Wendy Tan White leads robotics software and AI company Intrinsic, which is focused on making industrial robots easier to use and train. Intrinsic initially sat outside of Google at Alphabet, as part of its “Other Bets” group, which includes businesses such as Waymo. But in February this year, Intrinsic was brought into Google as a separate entity, working closely with DeepMind and using Google’s Gemini models and cloud services.

In September, Intrinsic announced that it would be making its basic software—needed to build and operate industrial robots—freely available, or open source. The aim is to give smaller manufacturers the ability to use robots without having to spend vast sums on custom technology. In the longer term, Google and Intrinsic hope to become the providers of the platform that powers what gets built on top of that software. Tan White sees this as the “Android for robotics and AI.”

Tan White has experience as an entrepreneur, investor and operator. She co-founded U.K. website-building platform Moonfruit, later becoming a partner at the British Growth Fund’s £2.5 billion growth capital fund, and general partner at Entrepreneur First, a global tech talent investor in AI, robotics and biotech. Tan White was then recruited as a vice president at Alphabet’s X, the moonshot factory, and in 2021 launched Intrinsic, which spun out of She spoke with TIME about how she sees robotics helping small manufacturers, Intrinsic’s relationship with Google, and what skills young people need in the AI age.

This interview has been condensed and edited for clarity. Tell me about the history of Intrinsic.

The manufacturing process hasn't really changed much for many years [since the Industrial Revolution and the advent of the Henry Ford model]. And the problem with it is, you’ve got to make it cost effective. It has to be high volume, and it has to be very similar in product. And so, from a customer's point of view, you can't always have something very specific customized to you at a price point you can afford. And also, from a production point of view, the lines are very brittle and very standardized. One of the things we were looking at is if you brought in robotics, could one let's say robot cell do more than one task?

Could you get a robot cell that's more intelligent to make a phone and a server and it's not it's not as sophisticated as this yet, but maybe a whole other type of product in one line. Could you customize it, and could you only actually build a few items? The reason that's important is you don't waste inventory; you actually build what you really need. And a factory can actually effectively be much more flexible. The other advantage to something like that might be there are certain tasks that humans do in these line production, things that actually they shouldn't do.

There are dangerous tasks. Through COVID, a lot of skilled labor workers in manufacturing retired because they don't want to do those jobs. In food production, there are some people who have to sit there deboning chicken. It's not a job humans should be doing, and there are just lots of dangerous and dirty and very repetitive jobs like that.

So, I think with robotics and AI, what that brings is the ability to have much more flexible making and production more intelligently, so actually that's really what most of my team cared about. How can you democratize access to that type of technology and tooling so that both small and big businesses can have access to it? How do you make sure humans don't have to do those sorts of jobs? But also how could you bring some of this manufacturing back more locally as well, because otherwise, what we also do is we ship a load of raw materials to China, build it there because it's cheaper, and bring it back again.

That seems like a real waste from both an economic and climate point of view, but also, it's a crazy supply chain to be relying on. So at that insight, actually, we proved that we could do cell and robotics with AI that could be more flexible and also could be a smaller unit or even in a small town or a small warehouse, but also could be used in a modular way in larger factories too. So that was the insight. We did a prototype, and then Alphabet got behind and said, "Actually, we think you're ready.”

I think the way Google saw it was, it was like the Android for robotics and AI. They've been very good with innovating Android as a platform to democratize building applications for mobile phones. So I wanted to do the same. Intrinsic was brought into Google in February. What impact has that had?

The reason we were brought back in is that robotics and AI come very front and center. So far, it's been about generative AI and agentic tools, but it's still very much in the digital world only. So it's for video, for images, for language. The next bastion is what we would call physical AI, which is: Can you bring the same ability to create intelligence around vision and touch and mobility, which we take for granted as humans. And that's what robotics really is.

What we can do with AI now is add vision, touch, and mobility to robotics, and democratizing that, making it open in a software sense, is something we had proven we could do, and in a way, given Gemini and given where Google is going, it was a very natural way for them to go. There's going to now be a platform for AI into the physical, and they thought we would become a very core part of that.

We had a lot of other partners, so we ended up doing a joint venture with Foxconn, the biggest electronics manufacturer in the world, with collaboration with Nvidia, and a lot of people started wanting to invest in us, and I think Alphabet felt this should be a core part of our next generation of software for physical AI, so they brought us back in. It's been helpful because we work very closely with DeepMind as well. It's given us more certainty to really focus on the democratization of the tooling rather than worrying about somebody wanting to invest in us.

So far so good, but it's obviously psychologically a change, I think. How does that change how you approach things like day-to-day decision making?

They've actually unusually kept us separate, so in a way, we've been brought in a bit more like DeepMind was when it was first acquired, and they've kept me CEO. They've kept our brand as Intrinsic. We've got two buildings of our own that are not in the core campus. So we have our own identity, and the reason they've done that is they know that we're still scaling and incubating.

And if you just try and squeeze us in, it's a back to innovator’s dilemma. You put us back into a standard product area within Google. It may not get the airtime it needs to breathe and to keep innovating. Now we're innovating the business itself, not just the technology. So I think that's been really helpful.

But the advantage we do have is we have much more line of sight to the senior level itself, and so they understand what we're trying to do, and we get decisions made faster, and that's been really helpful for us. The flip side is we also have to consider now that if we're representing part of the Google family as such, we can't just run. We can't just think “oh, we're just a startup, we should just take the risk and break something over here.”We actually have to consider the impacts that might have as well. What will be the initial tasks that we will see your technology being used for?

There's a lot of hype at the moment around [whether it’s] going to be humanoids being at home with you. The thing about robots and AIs is the development of their senses so that they can be adaptive and safe within the home—there's still innovation to go there. So we made a choice to do this within manufacturing because it's a semi-constrained environment, and they've already got some understanding around safety and how to organize that. So we started there.

Smaller machine workshops [that] do things like make a part of a phone, they'll use a CNC [computer numeral control] machine to do that. Typically, they haven't been able to afford things like robotics and AI, and we've already started working with a couple of different companies in the U.S., machine shops who effectively need their machines tending because it's a quite a boring job for humans to do, but also some of them have so much demand now for things like computers, the servers that produce the processing for the AI chips, and then they get put into data centers.

So increasingly, that some of the machine shops in the U.S. are being asked to build component parts of these machines. They don’t have the skilled labor anymore to do it. They don't necessarily have the space to do it. [We provide] small cells for them, so that they can actually run an extra production run overnight themselves, without that labor. So suddenly they can meet that demand.

But what's interesting is the same cell can do more than one type of task. It can tend a machine, but it can also inspect [a part] and do other tasks. So getting something that's cost effective and able to be flexible for a small machine shop like that is a game changer for them. It's a platform, so the same cell can be configured through software to do a different task, and so you don't have the cost of actually building another cell.

We've also found we can plug them together and orchestrate them, and if you build up a whole set of flexible cells like this, you have a whole factory, and that's what we're doing with Foxconn. So the same cells that are going into small machine shops are now also able to go into whole factories as well, with the same platform and the same sort of structure, and that's really where we're starting to democratize it. So, the sort of task we're talking about is machine tending, polishing.

We can look at inspecting. Some of the more innovative stuff we're looking at is things like assembly. So the thing that robots and AI haven't traditionally been able to do so far today is dexterous tasks. It's what we're good at, humans, like plugging things together, inserting things, screwing things. All those sorts of tasks are now more possible with AI, and so we're adding that in too Can you tell me more about the Nvidia partnership?

With Nvidia, they do a lot of work in their site called Omniverse, which is simulation. So we're working with them and looking at how we do more of the orchestration of the cells I was talking about in factory. They also have some what we would call specific things like how do you grasp something? The Gemini team also have some of those. We are an agnostic platform. So we're a bit like Lego. We can wrap everything into a skill.

So we can take a grasping skill from Nvidia or a grasping skill from Gemini, or we have some ourselves. So we have a world-class vision model at the moment that you can basically see with it very quickly, can look at a new part, see where it is, and pick it up. That sort of thing wasn't possible in the past. We can wrap those in skills and add them to a platform, so that people effectively can assemble skills.

Like, can you pick this up? Can you move it here? Can you plug it together? All those become like Lego bricks within our platform, and we can use it on different types of hardware. So, we're working with several partners now. Nvidia, but also our own team, Gemini, and many other startups now doing these different models. We can also wrap different pieces of hardware like that too.

So we've got a big partnership with the biggest robot manufacturer in the U.S., called FANUC [headquartered] in Japan. We've been able to add software and AI capability to their robots, which is the vision, touch ability to start to do things like insertion assembly that they've never been able to do before. So, in a way, we're a massive enabler. We wanted to make this open to many different types of people to access, and we also have a user interface layer that makes it easier for people who are more like operators to be able to use our tooling.

This was again the democratization point. If you can get a standard operator within a factory or a machine shop to really use our tooling, it's a bit like using apps on phones.

They can do things they've never been able to do before either. You give them access to that technology, so we're trying to provide that too at the moment.

Do you have any specific rules or guardrails around what your products can't or shouldn’t be used for, like defense?

We've always taken quite a strong view on where we can make the biggest difference and I think when I'm in the way, I'm mostly interested in helping the land of smaller suppliers. How can they actually have access to this?

Whether it's U.S., Germany, Japan, we work in different countries in that way. If you take Germany’s economy, most of it is driven by the Mittelstand, which is that whole supply chain machine shop group in the middle suppliers, but we're also trying to understand what is needed at the bigger factory end as well. That's why we've got a partnership with Foxconn. But we just work in the factories right now in the U.S. with that to learn how to do that.

I think we've stayed there because there's enough of a need around production. I think people forget just how hard it is to meet demand. In the U.S. 53% of skilled labor jobs are not fillable anymore, and the young are not coming into it, and the old have retired. A lot of what we do is more in that educational space. What's interesting is the next gen will come back in if they believe the tooling is more sophisticated, more like an application or it's got AI, and you get that a lot with the mom and pop shops because often they're family-owned businesses and they're finding they can't get their kids back in it [if] it looks really old-fashioned. Whereas, as soon as you bring a bit more technology in again, they'll come back in, so we're focusing more at that end of it, to be honest.

There's been a lot of conversation around how AI is going to impact the labor market, particularly at the entry level, and what skills people should be learning. How would you advise a young person?

My son's actually 21. He's an industrial designer, which, if I'm honest, is a great place to go, because he can code, he's a good engineer too, and he chose industrial design because he wanted to make, physically. And the design thinking process is really about your unique take on a problem solve or a product design. And actually, I think in a world where [AI can do] the coding for you, the design thinking process and the editing and the curation is going to become more important as the direction, to some extent.

I think that matters for all the next gen. If that tooling is available to you, your unique taste and ability to direct it is going to become more important. And do you know that process of how to make that consideration in your own mind? I think that's going to happen to every industry. I think taste-making is going to become the thing to really hone. In my daughter's case, she's gone for neuroscience. So, same thing. She could have gone blank coding, and my husband and I encouraged her to go more down a creative route, again, because she's gone down a more science creative route, which is, how do you do the breakthrough thinking?

So I would say for all the next gen, it's more about what do you love and are you great at, is what's going to become super valuable because that's all there's going to be, in some ways. It's what makes us human, and I think we're building things for other humans. That's the other thing people forget. What other humans that are interested in buying—things [or] having experiences—that resonate for them, that somebody else has designed. So that's what makes the connection. There's been discussion about women lagging behind in terms of AI adoption. What do you make of that conversation?

I've been in tech my whole career. When I did computer science, [it was] with seven women on our course at Imperial College London, out of 120—it's got better. It is important, though, that I actually think this new world should suit women. I don't want to overgeneralize but we are very multi-hybrid in our skills, and often what happens is that we put and integrate things better together. Coding, for example, or pure technology is often quite focused, and sometimes the women on my team find that either too low level or they tend to like to do broader [work], so they make great product managers.

I think we've got a media marketing problem in really putting it out there. I genuinely think women are going to be very good at using the tooling.

The way I tend to use it now is, it's almost like a collaborator. So it's somebody. I tend to solve problems by talking to people, interacting. So what I'll do now is use an agent or something to actually just talk to it. And I test it too, like another person—I know another person has a perspective, I think you're an interesting person, I know you've got this information, I don't necessarily think you know everything, and that you're right.

So what I want is a sort of debate. So, I think we can have better tools where it's more collaborative interaction. I think we'll all get more out of it, and I think then there'd be better adoption. I think the user experience side of it is going to become critical. The adoption won't be there otherwise, but I think it's coming. Is there anything that you definitely wouldn't use AI for?

I'd never use it as a definitive answer. I know there's a lot of debate about critical thinking at the moment, but, even when I talk to one of my mentors, a human being, I will take the information and still try and think about it within my own context, and is it actually relevant or valid? Same thing here. Don't think it's a definitive answer. I think you need to use your own wisdom and brain and intuition if this is OK or not.

I. I think it's the same as normal life. It's just a tool, right? So don't do anything with it that you don't feel fits with your moral compass. Just don't do it. I think these things are always going to be challenging. So, what's possible with it? I would just make sure you use it as an enabler, stay creative with it so that you don't sort of atrophy your own brain.

You've worn many hats over the course of your career. What are some of the most important lessons that you've learned about leadership?

I probably matured in that, because I was a CEO at 28 the first time round, I think I was hungrier then, and it kept me moving. But now I'm more patient, and I'm better at trying to get a full picture of information before I make a decision, but within constraint. So, one of the things about having done startups is you learn that some decisions are like 48:52 probability. You make a decision, that's your job as a leader, but I've got much better at fast trying to really get as wide a purview as I can in the time constraint I have before I make that choice.

And then when I make that choice, I make it knowing that I may not have all the information. And so I'm also prepared that if something turns out not to be right later, to adapt to that, iterate on it, and be OK if I'm wrong [and] to admit it. I think a lot of people struggle in leadership or don't know how to own it and then to iterate, and that can cause more damage over time, or it can actually slow down, ultimately. What that's about is, you need to learn to drop your inner ego, for the outcome.