Inside a YC Startup Disrupting a $900B Industry That Runs the US

0:00 Enterprise companies don't want 90% accurate. They want 99.999 as many nines as you can give them. >> I barely interact with my software anymore. If my agents can't use it, then we're going to have to switch. >> Our AI agents are digital identities that can do useful work across multiple channels. >> Almost everything you own passed through hundreds of touch points between shippers, brokers, and carriers before it ever reached you. Rate is an industry that quietly runs the world.

0:25 And like most legacy industries, it still runs on phone calls, emails, and spreadsheets. But in this trip, I met a startup that's raised venture capital on the belief that all of this is about to change. AI allows us to take on more complex, flexible problems. But you can't just throw LMS blindly and expect them to do things. You need to make sure that you're modeling the problem well. >> Love the problem, but was looking for ways that I could be contributing more immediately to the industry, to the world, to society. Today I'm going behind the scenes with Vmer, a YCback startup building agentic architecture for freight and logistics, handling everything from quoting a shipment to tracking the vehicle and collecting payment on the other side.

1:06 They've raised over $16 million with a series A led by Craft Ventures and their infrastructure is already inside some of the largest freight businesses in the US. Trucking is a $900 billion industry projected to reach 1.46 46 trillion by 2035. >> The future that we see is where you've got logistics professionals that really understand the space, that know all of the complexities, but have an army of AI workers that are operating on their behalf to help them do more. A startup like this would not have been able to exist more than 5 years ago. It just would have been impossible, but it is the industry that literally runs the world. So, there's enough room for growth. But the real question I want to understand is how a startup this young actually wins in an industry this old, one that's resisted automation for decades.

2:00 So today we're going behind the scenes with a company called Vuma. They were originally YC backed. They've already raised Series A and they are specializing in freight and logistics, specifically software and AI tooling for companies around the world. They're at a really interesting inflection point and it's a good time for us to jump in and see what's going on for them because the actual freight and logistics industry is kind of predicted to have this huge surge in the next 5 years. It's been in a bit of a lull as as a whole and great suppliers of software and AI tooling have been able to grow during those lulls, but now it's about to take another step up. And so gearing up sales teams, gearing up product and engineering teams to actually account for this growth and make sure that they can adequately capitalize on it is going to be a huge one for them. Jesse, who is one of the founders, is an Aussie, which obviously is super cool. Yeah, should be a bit of fun. Let's go do it.

3:11 dude. So, I know we've got like a lot of stuff going on today. >> Yeah. >> So, I mean, do you want to take us through kind of what's what's going on? >> Yeah. So, Monday mornings are a lot of about like kicking off the week as a team. So, we have various products that we have. So we'll have like individual product planning sessions and then we have a company all hands where we all get together little bit of celebrating the weekend and then we align on what the plan is for the week. It's very like tactical.

3:39 >> Could you give us like a quick rundown on like what Vuma actually is? >> Yeah. Vmer is a platform that allows logistics companies to build AI co-workers that do work from quote to cash. So the agents will quote business, allow logistics companies to win freight, build that shipment in their system, schedule appointments for the trucks, cover the trucks. So like finding folks to actually move it and then tracking it end to end and getting paid on the back end. So it's really a complete platform where our agents are communicating across email, voice, and text to be able to get work done. Sort of tagline is to like win a move more freight.

4:17 >> What is that? That looks familiar to me. >> I know. This is like our tribute to to Y Cominator. Like just a good reminder to stay focused on what's important. Making something that people want. There is a pretty high priority platform project for for this month on sort of general data model updates and road mapping. I I think there are there are some immediate changes that we could make to the existing entities that we have, but I'd actually if it's not urgent, I' I'd like to sequence those after having a chance to do some of the sort of longer term work because I still feel like we're we're maybe missing a bit of a concept around like coverage strategy in in general.

5:06 >> We might have just touched on this, but the auto assign carrier contracts just for my main model that does depend on the state of model change, right? Not those like the whole thing but the specific one and then but there's also other prospects I think that needed. >> No it's it's primarily for VP logistics. >> Got it. Okay. So those are two different >> but we have Yeah, we have time. I think we should build this. It will support many customers transitioning to us.

5:32 >> Yeah. >> Okay. So goal is to try to get something built on that this month but probably in the last week of the month to actually >> that's probably where it would fall. Yeah. >> Good. >> Cool. >> Cool. >> All right. Great. >> Thank you. Oh, there's no all hands in everybody. >> So on Fridays we award the chef. so the chef is whoever cooked the hardest last week. So they get to wear the apron and the chef's hat. And so then the chef from the prior week leads everyone in a final clap at the end of the start of week meeting. So you'll get to participate in that in a minute.

6:05 >> I'm looking forward to it. >> >> We do these marketing videos and we did you ever see the Volvo commercial where Jean Cord Vanam was standing between the two things? Okay. So, we did a remake of it. It is AI generated. So, it's me standing between two Boomer trucks and then they do the splits and it was a bit of it it was great. but then >> look real.

6:47 >> It looks pretty good. A lot of people I think a lot of people that don't understand how fast AI is evolving actually thought it was me doing it. I mean, it was super clearly a AI. >> Yeah. >> But then then I was like, "Oh, I should I should pretend like JeanClaude Vanam has beef with us because I like stole his trick." And then I was like, I went on a cameo and he's he's on Cameo. So we then we just we got Sean Clan Vanam to send a cameo being like Jesse, you know, like you stealing my trick like.

7:16 >> So it just came through. >> This is >> and we're going to post it later this week. It'll be fun. >> Hi Jesse. How are you? This is Jean Vanam. Apparently you tried to do my split. It didn't work so well. I heard you have a beautiful mustache, which is great. And listen, do the split. Try again. Don't hurt yourself. Make sure to warm up before. And if you don't do it, what's going to happen to you?

7:50 Yeah. Love you, man. Talk to you soon, Jesse. >> All right. Should we get our weekly plans going? >> Let's do it. >> Here. Do you want to abate to the team just how we're thinking about this like backlog user feedback? >> There are like I think 10 orgs that are currently on like the track backlog waiting list sort of waiting to go live. We want to do I don't know if user research is the right positioning but basically like a check-in call with each of them to share you know the product progress over the last few months what's live today what's coming down the pike because hypothesis like some of the users may be good to go live with what we have and we can do that sooner andor we can get feedback on what specifically is missing and what they need to help inform like the next you know pieces of the road map. The idea there is to try to like pull the learnings forward as much as we can because something that we sometimes do is we've got a product working and then we like go full force into like get the deployments live as fast as possible. And what we want to do here is like bring all the learnings as early as possible about like what are the gaps so that we can do feature builds while there are relatively fewer customers using it, relatively less backward compatibility to think about.

8:59 we'll be able to get those in place and accelerate our our go live after the fact as opposed to getting like, you know, three months in and we find out that there's a important missing feature for some customer and like having to sort of chase our tail there. >> Okay. Amazing. Who was chef last week? >> Chef. Well, and for for this one, he's needs to put the the apron back on with a with a hat for sure.

9:22 >> It's for your first sales call. >> I signed it. >> Yeah. Yeah. That was per tradition. >> Nice. >> All right. You want to count us down? >> 3 2 1. >> Okay, dude. So, for those again who don't understand who you are, what you do within the company, >> would you mind running me through a bit of an intro?

9:55 >> So, VMA, we have a a pretty heavily technical problem. you know, we're building AI agents that can do a really broad array of things in freight. >> and it's not always true that you can kind of just take some something off the shelf and have it work super reliably at the levels of autonomy that we need our products to to be able to operate. >> Yeah. >> so my background came from the world of self-driving vehicles. Worked on self-driving cars and self-driving trucks. I loved the problem. It was probably one of the best things that at that time when I was sort of coming out of grad school you could be working on as an engineer who wanted to be, you know, working on very hard problems.

10:29 >> Yeah. >> but it did have very long hardware timelines. >> It's a fastmoving industry, right? >> Oh, it absolutely is. Yeah. And just like so much room for automation. The the one of the reasons why I love the industry so much is that it sits at this layer of the economy on top of which like everything is built, right? So, it's a it's a necessary part of basically any physical product getting to its end user.

11:06 >> From your perspective, what's the biggest bottleneck for you guys right now? >> Yeah, I mean a lot of it is about how quickly can we build into this space? like we started really narrow focusing specifically on brokerage and and specifically on the order entry product in brokerage and then over the past couple years we've expanded. We do quoting, we do load coverage, scheduling, track. and now we're really seeing sort of all of Freight as like, you know, the big sort of connected ecosystem that that it is. And we think there are huge opportunities to like build across that space because that's actually the biggest challenge, how different the problems are that that every business is solving. The power that LLMs have to flexibly handle problems across domains like that is a lot more powerful. It unlocks things that were just never possible before.

11:53 >> It's really important for people to understand how complex this industry actually is and how fragmented everything is and and kind of >> would use the word chaotic in a way, right? Definitely. >> So having having like agentic architecture to actually go and solve for that you couldn't do that before, right? It's a very good description like it is it is chaotic and like every business sort of has their own way of doing things. That's really why this industry has been historically so like resistant to to automation because when you were when you had to build rigid deterministic software just the like the problems are are solvable but just the scale and the variety just didn't make it economical. and now that equation's changed a little. and it's it hasn't gone away completely. There's there's a really interesting tension between how how accurately do you try and model some canonical view of of the industry and and the problems versus being flexible enough to adapt to what an individual customer has in terms of you know this is how we think about our loads about scheduling about about coverage. so that that's that's what that tension is what really makes it interesting.

13:05 >> >> Yeah. So, number number one, we're still going to do like a basic algorithmic test cuz we feel like that's table stakes for engineering. but number two, we want to make this more realistic for how we write code of the new era, which is a lot of reviewing code.

13:39 >> which is a really great skill set to test, I think. So we set a prompt to an AI agent, tells them what code, how do we know it's good, what's broken architecturally, is it sound, is this good to go, all that stuff. Basically, we want to create a fake repo and have some PRs for it that are AI generated and broken in very specific ways. >> What does the repo do you like? What's the >> Yeah, so this was the other question.

14:02 How big do we want to scope it? It's we have two repos. It's like one mono repo but two folders within that. One is a typescript project, one is a python project. >> Oh, that's >> each one is like a tasks list like a very basic feature. >> Like one is a rest API with traditional MVC architecture and one's GraphQL type script. >> Okay. >> So you can kind of choose whichever you're more familiar with. >> Yeah. developers aren't going to be exactly familiar with our exact stack.

14:30 So it's not about testing those specific things, but just those different architectures. >> So just to jump in for the audience's context here, this is your screening for potential new hires. Is that right? >> Yeah, exactly. And we're kind of redesigning it because we're working with very AI forward engineers and the process is changing so quickly, traditional evaluation techniques don't necessarily scale or don't necessarily measure what we're trying to measure anymore. So trying to tow the line between traditional engineering skills required but also people who will be really successful in the new world I think as you called it weapons.

15:04 >> Yeah. Right. >> Yeah. Exactly. Like we if all we did was check like can somebody write code and then they come in and they start working and AI is writing 80% of the code like we're missing 80% of the signal basically. And there's a whole new there's a completely different set of skills to think about putting AI to work in sort of the most optimized or the most leaned in way. and what we're trying to solve for is making sure that our interview process is getting signal on how well candidates do that.

15:34 Yeah, it's super interesting how they're thinking about this in terms of hiring because in this new AI enabled era, the hiring process needs to be so much more thought out, very intentional and and also just very detailed in what they're actually requiring out of these potential candidates. I think the other thing that is probably most relevant here is that they have a lot of competitors. There's bigger incumbents. There's other startups at play here as we know, we've seen them on the channel, and they need to be ruthless about the selection of people that they bring into this organization if they want to be the best All of this actually works autonomously, but for the sake of the demo, we're kind of running through this like a bit manual, right?

16:36 >> Yeah, exactly. So, our tracking product will operate autonomously in the background. And you know, there's various data feeds that are going to the agent that it's continuously evaluating. So, we're getting GPS pings from the vehicle. It's assessing is it going to be running late? Is it veering off course? Has it stopped moving? Is that expected? Is it the middle of the night? and the agent may decide to do different things at various pieces. And then typically when the the load delivers, you want to confirm that that is the case because sometimes you've got to start collecting documentation from the driver because you can't pay them and you can't get paid until you have proof of that. And so our agent will initiate and complete a lot of those workflows.

17:12 So I'm going to hit create milestone here. And what that's going to do is trigger an agent that is going to give us a phone call here. So I'm actually getting a call right now. >> Wild. Hey Harold, this is Ashley with the Buma tracking team. I'm calling to confirm if you've unloaded at the delivery location for load 10003 28066 in Atlantic City, New Jersey. Let me know. Are you fully unloaded? >> Yeah, I got unloaded about an hour ago.

17:39 So, we're all set. >> Got it. Thanks. I need a couple details. What time did you arrive at the facility? >> it was 2 p.m. when I got here. >> And what time did you finish unloading and leave? >> it was like 5:00. >> Thanks. Just confirming that 5 p.m., right? >> Yeah, that's right. >> Appreciate it, Harold. Thanks for your time. >> All right. See you later, Ashley. Thank you. >> Ashley. >> Ashley. That's what we call.

18:05 >> What advice would you give to that person who's earlier in their journey who's really trying to figure this out and get some footing in this space? >> One is like don't listen to too much advice because I feel like if you're if you're going to be a founder, you've just got to figure out for yourself. I definitely early in my journey relied on the crutch of wanting people to like have the answers for me, but really like part of the journey is kind of how quickly can you figure things out and respond to feedback that you're getting from the market.

18:34 But the other thing I think is like think really big and be prepared to like shift and change your ideas really quickly early on until you hit on the thing that feels like there's something here worth committing a big part of your life to. >> And so some of the best teams and I think you know the thing that we've tried to channel is that it's like constantly moving to find the thing that is like the most powerful and impactful and then trying to be the best in the world at it.