Why is AI adoption now a defensive imperative for manufacturers?
If your company has not set up generative AI properly, your employees are already using it without you. Jens Mobius, managing partner at Infinity Partners, makes the case on The Artificial Intelligence Podcast that Gen AI has moved from a nice-to-have to a defensive imperative: 82% of knowledge workers use AI, according to a study commissioned by Google and carried out by Harris Poll, so an unmanaged rollout is shadow IT.
Mobius says the window changed. Ten months ago he told manufacturing clients it was the perfect time to adopt, because there was a year and a half of best practice to look back on without having to experiment with everything. Now the argument is defensive. "Gen AI is an absolute imperative to not miss the boat," he says. It is not too late to leapfrog with agents and robotic process automation.
What is the spaceship dilemma in AI adoption?
It is the choice between leaving now and waiting for better technology. Mobius uses Proxima Centauri as the illustration: with current tech the trip takes 6,300 years, so do you go now or wait for something faster? The same applies to AI tools. You have to decide on a tool or some form of adoption at some point, or you miss the boat. The opportunity he points to is catching the humanoid robot wave.
What do humanoid robots cost compared to manual labor?
A humanoid robot costs about $10 per hour to operate today, and that falls toward $1 per hour by 2035. Mobius defines the cost as purchase price, capital costs and maintenance divided by the hours you can operate it. Manual labor runs at $18 per hour. So humanoids enter at 50% of the manual labor rate and go to 5% across one decade. He compares that with the steam engine, which came in at a 20% performance improvement, and says humanoids will be a factor of 2.5 to 25. Generative AI alone delivers at least 25% productivity improvements across studies, he says.
Which humanoid robots are already on factory floors?
Bipedal, Gen AI powered, general-purpose bots are the ones Mobius means. Figure 02 by Figure works at a BMW plant in Spartanburg, where it can put fitted sheets on precise fittings. Apptronik's Apollo works at a Mercedes-Benz plant, mostly material handling. Atlas by Boston Dynamics is autonomous and general purpose. Tesla's Optimus is the one he calls far ahead and tailored to manufacturing. They can also do visual inspection while carrying things, quality control and assembly.
Why are general-purpose humanoid robots different from traditional factory equipment?
Because they are versatile, so you do not get the sunk-cost problem. Manufacturers are used to buying equipment as capex that is good for one thing. If your strategy changes and you no longer need it, that money is gone. With humanoids you can repurpose them to do something else, and they lose much less value. Even if you dissolve the venture, Mobius says, they will be just as valuable or more valuable to someone else.
When does it make sense to put humanoid robots to work?
The perfect storm is adverse or dangerous conditions, where putting a human to work carries a lot of risk, or a labor shortage. Mobius says that is the no-brainer case, beyond simply getting into pole position for when industrialization takes off.
How do you pick an AI tool without wasting money?
You flip the sequence. Traditional IT projects were process first, then the tool that fits the process. With Gen AI, Mobius says it is the tool first, then you experiment, and the process comes out of that. Everyone should get training on the basics and start using it, safety considerations aside. The experiments still need to be aligned with the company's priorities, not wild. Mobius acknowledges the jagged frontier: you do not know what the tool is good at until you try.
The concern that you buy a tool and then have to justify it is real. The answer is discipline and change management, not a longer selection process.
What to do
Commit to a tool and start experimenting now. Waiting for perfect technology is the spaceship dilemma, and Mobius says you have to decide at some point or miss the boat.
Train everyone on the basics and get them using Gen AI, with safety considerations aside.
Align the experiments to what the company is trying to achieve. Mobius calls for defined priorities rather than wild experimenting.
Start with six to eight clear initiatives, each spearheaded by an organizational leader such as a plant manager or C-suite member, not a project manager.
Hold program group meetings every two weeks. Mobius cites Stanley McChrystal's Team of Teams and the concept of shared consciousness: everyone leaves understanding not only their own project but everyone else's, which spreads best practice and creates accountability.
Use one-on-one coaching with project leaders between meetings. Bring in an expert or two to suggest the right tools for the problem at hand, which keeps expensive experimenting to a minimum.
What is the risk of waiting?
You end up playing catch-up. Mobius says it is much more fun to surf the wave when you are ahead of the curve. "If you decide to catch the wave, you better paddle out there," he says. You might not catch the first wave and you might sit around for a bit, but if you are not out there positioning yourself, you will not catch it.
Original show notes
Can AI and humanoid robots actually transform manufacturing? Jonathan Green sits down with Jens Mobius (Managing Partner, Infinity Partners) on why Gen AI has gone from a nice-to-have to a defensive imperative for mid-market manufacturers, the "spaceship dilemma" of when to adopt, the staggering economics of humanoid robots ($10/hour today toward $1/hour by 2035 vs. $18/hour manual labor), the real bots already on factory floors (Figure 02 at BMW, Apptronik's Apollo at Mercedes, Tesla's Optimus), and the change-management discipline that makes adoption actually stick.
Key Takeaways: • Gen AI is now a defensive imperative — 82% of knowledge workers already use it, so if your company hasn't set it up, it's running as shadow IT. You can still leapfrog with agents and RPA. • The "spaceship dilemma": if you wait for perfect tech you never leave the dock — you have to commit to a tool and start experimenting now, or you miss the humanoid-robot wave. • Humanoid economics are staggering — ~$10/hour today, dropping toward $1/hour by 2035, against $18/hour manual labor. General-purpose bots are versatile, so they hold value even when your strategy changes. • With Gen AI the sequence flips: tool-first, experiment, THEN build the process — the opposite of traditional IT. Just keep the experiments aligned to company priorities, not wild. • Adoption is change management: 6–8 leader-owned initiatives, biweekly "shared consciousness" meetings (McChrystal's Team of Teams), and 1:1 coaching spread best practice and keep costly experimenting to a minimum.
Notable Quotes: "Gen AI is an absolute imperative to not miss the boat." — Jens Mobius "If you decide to catch the wave, you better paddle out there." — Jens Mobius
Connect with Jens Mobius: LinkedIn: https://www.linkedin.com/in/jmobius/ Website: https://infinity-partners.co
Enjoyed this? Follow The Artificial Intelligence Podcast and share it with someone in manufacturing.
Auto-generated transcript, 3,550 words. Timestamps link to the moment in the episode.
Can AI make a difference in manufacturing? Let's find out today's amazing special guest, Jens Mobius. Thank you so much for being here for today's episode. think mostly when people think about AI right now, they just think about content creation, blogging, social media, video creation, the things that get the most attention on social media and go trending, but it's not really where the needles move. So I'm really fascinated by... how AI can make a difference in manufacturing. So let's just start with kind of what you
do in that field and let's just dive right in. Yeah, thanks for having me, Jonathan. So we serve the transatlantic mid-market deep tech manufacturing sector. So we split that into two, the smaller ones, let's say up to 20 million revenue. We help them with market entry, especially in the U.S., know, business development and growth topics, including expanding their footprint. Because many are pretty soon also bring in manufacturing here to the US because customers are not paying tariffs. And then the larger companies,
that's called the 200 million to $2 billion range, we do business transformation and step change performance improvement. And everything in the middle, that's kind like the gray area where it could be either, neither or both. And, you know, for our manufacturing clients 10 months ago, I would say, you know, this is the perfect time to adopt Gen. AI, because you have one and a half years of best practice to look back on, you don't have to experiment with everything,
but you're still kind of ahead of the curve. And now that has changed to a defensive argument. So Gen. AI is an absolute imperative to not miss the boat, right? So according to a the latest study that I saw that was commissioned by Google and carried out by Harris Poll, 82 % of knowledge workers use AI. And if your company doesn't have that set up properly, that is shadow IT basically. So most at the moment use it as shadow IT, but it's not too late. there's...
opportunity to leapfrog with agents and RPA, so robotic process automation. And everyone faces this spaceship dilemma a little bit, Where with AI implementation or adoption, so are you familiar with the spaceship dilemma? Why do you start by explaining that to us? Because not everyone listening will know it. Yeah, so basically, if you want to go to the nearest star, which would be Proxima Centauri, with current tech, it would take you 6,300 years to get there. And so the question is, do you leave now
or do you wait for better tech to be available to make it maybe in half the time or really cut it down to a fraction? And you have a similar dilemma with... adoption of AI tools. But now you have an opportunity to... You need to decide on a tool or on some adoption at some point in order to not miss the boat, so to say. And the big opportunity now is to catch the humanoid robot wave. So before we hit the...
Sputnik or COVID or, you know, GPT moment there, however you want to call it, where everybody suddenly becomes clearly aware, you know, that, not that it's a thing, but to the need to act immediately. So realizing you're now in a different world. And so I know I'm preaching to the choir here, but speaking to manufacturing people, because they historically haven't used it so much or are slow to adopt. So they need to prioritize adoption and what's striking is just the order of magnitude of performance improvements,
right or this productivity gains because across studies just geni per se you get at least 25 % of productivity improvements and You know with humanoids that's gonna be even an order of magnitude more so they enter the market at a costs capability, so that's the cost to operate a humanoid per hour. So the cost across purchasing it, capital costs, and the maintenance costs divided by how many hours you can operate it. It comes in at $10 per hour.
And that will go down to $1 per hour by 2035. So if you compare that. with the manual labor rate currently, it's $18 an hour. You see the huge difference you have there, right? So they come in at 50 % of the manual labor rate and goes all the way down to 5 % across one decade. And so if you compare that with the steam engine in the 1800s, came in at... 20 % performance improvement and look what that has done.
the humanoid robots will be a factor of two and a half to 25 that across one So it'll be monumental. And the reason why I really urge now AI adoption and then very in quick succession humanoid adoption with manufacturers is that it's so much fun to surf the wave. If you're ahead of the curve and not prepared. rather than playing catch-up, which many find themselves in that position now in terms of gen.ai adoption. Let's talk about exactly what you mean by humanoid robots,
because I think that most people are used to seeing these robots from Boston Dynamics or from Tesla that can walk around or they can jump, or we've seen that four-legged dog that runs around. But what exactly do you mean when you talk about humanoid robots in manufacturing? Can you be more specific so people can kind of visualize it? Yeah, right. you know, the plon of robots per se, you know, has been underway since the 60s, basically, even, you know, co bots,
they call them with human form factor, but often without legs. So there will be the, for example, a coata nextage or a BB UMI. They do repetitive precision tasks and are deterministic and I talk about when I say humanoid now is where at this moment where it's about to adoptionary take off is the bipedal, gen AI powered, general purpose bots. So they have real disruptive potential because they're so versatile. So examples would be, for example, the figure two by figure that a BMW plants
in Spartanburg and they can put fitted sheets on precise fittings or the Apollo robot by Aptronic that works at a Mercedes-Benz plant, Atlas by Boston Amics, of course, you know, very autonomous in general purpose. You can teach them new things pretty quickly and then they just do a different So basically, ~ you know, terms of for a manufacturer, you're used to buying equipment, and capex, and then it's good for one thing, but if your strategy changes and you don't need that anymore,
that's all sunk cost. Well, with, ~ with humanoids, since they're so versatile, you don't have that phenomenon as much. Like you can just repurpose them and they do something else or they basically, they lose much less value, because even if you even if you have to dissolve your venture, they'll be just as valuable to someone else or even more valuable. And then anyway, the last in the series that's kind of far ahead and prominent and tailored to manufacturing, there's more, there are more in the
healthcare space and so on, but for manufacturers would be Optimus by Tesla, of course. And so yeah, what they all have in common in terms of USPs is both the adaptability that I talked about, but also just this dexterity. so they can do very intricate tasks. what they can be used for is right now, of course, material handling. That's what Eptronix Apollo does mostly, carrying things around. But also visual inspection while they're carrying things around. They also do quality control. And then assembly, like I mentioned,
the figure two, the BMW plant does that very well. And yeah, in terms of, you know, when it makes sense to put these to work, like the perfect storm for adoption, where it's like an absolute no-brainer, besides just the purpose of getting into pole position for, you know, when industrialization takes off, is when you have very adverse conditions or it's dangerous and where you would you know, lot of risk putting a human to work and or when you have a later shortage.
So far from most of these robots that we've seen that claim they can do lots and lots of different tasks, they can't really do it. The Optimus robot, very recently they had a campaign and there was a guy driving it. It wasn't actually making the drinks, like they got caught right away. So, and every time they shoot a video, it's more and more shadow, right? There's always something weird about the videos and while we've seen the Boston Dynamics robot, do a backflip or walk around.
What we haven't seen it do is chemistry or... surgery or do these fine things that like I see Lee promises all the time How far away are we from where a factory can be all robots? I don't know. I can't tell you the answer to that. What I know, and you have a similar phenomenon with Gen.ai as well, right? Where you have this jagged frontier of things that it's good at and things that it isn't good at and you don't really know until you
try and find out. So from that perspective, it's very, let's say, experimental in nature. But you don't get around having a foothold in it. basically, the more you find out and get a feel for it, the more... It's hard to imagine that the trend doesn't continue in an exponential fashion very fast. So it's not like... you put even a GEN.ai tool or you put a... And they're experimenting with it too, The BMW and Mercedes-Benz. And obviously they have a lot of capital behind it.
But you need to basically have a feel for it and be very close to the cutting edge in order to catch the wave when it inevitably comes. So, you know... using that analogy a little more. It's like, if you decide to catch the wave, you better paddle out there. You might not catch the first wave and you might be sitting around for a bit and so on, but... But if you're not out there positioning yourself, you definitely won't catch the wave, right?
Let's take a look at it from a different angle. For a company that's like, well, what's the decision making calculus for if they should take a look at this type of machinery, this type of physical use of AI? What's that process? These companies like BMW and Mercedes, they have massive budgets, massive warehouses, massive campaigns. But for a lot of companies, one of the reasons manufacturers don't change tools very often is that the cost of a machine is millions of dollars.
So they have to, it takes a really long time for them to recover that investment. So for them to switch, right, before they recover investment for the previous thing, throws off all of their old numbers. So I think that's why we're seeing a lot of hesitation with adoption manufacturing, because that's why they have like multi-year buying cycles before they buy each new machine, there's a huge process. So let's talk about that decision-making calculus for a company that's in manufacturing. What are the questions that they should
ask themselves to see if this is going to let them get ahead for their particular industry? Yeah, so, you know, for all practical intents and purposes on the adoption side, with Gen.EI, everyone's a second or third grader, right, including the people that build the foundation models, the creators of those. And, you know, with humanoids, school is basically just starting. So, Um, it's relatively easy to become a leading edge, uh, if we want to, right? So wanting, uh, becomes a key word. And, uh,
that's where the constraint let's say is shifting to in the human, uh, no, it or more AI era. So it used to be, if you think about it, it used to be a number of bodies, uh, back in the middle Asia, all the way up to the industrial revolution. And then it shifted to energy availability. and costs in the Industrial Revolution. And then it shifted to knowledge and know-how down the line. Now it's shifting to, you know, ambition and entrepreneurship and willpower.
So basically wanting. So if you imagine a, know, a miller in France in the Middle Ages, it didn't matter what he wanted. Like it was pretty much predetermined. what his life and his venture is gonna look like versus the knowledge worker heard a lot about, if you just want it enough, you know, you can make it happen and so on. And which is, know, more true than for the French mill in the Middle Ages, but still only a half truth. Now,
you know, know how and those things are commoditizing away then really the AI powered entrepreneur has way less of that constraint. So it's even more wanting and that holds for all size of companies and especially the mid market ones as well because there are less of a disadvantage now with respect to say the Fortune 500 companies, right? And then it really comes down to best practice of deploying, you know, Gen. AI, for example, versus traditional AI tools, right? So traditional AI tools are deterministic
and they'll keep their role while Gen.AI is creative. And so, you we have the Jagged Frontier that we talked about and so it's always evolving. And so you need to experiment quite a bit. And so in terms of the implementation sequence, you know, with traditional IT tools, it would be the tool would be backloaded. So the sequence would be first get the process And that's something that was pretty much an evergreen when it's a project of changing performance and everyone says,
hey, we need a tool. It's like, no, you need a process first. And then the tool comes later once you have a process in place, a tool that fits that process. And now, but with with Gen.ai, it's the tool first. and the new experiment, and then you get the process out of that. And so ideally, everyone should get training on the basics and start using it. now, safety considerations aside, it needs to be very much aligned with the company's priorities. So it's not just wild experimenting,
but you define what is important, what you're trying to achieve as the company. then these experiments, if you want to call them that, or these adoption initiatives cater to that. And that's where AI and human adoption best practice really converges with change management's best practice. And so we can get into that if you want or later on. Yeah, where do want to go from here? Let's dial into this corp. This idea of tool first and then figure out how to use it has
a super high error rate. Because what happens if you buy a hammer? When I go to the hardware store with my wife, if I buy a measuring tape, when I come home, everything's getting measured. I'm measuring the kids. I'm measuring the wall. I'm measuring the bed. If I buy a hammer, something's getting hammered. So we have this problem, and we've seen this where a lot of companies will jump on an AI tool and then suddenly they have to justify that purchase or expense.
It's like, I spent this money so we have to use it even if it's the wrong tool for this situation. So my experience is much more find the problem and then let's solve it, right? Because most people don't know what tools are out there, what is capable of, because there are so many tools, so many platforms. So this is really where I think the crux of challenges for most companies is, is they don't want to make the wrong decision.
They don't want to be driven by desire. They want to be driven by usefulness and ROI and recovering their investment. So what's the right strategy for picking a tool? And let's be more specific as far as implementation of how a company can, even if it's running an AI pilot program, how can they actually get a win very quickly and kind of start to get usefulness out of their team? Because it's very... There's a lot of challenges with AI adoption. know, there's inertia, there's getting
your team to switch over, people are worried they're going to lose their jobs. How can you kind of go through this implementation phase in a way that helps see this as a new tool and how it kind of gets your company a win as a whole? Yeah, so I think it has a lot to do with discipline and going through the right sequence. so it actually goes hand in hand with this change management best practice that I was referring to earlier,
which what we do when we go in and make a transformation or really push for a step change, performance improvement or turnaround, we start with six to eight very clear initiatives. So those are the most important change efforts of the company. each project, so each initiative gets spearheaded by an organizational leader. So it's not some project manager, but really some functional head, maybe the plant manager. or C-suite member that spearheads some effort and has all the context that they
to push for that improvement. And then we set up program group meetings every two weeks. It seems to be a pretty good cadence where all the project leaders come together and that creates a couple of things. So the first one is shared consciousness. That's a concept popularized by Special Operations Forces commander Stanley McChrystal in his book Team of Teams and he describes it very well how you know, they were losing the the fight against the terrorists in Iraq and Afghanistan and because they were just
too slow and the different functions didn't know exactly what the other ones were doing and how their efforts were you know played into that and so That's what we really achieve with these large group meetings where every guru would say, that's very inefficient. That's too many hours that we're burning through. If you have 10 people in a room for two hours, that adds up quickly. But everyone leaves from there with really good understanding not only of their own project, but of everyone else's projects too and
how they tie into what they're trying to achieve. And, ~ and that really helps, yeah, both with, with buy-in for, the others projects and giving input so that, ~ they're actually, they have all the context they need, so to say, to, to target the right thing. And, and then quick dissemination of, best practice, right. And that holds for, for AI adoption as well. where if, if someone finds, Hey, you know, I realized I can do this. tool that helps me keep part shortages low
with this AI tool, then the others can pick up on that and do the same thing. that keeps the learning cycles at a minimum. Let's say only one team has to find out and then the other ones know at least and some can use the same best practice. And then of course it creates these group meetings, create this accountability where you see constant progress every every two weeks or so. then in between the weeks in between, we do one on one coaching with the project leader,
basically, you know, sit together and work through a problem. And that's where so you historically it's been a lot of infusion of functional expertise, right, commercial excellence, supply chain, practice and so on. But that can also be used very well for AI adoption until now and now in the future also, humanoid adoption best practice where maybe you have an expert or two who based on the various situation that the company's in, the project is in, what the project tries to achieve can bring in
the right or suggest the right tools for tackling a certain problem. And then basically keep that expensive experimenting that you talked about at least to a minimum. You won't avoid it, but you can at least minimize it. I think that's a really useful strategy. I think people are gonna find this really, really interesting. So, and for people who wanna know more about you, wanna work with you, kind of see the kind of projects you work on, who have mid-Atlantic businesses,
where can they connect with you online? Where's the best place to find you and see the projects you're working on, maybe even work with you? Yeah, so you can reach me on LinkedIn. the link is linkedin.com, jmobius, slash jmobius, or www.infinity-partners.co. this was amazing. Thank you so much for being here for another amazing episode. Fantastic, yeah, thank you. Awesome. Thank you so much. And bye, Johnson. Have a good one.
More on ai in manufacturing, energy and infrastructure
