From self-flying cargo drones to electric boats and autonomous trucking, Kofi Asante has been at the forefront of transforming how goods move through the world. As VP of Business Development and Market Expansion at Arc Boats—and with prior executive roles at Elroy Air and Powerloop—he brings a rare, cross-modal perspective on autonomy in motion.

In this conversation, we dig into the complexity of real-world autonomy: what it takes to navigate open skies versus tight factory floors, how battery tech shapes the future of robots, and why safety and reliability must be designed from the ground up. Whether in airspace, on water, or across roads, Kofi’s insights illuminate the practical realities of scaling autonomy in logistics—and what it will take to move from prototypes to real-world performance.

From Air to Water to Road: Designing Robots for a Multimodal World

Ashutosh Saxena: You’ve worked on robots like electric drones, boats, and trucks. How do the autonomy and power management challenges differ across these environments, and what’s surprisingly similar?

Kofi Asante: I’ve had the chance to work across a few very different domains—starting with autonomous trucks at Uber Freight, then building hybrid-electric aircraft at Elroy Air, and now electric boats at Arc. Each mode comes with its own set of constraints—and opportunities.

With trucks, the powertrain tech is mature. You can build a capable electric drivetrain that fits the vehicle and performs well. The real challenge is infrastructure. You need charging networks at scale to support long-haul routes. That’s where the bottleneck is—it’s less about the vehicle and more about the ecosystem around it.

Aerospace is a whole different problem. You’re designing propulsion systems that need to fit within tight volume and weight limits—and meet intense safety and certification standards. When we built the powertrain for our VTOL cargo aircraft at Elroy, we spent millions just getting that system right. It’s one of the most complicated problems in autonomy. Many teams underestimate this—and it costs them years and significant capital.

Maritime is what pulled me in next. It’s much more forgiving from a physics standpoint. You can build massive battery systems—multi-story battery packs—and weight isn’t as much of an issue. If it floats, it works. The interesting part becomes infrastructure planning. We focused on tugboats at Arc because they operate on fixed routes—point A to point B. That repeatability makes charging much more manageable.

And just recently, we launched the first electric tugboat in the Port of Los Angeles and Long Beach—the largest port in the Western Hemisphere. It ran beautifully. That kind of milestone helps validate what’s possible when you match autonomy with the right operational profile.

So to break it down:

• Trucking has the power tech, but needs supporting infrastructure.
• Aerospace has extreme design and regulatory constraints—it’s a high-stakes problem.
• Maritime gives you more flexibility, and the fixed routes make autonomy and electrification much more practical.

Across all three, the throughline is clear: the tech only works if the environment supports it. Autonomy doesn’t live in a vacuum—it has to plug into the real world.

Mastering the Air: Why 3D Autonomy Is So Hard

Ashutosh Saxena: Compared to ground vehicles or factory robots, drones operate in actual 3D space with fewer physical constraints. What are the biggest autonomy challenges in aerial logistics that people often overlook?

Kofi Asante: When I transitioned from working on autonomous trucks to aerospace, I assumed aerial autonomy would be a simpler problem—at least from a software perspective. After all, you're not dodging traffic or dealing with pedestrians. But I quickly learned that the real challenge in aerospace isn’t the autonomy itself—it’s certification.

Most of aviation has been built around the assumption that there’s a human pilot on board. So when you remove the pilot and ask the FAA—or any other regulatory body—to approve a fully autonomous system, it introduces a whole new level of scrutiny. You’re not just proving that the system works. You’re asking for permission to let a computer make critical flight decisions—at scale.

Now, in places with low ground risk—like rural areas or over water—you can make a stronger case for autonomy. And in fact, this is already happening. If you look at Rwanda or Ghana, drones are used to deliver the majority of the country’s blood supply. I visited one of these operations in Rwanda, and it’s amazing. My family is from Ghana, and it’s happening there too. The routes are short, the payloads are manageable, and the terrain is predictable.

But when you move up into larger aircraft—like the cargo VTOLs we were building at Elroy—you’re operating at the scale of small helicopters or planes, carrying thousands of pounds. That’s where things get complicated. There's no precedent for flying something that big autonomously, so you're skipping steps in the regulators’ eyes.

That said, our aircraft are already flying autonomously in test environments. The autonomy works. What remains is navigating the commercialization path—getting it certified for broader deployment. Meanwhile, the military is already flying large autonomous systems, but they have their certification processes that don’t translate directly to commercial use.

Environments in Motion: Complexity Across Modalities

Ashutosh Saxena: One of the core ideas we think about at Torque is how generative AI can help robots navigate challenging, unstructured environments—whether it's dust on a farm, traffic in cities, or dynamic obstacles in warehouses. Across trucking, aerospace, and maritime, what were the most challenging environments you had to design for?

Kofi Asante: Great question. I’ll break it down by mode:

Trucking: When I was building Powerloop at Uber Freight, we realized that long-haul routes were relatively predictable—you’re just driving in a straight line for hours. But short-haul routes? That’s where things got messy. You’re dealing with urban environments, stop-and-go traffic, pedestrians, dogs, kids running into the street—basically, all the edge cases.

Our solution was to “loop” power units between trailers: let humans handle the short-haul trips while autonomy tackled the simpler long-haul ones. It was safer, more efficient, and allowed drivers to stay closer to home. But with systems like TorqueAGI that can handle higher complexity, I think short-haul autonomy is becoming more realistic.

Aerospace: The airspace system is incredibly well-documented. The FAA classifies airspace by complexity, and operators typically aim to fly in lower-risk zones—often at military bases, where you control the entire airspace. That’s how we approached our early tests at Elroy. The complexity increases as you move into public airspace, but there’s a clear path for scaling once you know what class you’re operating in and how to get certified.

Maritime: Autonomy on the water has its unique challenges. On the recreational side, most people still want to drive their boats—it’s part of the fun. But there’s growing interest in assistive autonomy for docking, trailering, or route planning. On the commercial side, things are shifting more slowly. There are strong labor unions and legacy regulations that still favor having people on board. But even without full autonomy, the shift to electric powertrains is already saving operators thousands of gallons of fuel—and that opens the door to more intelligent, hybrid systems over time.

So across the board, the complexity is real—but it’s also contextual. Whether it’s traffic, turbulence, or tides, the question is always: can the system understand the environment well enough to act reliably in real time?

Fueling Autonomy: How Battery Tech Limits—and Enables—Robotics

Ashutosh Saxena: Battery performance remains a significant constraint across all robotics. What improvements in battery tech or energy management would most accelerate progress in autonomous logistics?

Kofi Asante: This, too, is a great question—and one that sits at the intersection of all these industries. The biggest unlock right now isn’t necessarily new battery chemistry. It’s cost.

If we can bring the cost per kilowatt-hour down by 50% from where it is today, that becomes an accelerant across the board. And the good news is: we’re getting there. Thanks to the billions of dollars invested in automotive and aerospace, we’re at a point where the core technology—cells, modules, battery management systems—is good enough for many robotics applications. You don’t have to reinvent the wheel. You can leverage proven, off-the-shelf components and focus on integrating them into your platform.

So from here, it becomes an economics challenge more than a technical one. Especially in sectors like maritime, where size and weight constraints are more forgiving, it’s a matter of:

• How much battery can you fit?
• And can you do it at a cost that beats or at least competes with diesel?

Then there’s the infrastructure question. Do you build out charging? Or do you go hybrid and put a generator onboard to extend range? Those decisions vary by use case—but they’re solvable.

At Arc, we’re especially excited about what we’ve been able to build with American-made battery packs. We’ve driven the cost down significantly, which is what ultimately enables broader adoption. When the economics line up, that’s when you see real acceleration—and that’s true in any industry.

Safety, Reliability, and the Path to Scale

Ashutosh Saxena: Designing for autonomy means managing risk—but those risks vary widely depending on the environment. How do you think about designing for reliability when the operational profile is so different for, say, drones vs. boats vs. trucks?

Kofi Asante: I usually think about this through two lenses: business and technology. And they overlap more than you’d think.

From a business perspective, there’s a basic sequence we try to stick to:

Demand → Margin → Scale.

You don’t want to build or scale anything unless you know there’s real demand for it. Then you validate the economics—can you make money doing this? And only then do you scale. If you skip steps, you end up burning a lot of capital and running into sustainability issues fast.

That approach guided how we built Uber Freight and Powerloop. We started in one dense region—the Texas Triangle—and focused on getting the experience right for just one driver, then ten, then hundreds, then thousands. The same pattern applied: prove demand, validate the margins, then scale. That discipline helped us grow sustainably without reinventing everything from scratch.

And we had great mentors internally. People like Kevin Novak—who had been there from the early days of Uber’s rideshare expansion—influenced how I thought about scaling. They had already figured out how to launch in one market, learn quickly, and then replicate that across new geographies. We applied that same framework in freight.

On the technology side, especially in aerospace, you’re dealing with far longer validation cycles. To give you perspective: getting a new aircraft platform from design to commercial certification can easily take 10 years and a billion dollars. And that’s baseline. It’s a hard constraint—not because the tech isn’t ready, but because the regulatory and safety bar is (rightfully) high. These are vehicles that need to work flawlessly in high-risk environments.

So we tested our aircraft in highly controlled environments—military bases, deserts—where risk to others was minimal. That gave us the space to prove reliability before going to market.

At Arc, we took a similar approach but in the maritime domain. We launched the Arc One, learned from that platform, then doubled the battery capacity for the Arc Sport, and eventually tripled it for our electric tugboats—600 kWh battery systems. We also deliberately chose to start with the Port of Los Angeles and Long Beach, the largest port in the Western Hemisphere and a leader in green infrastructure. The support there—policy-wise, commercially, and financially—gave us a strong foundation to grow from.

Just like Uber did with rideshare—starting in San Francisco, then expanding city by city—we’re looking to replicate that in maritime. Once you prove it at one port, you can take that playbook to other ports in the U.S. and eventually global markets.

The key is not skipping steps:
First, prove demand. Then, validate the economics. Then scale.
And always do it in environments where mistakes can be managed and learned from safely. Once those fundamentals are in place, that’s when you earn the right to scale.

Accelerating Time to Market

Ashutosh Saxena: That timeline you mentioned—a billion dollars over 10 years—highlights how long and expensive these iteration cycles are. At Torque, one of the things we emphasize with our AI stack is helping robotics customers shorten time to market. Whether it's in aerospace or agriculture, reducing iteration time can have a massive impact—not just for a single company, but for the industry as a whole.

Kofi Asante: Absolutely. That’s a critical point. If you can shave even a year off that cycle—or reduce capital requirements by a few hundred million—you unlock real progress. It’s not just a cost-saving. It accelerates the pace of innovation across the entire ecosystem. And often, it also means cleaner, higher-performing solutions get to market faster—which is a win for everyone.

From Pilot Purgatory to Real-World Scale

Ashutosh Saxena: From your vantage point—as both an operator and investor—where are you seeing autonomous logistics systems gain real traction? And conversely, what’s still stuck in what we sometimes call pilot purgatory?

Kofi Asante: Yeah, that “pilot purgatory” term is spot on. I’d say the systems that get stuck are the ones grappling with edge cases—especially in environments like last-mile delivery. You can often get a system to handle 80% of scenarios fairly reliably. But that final 20%? That’s where it gets expensive. Solving for those long-tail conditions can take billions of dollars and years of iteration—not just for one company, but across the entire industry.

You can demo a system that doesn’t have a driver or pilot onboard, and it might look impressive. But unless you can make it robust enough to handle real-world complexity, you don’t get the economic upside. That’s why TorqueAGI is so exciting—because you’re focused on solving the hard stuff. Those last-mile, last-percent challenges that would otherwise eat up vast amounts of capital and time.

Take autonomous trucking, for example. I was working on that back in 2017. The vehicles could drive—technically—but we’re still not seeing widespread deployment today. The autonomy works in many scenarios, but if you still need a driver onboard, you haven’t unlocked the real economic value. You’re not getting the full benefit until you can entirely remove that human requirement.

Ashutosh Saxena: Right. That’s the inflection point.

Kofi Asante: Exactly. And while consumer-facing use cases like last-mile delivery get a lot of attention, the real traction is happening in industrial environments. Honestly, the least “sexy” applications are often the most promising.

Think:

• Mining operations
• Rural logistics
• Construction
• Advanced manufacturing

These are environments where you're not interfacing directly with consumers, and the operating conditions can be more controlled or repeatable. Many founders I know are building B2B autonomy companies in these sectors—and they’re finding real product-market fit. These applications may fly under the radar, but they power the backbone of our economy. They scale quietly—and profitably.

From Demo to Deployment

Ashutosh Saxena: That reminds me of something we talk about at TorqueAGI all the time: demos are easy. Building autonomy that works in a controlled environment is one thing. But getting it to work in the real world—where variables aren't fixed, and obstacles aren’t staged—that's the real challenge.

Kofi Asante: Exactly. In aerospace, especially, getting from a drawing board to a flying vehicle is an emotional moment. You’ve poured so much capital, time, and effort into getting a system off the ground—literally. But that’s just the start.

Now you have to land it, operate it repeatedly, and deploy it in new environments. And that’s where things often need to be reworked. You’re not just solving for “can it fly?”—you’re solving for can it be deployed at scale, across diverse environments, with high reliability?

That’s where teams like yours at TorqueAGI come in—bridging the gap between impressive demos and real-world deployments that scale. 

Beyond Perception: The Role of Intelligence in Autonomy

Ashutosh Saxena: As vehicles become more autonomous, how important is it for them to not just navigate, but to truly understand and reason about their environment? What role do you see for AI models that go beyond just perception?

Kofi Asante: When I look across the different vehicles I’ve worked on—trucks, drones, boats—what they all have in common is the need to be both safe and reliable. It sounds simple, but that’s where all the complexity lives. If a human operator—whether it’s a driver, a pilot, or a sailor—can make decisions that keep the system running smoothly, then the bar for autonomy is to match or exceed that level of decision-making.

You can’t just swap in autonomy and expect the same outcome unless the system can reason about its environment, not just perceive it. That includes understanding edge cases, adapting to changing conditions, and making judgment calls that humans are often very good at—especially in industrial settings where downtime is a significant cost.

And you can’t trade off safety for autonomy. Not even a little. Especially in sectors like logistics or maritime, where the operational profile is highly demanding. These systems power critical infrastructure, and reliability is non-negotiable.

So yes, performance needs to match—or more realistically, exceed—human capabilities. Not because machines are held to an unfair standard, but because the stakes are high. Even if your system is statistically ten times safer than a human driver, a single high-profile incident could shut down an entire program.

That’s been consistent across all the modalities I’ve worked in:

• Is it reliable?
• Is it safe?

Those are the first two questions everyone asks—whether you’re in aerospace, trucking, or maritime. And getting to “yes” on both requires intelligence that goes beyond navigation—it requires real-world reasoning.

Build or Buy? Deciding Where to Own the Stack

Ashutosh Saxena: You’ve helped build companies and now you back them as a VC. For robotics and logistics players, where should they build in-house intelligence—and where should they partner?

Kofi Asante: It depends on the company’s strategy and what they want to be known for. We are interested in both types of companies:

• Ones like TorqueAGI, which offer modular intelligence that others can plug into
• And ones that need to build deeply in-house because it’s core to their value proposition

That said, there’s a general rule I’ve seen work well:

If there’s a solution off the shelf—and it continues to evolve faster than your internal team can keep up—don’t try to outbuild it. Use it. Don’t burn capital reinventing the wheel.

Even massive companies with billions in the bank can get this wrong. Suppose they try to own autonomy end-to-end—or build out their battery tech or charging infrastructure—without a clear edge. In that case, they end up wasting time, losing momentum, and sometimes even demoralizing their teams. I’ve seen this firsthand in EV companies and maritime infrastructure. It’s tempting to vertically integrate everything, but very few companies can do that well—and none start that way. It takes decades and billions to earn that level of control.

“If getting it wrong doesn’t break your business, it might not be your core. Outsource it.” — Kofi Asante

So I always ask:

If you get this piece wrong, does your business fall over?
If the answer is no, then it might not be core—and you might want to outsource it to someone whose entire business depends on getting it right.

At Arc, our electric propulsion is core. We have to be the best at that in maritime. But that doesn’t mean we need to build everything else ourselves—not today, anyway. Over time, we may earn the right to integrate more. But you have to be strategic about where you spend your cycles, especially in capital-intensive industries.

Ashutosh Saxena: That’s such a great framework. You started by saying you didn’t have a clear answer, but that was incredibly clear. It comes down to focus—deciding what you want to be the industry leader in. If your core business is farming or mining, consider not building autonomy from scratch. Let a partner like TorqueAGI handle that. But if AI is your edge, then go deep.

Kofi Asante: Exactly. And thanks—appreciate you helping distill that. It’s something I’ve learned through a lot of hard-earned lessons, both as a builder and now as an investor.

Future Forecast: What Are You Most Excited About in Autonomous Logistics?

Ashutosh Saxena: Whether it’s a breakthrough in autonomy, an energy innovation, or a real-world deployment—what frontier in autonomous logistics are you personally tracking with the most excitement?

Kofi Asante: For me, it keeps coming back to autonomous industrial applications. I’m bullish on that space.

A few examples: there’s a company called Hadrian, part of the same A16Z American Dynamism portfolio as Arc. They're building automated manufacturing systems that are 10x faster than traditional setups—and I just toured their facility. It was incredibly impressive.

Another is Bedrock, which is tackling industrial autonomy in a different domain. The team spun out of Uber and includes some of the smartest operators from Waymo and elsewhere. They’re backed by Eclipse Ventures, which also led Arc’s Series B. I know those folks well, and I think it's essentially an execution game—and they can execute.

And of course, there’s TorqueAGI. When we decided to partner, it was because we understood just how much complexity companies face when trying to operationalize autonomy. Having a team that’s focused entirely on making that complexity manageable—that’s a huge unlock for the entire ecosystem.

So, whether it’s mining, construction, or manufacturing—all these sectors that most people don’t interact with directly—I think that’s where autonomy will gain a strong foothold first. The stuff that’s out of sight tends to be where fundamental transformation begins.

Ashutosh Saxena: That’s a great note to end on. Thanks so much, Kofi—this has been a fantastic conversation.

Autonomy That Moves the World

As Kofi Asante makes clear, the future of logistics won’t be built in a single mode or medium—it will emerge across a spectrum of environments, each with its own physical constraints, safety requirements, and energy demands. Whether navigating the open skies, the open ocean, or dynamic ground-level infrastructure, autonomous systems must be intelligent, adaptive, and reliable by design.

What unites these domains is the need for scalable intelligence—AI that does more than steer, that can understand, reason, and respond to the unexpected. That’s where the next breakthroughs will happen. And as autonomy moves from R&D to real-world deployment, leaders like Kofi—who’ve operated across air, water, road, and venture—remind us that scaling autonomy is as much about practical execution as it is about technical vision.

The journey from prototype to performance is already underway. And it’s powered by intelligence that’s built for the real world.

Whether you’re dealing with dynamic environments, moving objects, or difficult weather conditions, TorqueAGI is ready to add even more intelligence to your robotic stack. Contact us to schedule a demo and discover how we can assist you.