:40 – 8:15

   Pivoting on Demo Day to chase a bigger vision

   1. DHDiana Hu5:40

      what about that, that previous idea? Because you, you already had some inkling during the batch. I remember doing a lot of office hours with you guys, and you, you couldn't see how you could build a large company with that one.

   2. SPSpeaker5:49

      No, YC helped us understand who our ICP was, who are we selling to. I remember ta- chatting. I was like, "Okay, who's your customer? Like, there's this, this, this. Like, can you get a working business by only focusing on one?" And we started, like, crossing them out and at some point realized, "Okay, there's, there's not a business here. Like, who is our customer?" So it was-Selective in that we decided to focus on what are we doing, and we started, like, realizing there wasn't, like, a big enough market for us. So YC was very, very helpful in understanding who, who are we serving really.

   3. SPSpeaker6:19

      And for context, maybe we were building a, a computer vision platform, like, an auto labeling platform for a computer vision... Companies that were using computer vision, robotics companies. We tried self-driving car companies. Like, we ta- tried to talk to, uh, Cruise at the time, and it was like, "No, we build that internally."

   4. SPSpeaker6:34

      Yeah.

   5. SPSpeaker6:34

      So it was a lot of build versus buy. Uh, and then we ended up going for, uh, satellite imagery. We realized that it was a lot of government.

   6. SPSpeaker6:41

      Yeah. 80% of the satellite imagery is consumed by government, so we thought, like, "This is gonna move very slow."

   7. DHDiana Hu6:48

      Which was the case.

   8. SPSpeaker6:49

      Yeah.

   9. SPSpeaker6:49

      Yeah.

   10. DHDiana Hu6:49

       I remember, um, talking with you guys, trying to close these large government contracts, and it took forever.

   11. SPSpeaker6:56

       I remember talking to this agency. They were like, "Oh, this is great. Let's talk in six months from now." We're like, "Are you kidding?" Like, [laughs] "We, we might as well be dead in six months from now, but..." Then we, we pivoted on around September of 2023. Went into a little bit of a pivot hell during those two months. We tried to tap into what we knew, what, what experience we had. Luis really wanted to dive into, into, uh, voice. I wanted to really see what LLMs were doing, and at the time, I think it was right about the time where 3.5, GPT 3.5 was coming out. And then Javi came from, uh, uh, logistics and supply chain. He was the CFO, as he said before, of, uh, the biggest olive oil distributor.

   12. SPSpeaker7:35

       Yeah, you guys put together an amazing tool, and it was a developer tool. Um, but I was like, my thesis was like, why don't we go to serve a vertical? Uh, because in the long run, this is gonna get commoditized, uh, so we build products, uh, for a vertical that we understand very deeply, and that's the one I knew.

   13. DHDiana Hu7:54

       Which was very interesting to me because I remember chatting with you guys and the went, the way you went about. You went through different verticals-

   14. SPSpeaker8:01

       Mm-hmm

   15. DHDiana Hu8:01

       ... and you attended different conferences in different industries-

   16. SPSpeaker8:04

       Mm-hmm

   17. DHDiana Hu8:04

       ... until you stumble onto the one that worked, and you guys didn't know anything about this industry. I mean, of course, now you guys are experts.

   18. SPSpeaker8:10

       Mm-hmm.

   19. DHDiana Hu8:11

       Tell us about, um-

   20. SPSpeaker8:12

       [clears throat]

   21. DHDiana Hu8:12

       ... that conference that you went that

5. 8:15 – 11:00

   Breaking into freight and supply chain

   1. DHDiana Hu8:15

      made you land into the idea that HappyRobot is today.

   2. SPSpeaker8:18

      Yeah, I mean, I, I can say I knew it as a customer of our customers. Like, because I was, uh, working in a manufacturer, a, a shipper, which is the ultimate customer of any logistics company. That was the insight, right? Um, I suffered the fines of retailers for delivering late. Uh, on time and in full, that is their main KPI, and that is where this ordeal of orchestration comes to play and where these logistics players have hundreds of peoples in call center, uh, just, just trying to find capacity to deliver on time, uh, this truckload, uh, to Walmart the next day. But as you're saying, like, we went to these conferences to explore, and this is the industry where everybody told me... I had a working demo already on the phone, like, pretty decent.

   3. SPSpeaker9:04

      The first two months we put together-

   4. SPSpeaker9:05

      Yeah

   5. SPSpeaker9:05

      ... like a d- like, Luis put together a demo. Uh-

   6. SPSpeaker9:07

      And they, yeah, and they were like-

   7. SPSpeaker9:08

      Like, seven seconds of latency, you know?

   8. SPSpeaker9:10

      Yeah. [laughs]

   9. SPSpeaker9:10

      Like, "Hello?"

   10. SPSpeaker9:11

       I wish it could-

   11. SPSpeaker9:11

       Seven seconds later, like, "Hi." [laughs]

   12. SPSpeaker9:13

       Yeah. A- and yet they told me, "Look, I have 200 people, 500 people in a call center." Not to say that their own employees in America, in, in the US, they also do that, but they have outsourced already so much and they were like, "I can give you that tomorrow if this works."

   13. DHDiana Hu9:29

       All... And it was very simple level of work. It was just people in call centers calling truckers or delivery or, or any capacity for, for logistics to check if they were available. That's it.

   14. SPSpeaker9:42

       Funnily enough, the use case we started with was actually more complex than that. We started saying, hey, Javi knows that these freight brokers, which ultimately was our first vertical, today we serve, um, freight forwarders, ocean carriers, um, trucking companies. But initially we really focused on these freight brokers, which ultimately is an intermediary between someone that is shipping something, uh, that olive oil company that Javi used to work for, and, uh, someone that has assets to move things, a trucking company, uh, a boat, a plane. So we were targeting these freight brokers with whom Javi had, like, worked with or, uh, kind of, uh, uh, dealt with. Uh, they were l- the ones, uh, may- maybe missing some of the deliveries for him and things like that, so he knew that existed. So we, we said, okay, like, there's this check call, like, this call that someone in a call center is doing to a driver to see if they're actually gonna deliver on time. We're like, okay, this is real easy, no? Like, we should definitely start with that one.

   15. SPSpeaker10:37

       Yeah.

   16. SPSpeaker10:37

       That's the easy one. "Hey, Mr. Driver, are you gonna be on time?" "Yes." "No." "Why not?" "What time?" Sounds stupid that that call happens, but the reality is that as much as, um, the industry has a lot of GPS and a lot of, uh, these, this digital tracking, if you will, um, it's not as obvious as saying, "Oh, I see, uh, Pete, the driver, in Dallas. Uh, now what? Like, why is he stopped? Like,

6. 11:00 – 14:25

   Solving the hardest problem: voice automation

   1. SPSpeaker11:00

      why is he not moving? Or, like, what is going on?" Uh, so that's why there's these phone calls. So we... To, to answer your, your, your question, Diana, like, we started with these, uh, check calls in mind, and then we go to these customers, these freight brokers. I'll tell a story about how I met our first customer, and we tell them, "Hey, let's do check calls." And they're like, "No, no, no, no, no. Let's actually do a sales negotiation. Let's actually negotiate rates on loads." So we went for, like, the harder use case, which took us a little bit of, of time to, to develop. Uh, we actually ended up, uh, fine-tuning a Llama tool-

   2. SPSpeaker11:32

      Yeah

   3. SPSpeaker11:32

      ... and Mistral.

   4. SPSpeaker11:33

      Like, the, the problem was clear, like, the pain was clear. Like, the idea market fit, if you will, if you will, was clear. The product was not clear. Like, uh, customers wanted this, but they didn't know the technology was there. So we're, like, super focused on this specific use case, which is one of the beauties of, like, being verticalized, is that we could put all our efforts in, like, training the best voice AI or building the best voice AI for this specific use case, like negotiating calls. So we're find... Like, GPT 3.5 back then was kind of fast, but it was pretty stupid. Like, you couldn't put him on a phone. Like, it wasn't reliable for, like, a business, like a real, uh, production-ready operation. GPT-4 was very slow, like-

   5. SPSpeaker12:10

      It was decent for a, for a demo, but it was like seven seconds of answering and it wasn't production ready. So we went full on fine-tuning like Llama 2 or Mistral back then, which again, speaks to how we approached building the company in that it was clearly not a long-term advantage. Like, we knew something else was coming, like GPT-5 or 4 point whatever it is. There was something coming. But doing those short-term, um, advantages kept us ahead of the curve a little bit and showing that we're a technology company and we can get, uh, this built.

   6. DHDiana Hu12:41

      I think that was part of the impressive thing we're seeing with successful vertical AI agents is doing this playbook that AI might not be 100% there, but you meet... You fill in the gaps and eventually will get there, and you have to kind of bootstrap it and fine-tune-

   7. SPSpeaker12:59

      Yes

   8. DHDiana Hu12:59

      ... a custom model. And that got you to close that first customer, right?

   9. SPSpeaker13:03

      Yeah.

   10. SPSpeaker13:03

       Yeah. When I go to one of these conferences, someone tells me about a Discord channel where the nerdiest people in logistics, uh, will have their [laughs] conversations and, and I told Pablo, still I am at the conference, "Hey, get into this channel," and you got into the channel. You know better the story.

   11. SPSpeaker13:18

       Yeah. Uh, I end up in this Discord ser- Discord server, and I pitch the thing. I do a demo call and they're like, "Holy shit."

   12. SPSpeaker13:26

       Yeah.

   13. SPSpeaker13:26

       "Like, that's super cool." And then next da- next morning, I have like an email from a couple of guys. Turns out those two guys, one was like the sales director at the 10th largest freight broker in the US, and the other one was like at the 30th largest. So like I... We end up, uh, piloting with them. I guess that speaks to the power of conferences in person, like seeing-

   14. SPSpeaker13:44

       Yeah

   15. SPSpeaker13:44

       ... like those people in person. Like a- anything vertical, there's always somewhere where like these people are meeting. Uh, and if you can like engage with those guys at, at that level, at that enterprise level as well, that was really an unlock for us versus just being at home and just like coding things without talking to people.

   16. DHDiana Hu14:00

       And one of the things is that that was just the foot in the door for you guys-

   17. SPSpeaker14:05

       Yeah

   18. DHDiana Hu14:05

       ... with the voice agent and that first workflow for negotiating contracts. Eventually, you started growing a lot more use cases and workflows, so then you weren't just one, one of the things that people might be concerned is you're just building a wrapper on top of a lot of the AI models, but that's not the case. You've built a lot of very custom

7. 14:25 – 17:20

   From small pilots to seven-figure contracts

   1. DHDiana Hu14:25

      workflows that now going from your series A to your B, you've grown to closing seven-figure contracts. That might have started at like five figures. So tell us about that progression.

   2. SPSpeaker14:36

      100%. It, it did start at that five-figure, uh, number, which we felt pretty good at at the beginning. Like, oh wow, like we're closing a, I don't know, like a 30k deal. Wow, great. It felt really good. Uh, and the reality is that was our landing. You know? Like that land and expand motion that we're seeing in a lot of, uh, uh, AI agent companies, uh, that is really what unlocked, uh, our growth, just getting our foot in the door, working with these people, them seeing us as their AI partner. They truly see us as a trusted partner with whom they can build agents and, in this case, like transform their workforce. And first use case with company X was this carrier sales negotiation. But then it's like, oh, I actually now wanna do that check call that we were talking about before, and I also wanna like collect, uh, payments, and I also wanna-

   3. SPSpeaker15:25

      Yeah

   4. SPSpeaker15:25

      ... collect documents.

   5. SPSpeaker15:26

      Voice is just one modality, but the customer really wants to get job done. Like, they don't really care if it's through phone or through email. Like, there's also email carrier sales. So we... Then we got into like there's more than voice. Like, customers don't really care what it is. They just wanna move freight, in our case. So we got into like those text workflows as well, which arguably was easier. We just, we started with like the arguably the hard part of the voice.

   6. SPSpeaker15:48

      Yeah.

   7. SPSpeaker15:49

      Putting the, the, the LLM to like answer an email was arguably the, the, the easy part. But that was something natural. Like voice, they wanted like a holistic solution, not only like a point of voice solution.

   8. SPSpeaker15:59

      Voice is the easiest one to demo and, and for people, the wow effect is, is crazy. And also the workflow is very straightforward. They have call centers just doing phone calls. For other workflows, it's more convoluted within, within the customer, and you have to like plan the approach to put an FDE, and they just get to know everything, and it's a longer term product that you build. But the voice one is very clear. You know? Like you're replacing 20 people in a call center in this country and, and that's also, to the expansion point, it's a such a clear ROI. You don't have to justify anything to the CFO. It's like, okay, that's great. And by the way, it's more robust typically, uh, the AI than, than all these-

   9. DHDiana Hu16:39

      The humans? [laughs]

   10. SPSpeaker16:40

       Yeah. I, I-

   11. SPSpeaker16:41

       It's more consistent.

   12. SPSpeaker16:42

       It's more consistent.

   13. SPSpeaker16:42

       Like actually, customers really appreciated that. Right?

   14. SPSpeaker16:44

       Yeah.

   15. SPSpeaker16:44

       Like they, they said, "Hey, my team, they don't wanna be doing this job, first of all. Second, if they do it, they're gonna do it wrong because they'll not follow the script that I gave them, and I really wanna want them to follow the script. I always wanna ask for a shipment number before I ask for the motor carrier number," whatever. Like, this is like some terminology from the industry. But like they really want the, that process to be done that way. And it's not only about the how you carry the phone call or the email or whatever workflow. It's actually about the data that you're extracting.

   16. SPSpeaker17:18

       Mm.

   17. SPSpeaker17:19

       So after that call

8. 17:20 – 20:10

   Building trust as a “digital workforce” partner

   1. SPSpeaker17:20

      is done, that email is, exchange is done, they're in, in, in front of a lot of data they're, uh, never seen, you're, you've never seen before as a customer. They've never been able to capture that data from their team because Luis gets, picks up a call-

   2. DHDiana Hu17:34

      Mm

   3. SPSpeaker17:34

      ... he negotiates a rate with a, with a driver, and maybe that doesn't end up in a s- in a sale, no? Maybe like that is just like some exchange they had there, but like they don't end up working together, broker, freight broker and trucking company. But like we should be writing down that option, that offer on some system. And reps, like the representatives in those companies, they have better things to do, uh, so they just don't do that. But that's data that they're, they've been not capturing traditionally, and that was a huge unlock as well for them.

   4. SPSpeaker18:03

      YC's next batch is now taking applications. Got a startup in you? Apply at ycombinator.com/apply. It's never too early, and filling out the app will level up your idea. Okay, back to the video

   5. DHDiana Hu18:17

      The really cool thing about you guys, you've been building a lot and really keeping at the state of the art. So back in 2023, you were fine-tuning open source models or just building off of 3.5 GPT. And now fast-forward, you're basically becoming a system of record, right, for all these logistic companies. And you started to build more complex tech around custom memory.

   6. SPSpeaker18:42

      Yeah.

   7. DHDiana Hu18:42

      And you're even doing reasoning models. Tell us about those. What are the, some of the cool things around that that's at the bleeding edge?

   8. SPSpeaker18:49

      It's fascinating. Like voice per se is, is hard. There's a lot of challenges, uh, today in like... Back then, the problem was most of the time latency was too high, but now the industry went too far, and now the bots are like interrupting too much. So there's like a sweet spot in, in real-time voice where you cannot be too slow, but you can also not be too f- too fast. So there's all this end of turn and understanding of the conversation. Like humans, when I'm talking, you're like reasoning while I'm think- like while, while I'm talking, and if I make a brief pause, that y- you understand that I'm not done. So that's something that traditionally voice, uh, companies we have not done because traditionally like the, the agent is kind of sleeping while, while the, the person is talking, the agent is kind of sleeping. And then once the person stop talking, it starts thinking like super fast and trying to answer as quick as possible. So there's a lot of reasoning mid-call that the bot is doing using different models depending on the difficulty of the conversation. The memory part, like we're automating work across a lot of thousands of agents at the same time, so there's a lot of knowledge that they can share live. And there's an example, for example, when negotiating calls, like there's carriers calling in the broker, and there's 10 concurrent calls for the same load. Typically, humans would not be able to like say, "Hey, I have a very, very interesting carrier. Maybe drop the price, negotiate harder. This lo- this load is very hot." The agents can sync

9. 20:10 – 23:30

   Custom tech: memory, reasoning, and real-time voice

   1. SPSpeaker20:10

      all the data real time and kind of say, "Hey, I'm getting a lot of interest. Do you wanna book it for less? I mean, th- this is the best I can do. I have another offer right now." All that sharing real time of information or that memory that we, we create was not possible before. So it's a bit of that data unlock that we provide the customer, not only automating the work per se, but also providing visibility into what's going on.

   2. DHDiana Hu20:30

      So what are some of the interesting tech that you have invented that, um, maybe that you can share-

   3. SPSpeaker20:35

      Yeah

   4. DHDiana Hu20:35

      ... that is really at the bleeding edge, that not even the AI labs are doing? And you have only been able to do so because you've gone deep into this vertical use case, and this is one of the reasons why you're the best solution.

   5. SPSpeaker20:48

      No, thank you. Um, I think end of turn is something that is not widely spoken. Like we always think about voice in like transcriber, LLM, um, text generation, but we kind of knew that models were gonna get faster. Like it was a matter of time that the, the models were gonna be fast. So that understanding of the conversation, um, pace and when to interrupt, when not to in, especially in voice over phone, it's also hard to know when the other person is engaged. So humans typically like back channel or you say something. And I've tried other systems that whenever you say something, the bot stops because it thinks you wanna talk. So that understanding of the conversation, that interruption handling is extremely important. And when you get all the other pieces right, this is typically what fails. Like you can have a very natural conversation, but if it's con- constantly like stopping whenever it listens to like a background noise or someone is talking and, I don't know, there's a side conversation, and the bot is answering someone else, that's something that is hard per se because it's, it's, it's a hard problem that you need a know- a lot of like audio understanding. So that interruption handling is something we've, we've worked a lot, and I think we're, we're good at it. I th- I think something that we unlock is that we're now automating work for these companies. Like we're automating the, the phone calls, the emails, the text, like literally enhancing or replacing, if you will, those human labor that are doing repetitive tasks. This is unlocking a lot of data. We were talking like those, that memory piece, those conversations are now creating this system of record. You said like now there's external data sources, and there's data that the agents are generating. What we're thinking is putting what we call frontal, which is like a bigger super intelligence on top who is not doing the phone calls per se. It's literally just looking at the data and being proactive in like telling the u- the human insights, teaching the workflows how to behave better. So now you have this manager that's looking at every single phone call, and it's t- teaching agent, "Hey, you should maybe negotiate harder," or, "You should call, uh, this carrier because they don't wanna be text." Like you're now holistically looking at the thing. So there's this like reasoning agent twenty-four seven looking at the data and providing insights and teaching the rest of the system.

   6. SPSpeaker23:01

      I, I think another way to see it is we've built the agentic workflows. That's the foundation, and it pretty much looks like a bunch of notes, like no-code tool, like a Zapier-looking tool just connecting notes to each other, having like a, a prompt and a set of tools. That's what we've called agentic workflows. We have a layer on top, which is the AI worker. That's the entity that is now looking at the workflows that it has available that it can use. For example, we have this one customer in the manufacturing

10. 23:30 – 26:10

    Moving beyond logistics into global operations

    1. SPSpeaker23:30

       space. Their team of 300, um, buyers or procurement managers are just looking every day at an ERP, seeing like what order is gonna be late, and they have to reason about, "Hmm, I think this order is gonna be late. I should maybe send an email to the supplier to make sure that that order is gonna arrive on time." The next morning they wake up and see that no one has replied. "Okay, I, I now have to make a phone call." So we can... The way we think about it is that email workflow, that, that is, that is a workflow. That email communication is a workflow. That, a phone call is another workflow. There's an entity on top, that AI worker, that is reasoning about when to use one workflow or the other. It's almost like-It has capabilities. That AI worker has capabilities

    2. SPSpeaker24:16

       Planning and orchestrating everything. It's not doing the phone calls, it's just, like, understanding the problem and scheduling different parts to, like, do work in the outside world, and then getting the results and briefing about them.

    3. DHDiana Hu24:25

       Interesting. So you have multiple agents, and you have the manager agent.

    4. SPSpeaker24:28

       Exactly.

    5. SPSpeaker24:29

       Yes.

    6. DHDiana Hu24:29

       That's, that's cool. So I think a lot of what you're building is really cool, making a lot of these logistic companies more efficient, not needing as large of a operation with call centers. On the flip side, there might be a lot of, um, sentiment from the public that a lot of jobs are getting taken away. How do you tend to think about that?

    7. SPSpeaker24:49

       This is a little bit of that, um, AI will not replace you, but, like, a, a human using AI will, and this is really true. We're seeing in our customers how their representatives are actually more efficient. Uh, some of the representatives in these freight brokers, which is one of those verticals that we tackle, are booking more freight, more loads, uh, 25% to 30% more loads per day, uh, which actually gives them more money because they get commission on those loads. So now they see it as a, as a, as a tool, right? As a, as an opportunity to grow more than a threat, really.

    8. DHDiana Hu25:24

       Which I think is what we're seeing with a lot of AI startups-

    9. SPSpeaker25:27

       Yeah

    10. DHDiana Hu25:27

        ... is this case where it's not that it's, uh, replacing humans, it is freeing up humans to do higher level task. And, like, this company, they didn't necessarily let anyone go at all. On the contrary, all those former, uh, employees are doing even better tasks and even more fun tasks-

    11. SPSpeaker25:43

        Yeah

    12. DHDiana Hu25:43

        ... and the company's becoming more efficient.

    13. SPSpeaker25:45

        100%.

    14. DHDiana Hu25:46

        That's awesome. What are you excited about the future? Where, where does HappyRobot go from now?

    15. SPSpeaker25:50

        We started really narrow with a vertical of, uh, freight brokers. After that we said, "Hey, we're actually not, like, automating the work of these freight brokers only. We're actually automating the work in a industry that is much bigger." We're automating all of the non-physical labor, uh, that kind of keeps

11. 26:10 – 27:52

    Automating the invisible work of the world

    1. SPSpeaker26:10

       goods moving around the world, people moving around the world, services. We're now working with some of the largest, uh, uh, energy suppliers in, in the, in the world. So there's someone setting up an appointment, there's someone, like, um, making sure that that appointment is, is set at the right time. There's operations. Uh, we were at the keynote in Zamzarra, uh, the lar- one of the largest, uh, fleet management software companies out there, and they serve these industry of, of physical operations. The way we, we see our agents is they're automating the work that is needed to keep physical operations going. So that is really what is exciting us a lot. Now, like, we're actually automating the work that keeps a lot of the, a lot of the work in the world, uh, moving.

    2. DHDiana Hu26:51

       And you guys are hiring.

    3. SPSpeaker26:53

       We're hiring a lot, yeah.

    4. DHDiana Hu26:54

       Yeah.

    5. SPSpeaker26:54

       A lot of engineers, full stack, ML.

    6. SPSpeaker26:56

       Lots of... Yeah, yeah. Like, in every one of those steps, like the agent part, how do we get the bot to do better at the phone calls? At the memory piece, that worker that needs to, like, plan and schedule and orchestrate the thing. The higher level intelligence of... L- there's a lot of in the tech stack that we're, we're, w- like, we're building ourself. Like, we're an AI company. We have an AI roadmap, and that's something our customers appreciate, that they wanna partner with someone that's really pushing the frontier of, of what's possible. So we're hiring, um, a lot of engineers as well.

    7. SPSpeaker27:23

       Forward deployed engineers as well. We haven't talked about that, but we're hiring a lot of, uh, of that concept that Palantir brought, like that forward deployed engineer, that customer engineer almost. That, uh, uh, person that is technically savvy, but also, like, enjoys the work with customers. Hir- hiring a lot on that front as well.

    8. DHDiana Hu27:39

       All right. Well, congratulations guys, on your Series B.

    9. SPSpeaker27:43

       Thank you so much.

    10. SPSpeaker27:43

        Thank you very much. [outro music]

Episode duration: 27:53