The best robots do the jobs humans shouldn't.
It does not create much societal value to build robots that fold laundry, but a robot that can paint the hull of a ship keeps a human away from work that is flammable and toxic.
Rishabh Aggarwal is the CTO of Raise Robotics, a San Francisco company building autonomous robots for heavy industries like construction, manufacturing, and shipbuilding. Picture a golf cart on wheels carrying two giant two-meter arms, ringed by lidars and cameras: you hand it a drawing of a job site, and it rolls from stop to stop drilling holes, marking layouts, installing brackets, and checking quality on its own. These machines take on the blue-collar work that is precise, repetitive, and often dangerous, standing behind the edge guard of a thirteen-story building so a human never has to lie on their back at the edge. For Aggarwal, the robot is not a replacement for the worker. It is a smarter, more capable version of the hammer, a tool that lets people do their jobs more accurately and safely.
He arrived here by way of a long fascination with making things autonomous. Before Raise, Aggarwal worked at an agricultural-technology startup building robots that harvested berries, a fleet that grew to roughly 150 machines deployed across Europe and California. Watching perfectly good fruit rot in the field because no one picked it in time convinced him that automation could recover real value from stubborn, physical problems. He is deeply attuned to the manufacturing side of the world and has seen up close the strain of heavy, laborious work he would never want to do himself. That firsthand exposure pushed him toward heavy industry rather than consumer gadgets, reasoning that if he could build tools for pickers in a field, he could build them for the people doing the hardest jobs on a construction site.
Aggarwal is wary of the idea that robotics is nearly solved. Take a world model, bolt on a dexterous arm from China, and you have a demo, not a product, he argues, because the moment a machine works perfectly in the lab is the moment the real journey begins. Dust on a camera degrades perception, edge cases multiply, and hardware has to run reliably from day one to earn a skeptical customer's trust. He sees the foundation layer of robotics growing Lego-like, with batteries, arms, and compute you can buy and assemble, which frees his small team to compete on the application layer. And he rejects the notion of a universal form factor: the right physical embodiment depends entirely on the task, and there is no reason a machine should be limited to six feet of human height.
What drives him now is aiming robotics at work humans should never have to do. He wants to pull people away from silica, asbestos, and flammable paints, away from the hazardous jobs that make for lousy stage demos but real societal value. A robot painting the hull of a cargo ship will never get the applause a laundry-folding machine gets, and that, to him, is exactly the point. He imagines a construction worker who no longer risks the edge of a high-rise but instead stands behind a tablet, directing the machine and checking that it digs the right hole. The expertise stays with the person, the danger goes to the robot. If technology is going to reshape labor, Aggarwal wants it to start with the jobs that break bodies, not the ones that fill demo reels.
Read full transcript of interview
In this conversation: Josh Rubin (Host, CTO Studio) and Rishabh Aggarwal (CTO, Raise Robotics).
I always start these with the same question, which is often the most difficult one for people. If you could just tell me your name.
My name is Rishabh Aggarwal. I'm the CTO at Raise Robotics.
And what does Raise Robotics do?
We make autonomous robots for heavy industries. By heavy industries, I mean construction, manufacturing, and shipbuilding. These robots go on real-world jobs and do blue-collar work like drilling, bracket installation, layout marking, quality control, and painting. So these are the tools that help blue-collar people do their job more effectively and efficiently.
These are not small robots, and they're not humanoid robots. What kind of robots are we talking about here?
You can imagine a golf-cart-sized robot on wheels, like a golf cart with two robotic arms, giant two-meter arms with tools in front of them. And they have a suite of lidars and cameras for perception and surrounding awareness. You deploy it on a construction site and give it the drawing or the map of what it needs to do, and then it continues to walk around and do the stops one by one.
How autonomous are we talking about here? Lidar, robots working on a site, that's not brand new. But you've got AI in your domain now, so what does that mean?
We start with a manual teleoperated system to get the robot the data it needs from the construction sites. Then we go to semi-autonomous, where you park it at a place and it's able to do the job effectively, motion-plans on its own, no teleoperation. And then you go to full autonomy, where you give it work to do at multiple location points: it goes to a park point, does the job, then automatically moves, scans the surroundings, sees where the obstacles are, finds a path to the next place, parks itself, and does the job again. That's what full autonomy means, which is still pretty rare in heavy industries, or even in commercial B2B work, to have that level of autonomy.
How many robots do you have in the market?
Currently we have 10 robots, and three to four are deployed simultaneously on multiple projects. We have one in Texas, one near SF, one in Nashville, South Dakota. We've completed 18 projects thus far. Full construction: the St. Jude oncology center in Nashville, a 13-story building, completely done by robots.
Let's talk through this. Are we talking about a robot that can complete multiple kinds of jobs, or one job consistently and effectively?
In this particular scope of work, we were doing one scope for the facade, which was layout and quality control. So we'd go on the construction site and lay out on the edge of the building where things need to be placed, and then the workers place those accurately. And we tell whether the construction of the building is in accordance with the CAD or not. If there's a variation in the level at which the floor is, we can relay that information to the things that get installed, so the manufacturing process can accommodate the built-versus-CAD changes.
So that first robot is almost there as a site preparer. It comes in and says, this needs to go here, this needs to go here.
On the edge of the building, so that people don't need to prepare harnesses, go on the edge of the building lying on their backs doing the job on a 12-story building. This robot has arms, so it's able to stand behind the edge guard and do those jobs.
I imagine that was a pretty labor-intensive, and really annoying, job for a human to do.
Yes, not only annoying. The challenge was humans were not precise. Taking a measuring tape across a 2,000-foot floor, your sixteenths of an inch keep getting added up. So it's more about precision, and they would do the job and the thing wouldn't work, and they'd have to take everything out and reinstall it again, and that was millions in rework. So not only is it annoying and frustrating, it's also very susceptible to errors.
So what other roles are the robots serving these days?
We've been deploying robots for drilling as well. There could be like 100,000 drilling holes that need to be done on a site, so drilling is another application. Now we're moving toward painting for shipbuilding: large cargo ships, hulls of ships, that entire surface needs to be painted. That's also very labor-intensive and very precise, because the coating thicknesses need to be very precise for ships.
It's also, I imagine, toxic for a human.
Very toxic. These are flammable paints. So not only are they toxic, humans shouldn't be in the vicinity of those paints, but currently they take ladders and spray with sprayers in hand and keep moving the ladder. Because the paints are flammable, the machines aren't allowed, so we have to make special machines that can operate in these environments.
So how long has your company been around?
Three and a half years.
Three and a half years. So this tech has obviously been using things like machine learning and telemetry for years and years. A year and a half ago, things like Claude Code start coming out, the new LLMs. Have those tools impacted or accelerated any of your work?
Yes. Over the last six, seven months I personally have been using those a lot, and over the last three, four months we've made an organization-wide push to use them. They've really changed the way we've been doing the work. The field engineers, who used to just talk to engineers, "Hey, how do I solve this issue?", if they need to change a parameter they can now just do it on their own by running Claude Code. Or if a salesperson needs to present something to a client, they'll make entire websites instead of a slide deck, with calculators and everything. So those don't require developers anymore. Where developers, UI, and UX people were needed a lot, those things are getting more and more done.
But that's still very much front end. You're talking about firmware, things that are hardware-specific. Is it impacting that world?
Yes. So that's the non-tech people using the tech, and that's definitely increased. And on the tech side now, the iteration loop the developer is using, the developer on the back end who's working on ROS nodes, or creating the motion-planning algorithms or the perception algorithms, those are also leveraging Claude. And the iteration cycle, which was supposed to be three or four months, has shrunk down to two or three weeks. Now it's a matter of how well they use this tool, because Claude, at the end, as I say, is basically the power of a thousand coders. How you use those coders is up to you.
Anything that can be automated, anything you can toss a robot into and replace, I imagine people are thinking about ways of doing that. In construction, what are other areas you're trying to get into?
That's something of a big debate in robotics. Foundation models and world models are coming into the robotics space, where people are like, take a world model, take a dexterous arm from China, and now you can do anything and solve any problem. But the problem becomes much bigger than this. You have to get the hardware running at very high reliability. You need to get the edge cases done. Your camera cannot catch dust, otherwise the perception lags. So there's a long chain of problems that come up when the thing starts working perfectly in the lab. That is when the journey of robotics starts. And that's where operators who have had experience come in and say, this problem is not so simplistic that you take a foundation model, dexterous hardware, and then boom, you have a solution.
So basically, once you've got your idea up and running in the lab, that's when it starts.
That's when it starts.
What is a problem that you personally may not have solved yet, but it's the area you really want to get these robots involved in? What's the opportunity for you?
The basic premise is: one side of the industry wants to solve for consumers, make a robot that folds laundry, does the dishes, and so on. I feel like initially all these people need to come together and solve problems for more hazardous jobs. How do we get workers away from silica, asbestos, flammable paints, from doing jobs that are hazardous in nature that humans should not be doing in the first place? And how do we create technology that helps society as a whole? It doesn't create a lot of societal value to make robots that fold laundry, but that makes for a good VC pitch and a good demo robot on stage. If you have a robot painting ships, that's not a good demo-able robot on stage, so it doesn't get as much PR, and it's harder to sell.
If you have a robot down in the mine, it's difficult to showcase that at a demo day.
Demo day, exactly.
What drew you to this industry specifically?
I've always been into automation. How do we make things autonomous? Prior to this, I was working at an ag-tech startup where we were making robots that would harvest berries, helping pickers pick berries, creating tools for them. That company had 150 robots deployed across Europe and California, picking berries. A lot of food was getting wasted, fully grown but somebody didn't pick it, so tons and tons of food was getting wasted. That was a problem I saw before. This is another opportunity I saw: why don't we work for the heavy industries where it's so laborious? I'm very attuned to the manufacturing side of things. I've seen the struggles of people doing heavy work, which I would not want to do, so I thought, why not make tools for other people as well?
Do you think the future is humanoid robots, or everything but humanoid robots?
I think the question is the physical embodiment, the form factor. It depends on the task being solved. This set of tasks should be done by this physical embodiment, and this set of tasks by that one. I don't think there is a universal embodiment.
Just because humans are shaped one way doesn't mean robots need to be shaped that one way.
If we have the power, why do we need to be limited by six feet of height? Why can't we be ten feet and drill holes up ahead?
I've always wanted to be six feet in height, but I've never quite made it. That's a whole different issue. If a robot can solve that for me, possibly with boots, I don't know. What is moving your industry forward the fastest right now?
I think the entire foundation layer is stronger than ever right now. On the hardware side, robotics is becoming more Lego-like, where you can buy batteries, robotic arms, compute systems, GPUs, and plug them all together to solve a problem. So the foundation layer is strong, which gives us a leg up in making our problem easier to solve on the application layer.
How many of those products can you actually buy in the United States versus going to China? How many do you have to order from Southeast Asia or China?
On the hardware side, they have a very, very strong pull. And to be honest, it's not a cost game, it's a quality game. It's not a factor of cost. We've had experience where, even at 4x the cost, we were not able to get similar quality or similar lead time. On the hardware side, they've been building these things since 2010. Cameras, lidars, they're already manufacturing in the tens of thousands. Americans are coming up with thousands right now, so China has a maturity, and there's a ways to go before we can get to that level.
And they used it to do their construction boom in China. Do you travel to China? Do you see construction projects there? I imagine they're very different from construction projects here.
All projects are different. They're super fast in the way they do things. Bureaucracy is far less internally within companies, and dealing with vendors in China goes faster and smoother.
But what about the actual construction projects? If you're on a building site in China, that building is thrown up faster, I imagine, than in the past.
Yeah, that's a combination of prefab structures that are put up to create the construction, and then construction robotics used in phases, doing layout marking and some other tasks, but not as extensively. They're using more prefab structures to get things up.
So that goes to: why build in the States as opposed to building in China right now?
It all comes down to the application layer. The problems, and the way development happens here, are very different, so you have to solve for your own customer. If I went about and re-architected the construction paradigm, the way construction should happen, and made a robot for that, I would get killed and there would be no adoption. So we have to understand how things get done, how the operations are done right now, and embed the robot in there, and guide that toward a more sustainable building.
Oh, that's interesting. China was more of a greenfield situation. We're a brownfield in the States. And that's actually the legacy thing that's holding us back. America is the land of tech debt, is what you're saying, in the construction industry.
And it becomes a very circular loop. The construction folks we're selling to have been burned by robotics in the past. So there's a trust deficit in that area: the robot needs to work at very high reliability from day one, but we need some understanding that there will be a development process through this. So it becomes a very circular loop.
So how big is your software development team?
Software development team would be like four to six people.
Four to six people. And for the kind of robots, for the stuff you're pushing out there, that's what you need at this stage?
Initially we were planning to double the team. But the way that Claude has come up, now we want to run the teams a lot leaner. It's easier for us to erase our tech debt, refactor the code. A lot of the grunt work the software folks were doing is now removed, and we can just think about the architectures we want, the features we want to build, the reliability we want to build, rather than "this thing is not compiling because of a version mismatch." Those grunt-work things can be offloaded a lot more easily.
If you had unlimited budget to hire people, what would you be hiring for right now, if anything?
In robotics, iteration loops are very long. So I need somebody who's not just a good solver of problems, but who has good foresight of what the problems will be in the future and solves for them today. Somebody who has seen the journey and can say, oh, these are the bottlenecks we're going to run into, so we'll address this today so we don't run into them tomorrow. People who've been in the industry for seven, eight years, have deployed 100 robots, are a step ahead of us, people with that expertise come to us and say, oh, I've seen these errors, so we won't make them, because it's not possible for everyone to make all the errors on their own.
So only deep expertise. The problem is, the only way you get deep expertise is either you spend a lot of money to hire for it, or you train it. A lot of people are concerned about: is there room for junior people right now?
It's a huge topic of conversation right now. It becomes very difficult for the junior folks who enter the market. There's a huge chunk in between a junior engineer and, say, SDE-2 to SDE-3, and all this stuff in the middle is getting solved with AI right now. So these folks need to enter the companies as interns and level up to the SDE-2 level over their graduation.
You can say that, but shouldn't you actually be taking these software engineers and throwing them onto construction sites, making them work as an electrician, make them apprentice there?
It's the cost structure that makes it exorbitant. Now you bake that cost into the operational cost of the robot, and your unit economics doesn't work. So you can do it for some folks, and then you have to amortize it across multiple units to develop the training team. So the challenge is growing and growing in that regard. I'd be keen to hear how we solve this problem from you as well.
That's a hard one. Because the kids have got to do something. We're all getting older.
Yeah. And there needs to be some path for them to get into the workforce.
Get into the workforce. But there also needs to be a pathway for new ideas. That's the downside of experience. There are plenty of people with experience who can also think outside the box and non-linearly. But often you need outside perspectives to understand where things are moving, especially when things are moving as quickly as they are right now.
Right.
People aren't as afraid of robots. Silicon Valley doesn't have time for fear or anxiety, because you're just heads down and working, but the rest of the country is a little bit freaked out right now, more about AI, more about data centers. The robots aren't freaking people out quite yet, because they haven't quite seen them in that environment. What is your anxiety like? Are you more hopeful or a little anxious about where things are going?
I think robotics still has a long way to go before it becomes widespread. If you look at the industries robotics has penetrated, there are only two: warehousing and manufacturing. As verticals, you don't see hundreds of robots deployed by anybody. Even these billion-dollar companies have like 50, 60 deployments, which is very meager. So there's a long time before robotics evolves. And secondly, I don't think of robotics as a human-labor replacement. I think these are tools that help a human do their job more efficiently, more accurately. It's easier for a construction operator to hire a junior engineer who operates a robot behind a tablet than to get a junior engineer who goes and does the layout marking on the edge himself.
It's also, you're not mad at forklifts, you're not mad at backhoes. This robot is just a more advanced version of that, at least at the moment.
Yeah, it's just a smarter machine that does a little bit more. Every action is not manually told, that's all it is. It's a helping tool for people at the end of the day. We still need the expertise of the people on the field to do the job, and these are the tools they have. It's just a more fancy tool than a hammer.
Robots are in a better position right now than pure AI, because if pure AI is replacing thought, coding, white-collar soft skills, that leads to a lot more anger than something replacing backbreaking, dangerous labor tasks. That's the point ultimately, to replace the hard, grueling, deadly things. Traditionally what we thought was, I think therefore I am. If you take my thinking away from me, what do I have left? Whereas, oh, I can't dig a hole for a living anymore, okay, I can find something else.
I can be behind the robot and the robot digs the hole. I make sure it digs the right hole, and I do the quality check after it.
Thank you very much.
Thank you so much.
Be part of the
conversation.
Whether you're a CTO who wants to be featured, a company looking to sponsor, or an engineering leader wanting a seat in the room — there's a place for you here.