Subscribe
Share
Share
Embed
Deep conversations with the founders, investors, and operators building real-world AI - robotics, automation, industrial systems & AI infrastructure. Past the headlines, into how these technologies are really built, deployed, and scaled. Hosted by Bogdan Cristei, venture partner and former systems engineer.
Bogdan Cristei (00:00)
This is the OPTIM Update. I'm Bogdan Cristei, and today I have two founders who are building robots for some of the most demanding industrial environments on the planet - construction, shipbuilding, and steel fabrication. There's a lot of noise right now about humanoid robots taking over these industries. Gary Chan and Conley Oster, co-founders of Raise Robotics, have a much more nuanced view, and they have actually deployed robots in the field to back it up.
Without further ado, here's my conversation with Gary and Conley.
Bogdan Cristei (00:32)
Welcome - so happy to have you guys here. I think I've known you for five or six years now. This is awesome.
Gary Chan (00:39)
Thank you.
Conley Oster (00:40)
Thank you. It's nice to be here.
Bogdan Cristei (00:42)
Absolutely. Let's start with some quick intros. Gary, give me a little bit about Raise Robotics in one sentence, then tell me about your background and what pulled you into this problem. Then we'll go to Conley.
Gary Chan (00:55)
In one sentence - at Raise Robotics, we build robots to help people build big things. That's probably the simplest way to put it. Before Raise, I worked on robots pretty much my whole life. I got started in elementary school when I watched a cartoon and asked my dad if he could help me build one. He bought me clay and we put something together, but I told him it doesn't move - I want it to move. I begged him enough and he finally bought me a Mindstorms kit, and that's how I got started. Since then I've worked on different types of robots - most recently at Waymo, working on perception and simulation. Now at Raise I work a lot on software, and as CTO I also work a lot on hardware. We build mobile manipulators that go out to construction sites, steel factories, and very soon shipyards, to do tasks where we think people really shouldn't be working.
Bogdan Cristei (01:55)
I've known you for six years and I did not know that story. That's why I love doing this podcast. Conley, what am I going to learn about you that I don't already know?
Conley Oster (02:04)
I guess what pulled me into this problem is just the fact that it could be done - and not just hypothetically, it could actually be done. I find that fascinating. Before this I was working in the heavy equipment industry, in the crane and rigging space. A lot of new things hit the construction industry while I was in it, and by and large, nothing worked. It wasn't until Gary and I started talking - and I was very hesitant at first, because I knew how challenging it was just to get a crane on a project, let alone sell someone on a robot and actually get it deployed. But we spent about a year together sussing each other out, making him talk to my customers, putting together simulations, and seeing what would stick. The more Gary taught me about what's possible - leveraging a lot of the lessons from AV - the more I realized there's a huge market application here. And it's not that the technology has always existed. This is something genuinely new, and it's a ripe market.
Bogdan Cristei (03:19)
Got it. As an investor, I see an enormous amount of capital flowing into robotics right now. It's finally a hot space. Most of the conversation revolves around humanoid form factor - robots that look like people, designed to do human jobs in a world built for humans. Construction and heavy industry are two of the most commonly cited targets.
What I want to explore today is whether that framing is actually right. I think you two have a more grounded view based on our conversations over the past few years. Let's start with what the work actually looks like - not the output, not the finished building or ship, but what's actually happening on the ground. A lot of people who are excited about construction automation have never spent time on a job site, and I think that gap matters.
Conley, paint a picture for someone who has never been on a heavy industrial job site. I imagine you with a hard hat, steel toe boots - you know how to talk to these guys. What does the actual work look like?
Conley Oster (04:33)
Before going into industry, I'd worked at a couple of messy factories and always assumed a construction site would be similar - some disarray, but still some resemblance of structure. When you actually get on a construction site, that's not reality at all. Everyone has a plan, but you've got to work with a lot of variability to execute it.
If you were to just put on a hard hat and walk into a commercial construction site, a shipyard, or an industrial fabrication facility, you'd see pretty much the same thing. The moment you walk in, it's immediate sensory overload. You feel like you're surrounded by chaos - big equipment moving everywhere, extremely loud noises, bright lights. If you're not focused, it's easy to get caught up. And the most surprising thing for me coming in from the outside was that everyone is super focused and super fixated on getting their work done. It's kind of just your job to stay out of the way. Don't get run over by a backhoe. Keep your head on a swivel.
It's chaotic - it's not a factory - but there is still structure and still a process. You just have to be on site to really understand what's going on.
Bogdan Cristei (05:52)
Gary, what surprised you the first time you spent real time on a job site?
Gary Chan (06:02)
For me it was honestly the weather. Working at an AV company, you're outdoors sometimes, but not really outdoors for eight to ten hours a day. That was definitely one of the biggest adjustments. Working on a robot while you're getting baked in the sun all day - we had some job sites in Dallas in the summer, 100 degrees, and the hoist was broken down, so we were walking 12 flights of stairs two or three times a day. Those days were brutal. Coming from manufacturing settings, I'd worked in workshops and done hands-on building, but it's very different doing those same activities out on a live site. It takes a toll on your body.
Bogdan Cristei (07:02)
Maybe that explains why there's a labor shortage in construction and why fewer people want to do this work. What's actually going on there, Gary?
Gary Chan (07:12)
It's an interesting dynamic. The labor shortage is definitely real, but it's not as uniform as I would have expected. It's not that you can't find people at all - like any market, if you pay people enough, you'll get good people. But that's starting to create serious problems. As prices go up, it gets more and more expensive to build.
With skilled labor specifically, what we're seeing is that a lot of people have retired, and whenever I visit customers, they tell us the same thing: "We have this really great person who's basically keeping the company running. They're 60 years old, and once they're gone, the new people - the 20 and 30 year olds - are getting trained up, but not nearly fast enough. There's going to be a gap where the old people retire, the new people still aren't up to speed, and we just don't know how we're going to keep things running."
What's striking is it's not just one customer in one industry. It's every customer we've talked to across construction, shipbuilding, and manufacturing. It's a prevailing problem everywhere right now.
Bogdan Cristei (08:23)
And we've talked previously about how the work actually breaks people physically. What are the most common causes of injury? I know there's a big push now toward ladder-free work on commercial job sites - what does that mean in practice?
Conley Oster (09:01)
The biggest hazards in heavy industry are falls from height, struck-by and crushed-by incidents, and slip, trip, and fall injuries. It's backbreaking work. The reality is that for a lot of these scopes - working on the leading edge of a building, working from height, doing overhead work - if there was a better way to do it, people would prefer not to do it. If they could oversee something else doing it, they'd adopt that solution.
The challenge right now is that we're just starting to see technology enter these environments. We're starting to see the tail end of the marketing campaigns that most automation companies have been running for the last two or three years actually hit their target audience in the field. People are starting to bend toward getting things done with fewer hazards.
Gary Chan (10:06)
One thing that surprised me, which I don't think a lot of people realize, is just how simple tasks become very hard once you're doing them every single day. Even for me just working on our robots - kneeling on the floor assembling components, bending over. If you're assembling IKEA furniture at home, it's not that bad because you do it once every few years. But once you're doing that work eight hours a day, every day, it adds up very fast.
I was talking with someone from union leadership who told me that for 10 or 20 years of his life, he just kept working and never had time to recover. You go to work, work eight hours, go home, sleep. Next day you're still sore, but you keep working. He said that even though he's now in a management position, he feels like he's still recovering years later. It's pretty sobering how it compounds.
Bogdan Cristei (11:22)
Conley, we talked on our pre-call about the amount of training required just to attend a one-hour meeting on a job site. Can you walk us through that?
Conley Oster (11:35)
This is a soft cost that almost nobody thinks about. Being in crane and rigging, I had to visit every type of project - construction, utilities, you name it. But it really opens your eyes to how much prep and administration is associated with completing work, or even just facilitating conversations about work.
At national labs and certain utility projects, just to facilitate a meeting and do some initial field engineering, I had to go through a 40-hour safety certification course - one full week - plus three days of supervised oversight on a similar project before I could even show up to talk to someone about the project. I didn't think much about that cost at the time, but now that I'm running those same trainings for our team members, I can see how significant it is.
And that's just for the meeting. If you're actually working in those environments, the PPE requirements alone are substantial - FR gear for anything utility-related, proper ventilation for abatement work. There's so much overhead associated with hazardous work that most people on the outside never see.
Bogdan Cristei (12:53)
Quick one for both of you - first image that comes to mind when you think of a task the human body was just not built for. Gary, you first.
Gary Chan (13:06)
I have one that always sticks in my mind. I was on a site in Texas, and I saw a plumber who had climbed up a pipe he'd just installed. He was standing on a little flange, tiptoeing to reach the top flange and tighten it. Like, yes, he could do it - he was doing it - but that's clearly not what you want to be doing every single day. It's very easy to fall, and he wasn't tied off. If that surface had been even slightly wet, he could have fallen and hit his head. That image always sticks with me.
Bogdan Cristei (13:45)
Conley, any stories?
Conley Oster (13:54)
Any time there's a visual scale mismatch I always find it striking. At one of my previous companies, we were doing a bridge girder installation - tiny people trying to manipulate a 300,000-pound beam into position within millimeters of accuracy. You're putting yourself in a massive hazard to get struck by something that weighs that much. It just blows your mind every time you see it.
Bogdan Cristei (14:19)
All right. So we have this picture of work that's genuinely hostile to the human body, and a labor market that pretty much reflects that. The obvious question becomes - what does the right answer look like? This is where I think you two have a take that's more nuanced than what I typically hear publicly about robotics. The dominant narrative is humanoids. I'd like to dig into whether you think that framing is incomplete for this part of the market.
Gary Chan (14:53)
From a technical perspective, I love humanoids - they're very cool robots. When we got started we actually considered whether we should use a humanoid form factor for some of our applications. The problem I kept running into was that the human form factor just wasn't big enough - surprisingly. If you need to reach around a column and you're stuck behind something, people's arms just aren't long enough, and they have to do awkward things to get around that. A lot of the applications our customers need involve lifting really heavy objects - hundreds of pounds, sometimes thousands of pounds. And we literally build lifting machines like cranes because the human body cannot lift a steel beam by hand.
Conley Oster (15:48)
The cleanest filter for deciding which applications we take on is: are we looking at applications where human physiology is the limiting factor? If human physiology is the constraint and adopting a human form factor doesn't change that constraint, then we're dealing with the exact same limitations at the end of the day. If we're doing long-reach or high overhead work, putting a humanoid in a man basket is just replacing the human with a human-looking machine. It's not solved.
Bogdan Cristei (16:23)
That reminds me of the self-driving car parallel - the clearest comparison here. We didn't solve autonomous driving by putting a robot in the driver's seat of a taxi. We made the car itself the robot. Why can't we do the same thing for industrial automation?
Gary Chan (16:38)
With self-driving, the physical robot was essentially already built. The car was the robot. What you needed to do was add sensing on top of it and make some controls adjustments for drive-by-wire. The hardware was ready - it was primarily a software challenge. But in heavy industrial automation, there's no existing physical form factor that can be a direct drop-in. The robot needs to actually be built.
A lot of people assume you can just take a humanoid and drop it exactly where a human works today. But the biggest challenge we see is that the processes themselves - when people are working on these sites - it's actually the person's physical body that leads to a lot of the problems: mistakes, accidents, that's where the issues stem from. People aren't physically capable of doing these things perfectly, repeatedly, safely. So when you think about getting a robot into these spaces, you actually do have to rethink what the physical form factor should be, build that robot, and then put the software on it to go out and work in the real world.
Bogdan Cristei (18:06)
Conley, if somebody built a five-story humanoid with full human dexterity, is that a good answer for some of your use cases - or does the problem go deeper than scale itself?
Conley Oster (18:18)
It definitely goes deeper than scale. Conceptually, does it solve the problem? For some things, yes - scale helps. We draw a parallel to a massive gantry system that spans an entire job site or fabrication facility. It solves a lot of the discrete positioning challenges because you can encapsulate the whole object, similar to what you see in automotive facilities. But it's not the most practical approach if the cost of delivering that system is exponentially higher than the value it's delivering.
The approach we've taken instead is to recreate the effect of that gantry system through distributed metrology - a localized robot with onboard sensors that establishes what you might call a software gantry, rather than a massive physical structure. You get the positioning capability without the impracticality.
Bogdan Cristei (19:37)
Raise Robotics sits in an interesting position - not truly general purpose, but not really single purpose either. How do you describe where you actually fit?
Gary Chan (19:50)
I'd actually take a less technical approach to this and look at how work is structured in heavy industries today. Even the people who do the work aren't really general purpose - that's why you have trades. People go through years, even decades, learning to be a welder or a fitter or a painter, and that's what they do every single day. Yes, humans are technically capable of many things, but when we work in commercial settings we specialize, because that allows us to do work at a much higher quality and allows processes with multiple people to be much more streamlined.
That's how we think about our robots. They're general purpose in the sense that they can handle multiple environments and multiple tasks, but we don't need them to do everything. We want to automate specific buckets where the industry has already defined enormous value - drawing a line, drilling a hole, painting a surface - across multiple customers and multiple segments. If we can automate those tasks reliably, there's a real opportunity to scale a large number of robots into the world.
Bogdan Cristei (21:06)
Single-purpose industrial robots have been around for decades - drill presses, CNC machines, spray systems. What can Raise do that those can't?
Conley Oster (21:31)
You can package up any of those static tools - the drill press, the lathe, the CNC - and take pretty much any hand tool and integrate it into a localized, mobile delivery platform. If you have quality-critical work that needs to be completed at a specific location in a large, uncontrolled environment, we can deliver that. The platform enables high-accuracy work, unbounded by a fixed installation, in any environment.
Gary Chan (22:00)
With traditional automation, you have to literally enclose the problem in a controlled space. With a CNC, that's a field volume of maybe a meter by a meter by a meter, or on the large end maybe ten meters. Above that, you have assembly lines in automotive facilities - the line itself is the encapsulation of the problem. You can have multiple robots, but you're always in a controlled space. That breaks down completely once you get to either very large components or very boutique work where you can't have that repeatability.
At that point, you have to invert the approach. You can't have a robot that encapsulates the problem - you need a robot that can work around the existing environment. That opens up an enormous range of tasks that traditional factory automation simply cannot reach.
Bogdan Cristei (23:09)
Here's what I find fascinating. When you describe what the robots actually do - drawing lines, drilling, spraying - it sounds almost simple on the surface, but you're charging a lot of money for it. Help me understand why these tasks are so much harder than they look.
Conley Oster (23:29)
People make a living doing each of these discrete things - an entire career of drawing a line. I think it's a disservice to oversimplify it, honestly, because this is people's livelihoods. Unless you've done it, you're not going to understand the inherent challenge.
Getting an entire building facade laid out to a sixteenth of an inch is extremely challenging. You can't do it with a tape measure. It requires a solid understanding of geometry, the ability to process CAD inputs and verify they're correct, knowing how to interface with your project manager - there are so many skills wrapped around that one discrete deliverable. And we see it time and time again: something as simple as drawing a line in the wrong place ends up cascading into a much bigger problem.
Gary Chan (25:01)
On the technology side, we see the exact same thing. What a robot is actually doing when it performs these tasks is acting as a bridge between the design world and the physical world. That's what tradespeople have been doing for hundreds, thousands of years - taking a design that a team has worked on and figuring out, intellectually and physically, how to make that design real. That is a very complex process. As we collect data and work with customers, we're trying to figure out how to capture all of that intelligence in the machines so that the robots can take over this translation process reliably.
Conley Oster (26:19)
And that's what makes the labor shortage so much more serious than it first appears. As skilled workers retire, you're not just losing headcount - you're losing tribal knowledge. You're losing the human LLM that's been built up over decades to deliver these projects. That's actually why we've partnered with America's largest labor union - there has to be a transition from tribal knowledge to digitized knowledge. There are deeply ingrained ways that experienced tradespeople have learned to deliver projects that a roboticist coming in fresh would never arrive at from first principles. If you're looking at painting specifically, the perfect tool path isn't just a perfectly perpendicular spray at a fixed velocity. There are elements of craft embedded in all of these deliveries, and you have to tease those out of the people who are doing the job today.
Bogdan Cristei (27:28)
Turns out operating in the real world is a lot harder than operating in the digital world.
Conley Oster (27:37)
We wish it weren't.
Gary Chan (27:38)
Unfortunately.
Bogdan Cristei (27:39)
I want to get smarter about evaluating robotics companies as an investor. Without naming names - what do investors generally get wrong? What are the biggest misconceptions you run into in fundraising conversations?
Gary Chan (28:20)
The thing I keep coming back to is what I observed in self-driving around 2015. It was very hot - billions, even hundreds of billions of dollars being poured into the space with somewhat unrealistic expectations of how fast the technology would develop. And then it all came crashing down around COVID. A lot of companies went under or got acquired at disappointing outcomes because the technology took longer than their investors expected.
It's really easy to make impressive demos, and that's a lot of what we see in robotics right now. It's not really production work - it's initial entries into the market. Getting from a compelling demo to something that reliably works in the field takes time. That said - self-driving does work now. We have Waymos everywhere. I tell everyone to try one. And I think with the new sensors, new actuators, and the AI and machine learning techniques available today, there's something genuinely new here. It just requires being patient enough to see it through.
Conley Oster (30:11)
My version: investors over-indexing on specialization. The conventional wisdom has been that if you do more than one thing, you'll spread yourself too thin and do multiple things poorly instead of one thing well. Five or ten years ago I might have agreed with that more. But we now have the means to deploy new applications on top of existing platforms much more quickly than ever before, and to train models in ways that build in much more robustness much faster.
And from the customer's perspective - if you're trying to sell a specialized tool, it's a much harder sell than a general-purpose tool, especially in heavy industrial equipment. If I can get a Bobcat instead of a wheelbarrow, I'm going to get the Bobcat. A platform that can be used across more project types with more customers is far more attractive to a distributor, and as a customer you feel like you're going to get your money's worth. Specialized tools don't get the utilization rates people think they will - I saw that firsthand in equipment rental. We offloaded a lot of specialized equipment because we couldn't get it out the way we expected.
Bogdan Cristei (31:44)
Given new training techniques, if a customer asks whether you can do something new, you can probably train for it in days rather than months. Gary, what's the state of the art there?
Gary Chan (32:09)
It's a nuanced answer. With the newer imitation learning approaches, you really can train a policy for a simple task - picking something up, placing it, even folding laundry - with just a few hundred demonstrations, maybe a week of data collection. But once you start changing the conditions - what I'd call the domain of the problem - things get much harder. Folding laundry on one specific table is very different from folding laundry across 20 different locations, let alone on the edge of a building.
Once the environment starts varying significantly, the technology gets a lot harder. Task-level generalizability is achievable with current methods. Environmental-level generalizability requires a lot more work. Making those two work together is one of the core challenges we focus on.
Bogdan Cristei (33:24)
You spend so much time on actual job sites in construction and manufacturing. What does a job site look like in 2030, 2035? Not just for Raise, but for heavy industry broadly.
Conley Oster (33:49)
Two big themes for 2030. First, much better process data - customers getting real visibility into how they're actually delivering work and at what quality. Better inspection tools, whether robotics, handheld devices, or IoT. Second, robots will have substantially offset humans from high-hazard work. You won't be seeing people on lift platforms the way you do today. You won't be seeing people spraying caustic aerosols. Anything dealing with hazard mitigation or liability reduction - that's where I see autonomy playing the biggest role in the next five years.
Gary Chan (34:50)
I'd add the technology adoption side. When we started the company, heavy industry had a well-deserved reputation for being slow to change and technology-averse. That's really shifted in the last year or two. We're seeing customers become genuinely open to new technology and taking a much more structured approach to it - actually championing new tools, building implementation plans, working with technology providers proactively.
By 2030, I think a lot of these industrial companies will have well-developed, structured processes for adopting new technology - something like how software companies run pilots and A/B tests before rolling something out company-wide. That's largely missing today. Right now it's mostly: someone likes a technology, they figure out how to implement it, and they improvise from there. Cleaning that up will make things go a lot smoother for everyone.
Conley Oster (36:39)
And we're already starting to see it change. We just engaged with one of the world's largest shipbuilders. They shared something striking: they had posted a position for a welder and after two weeks had received only two applicants. They repackaged the offering - same position, same outcome, but now you're managing a robot doing the welding rather than doing it yourself, with the title "robot welding technician." They received thousands of applicants. The labor market is already signaling where it wants to go.
Bogdan Cristei (37:36)
All right, last question for both of you. What keeps you up at night? What gets you out of bed every morning? What's the hardest unsolved problem? Gary, start us off.
Gary Chan (37:52)
It's simple for me - how do we get more robots out there? When we started, the main challenge was building the technology. Now we have much clearer roadmaps for where the technology needs to go. The bigger challenge now is education - making sure that people in the industry, investors, and engineers understand what's actually available. We get first calls with potential customers where they say, "I didn't even know this was possible." Once they see it, it completely changes their thinking.
So the biggest thing for me is making sure people understand what's possible and what the current limitations are, so we can align on realistic expectations and move forward with more deployments.
Conley Oster (39:01)
Coffee.
Genuinely though - it's curiosity about why things are the way they are. Across industries, across processes, across the gaps between technology and what actually happens in the field. We've uncovered so many processes or ways of doing things that seemed odd at first, and sometimes there was good reasoning behind them. But more often than not, you dig in and find out it's just because that's how it's always been done. Nobody has asked why.
That's a constant question I apply to myself, to what we do, and to the industry as a whole.
Bogdan Cristei (40:07)
Last one - in a sentence or two each: what does success look like for Raise Robotics in five years?
Gary Chan (40:17)
Five hundred to a thousand robots deployed in the field. And when someone is starting a new project or scoping out work, they think: "We'll use Raise Robotics for this part" - the way people think about 3D printing or CNC machining. It's just the go-to tool. It's so reliable and so common that it's the first thing that comes to mind.
Conley Oster (40:58)
Getting more robots out is the foundation. Beyond that - being at a point where we've seen a measurable drop in safety incidents in the specific scopes and applications we've deployed in. And second to that, I want Raise to be the first company in this space where customers get an insurance benefit for deploying our systems. Nobody has done that yet. Everyone is circling around it. I think we're going to crack the code, figure out the deployment hours required, and start genuinely saving people money on insurance.
Bogdan Cristei (41:31)
Where can people find out more about Raise Robotics?
Conley Oster (41:35)
Our website is raiserobotics.ai - we're updating it right now, just as a heads up. And on LinkedIn we're at Raise Robotics.
Bogdan Cristei (41:42)
Awesome. Gary, Conley - thank you so much for being here. This was a great conversation.
Gary Chan
Thank you.
Conley Oster
Thanks for having us.
*The OPTIM Update - Real-World AI | Automation, Robotics & AI Infrastructure*
*Hosted by Bogdan Cristei*
*Subscribe at optim.vc*