Beyond the Bot Ep. 12 Live: Becoming “Future-Proof” in the Era of AI and Robotics

Tony and Steven recording live for Beyond the Bot Episode 12

In this Beyond the Bot live session, Anthony DeHart and Steven King dive deep into the practical realities of AI-powered automation—from robotic arms on factory floors to AI agents making strategic decisions. Speaking candidly with an audience of technologists, entrepreneurs, and students, the duo shares not just what’s possible, but what’s actually working—and what still needs careful human oversight.

Whether you're navigating robotics in manufacturing, deploying AI tools in knowledge work, or simply trying to upskill in a fast-moving digital landscape, this conversation offers real-world lessons on transparency, implementation, and the ethics of automation. This episode is full of insight, humor, and tactical guidance for anyone wondering where to start—or how to grow—with automation.

Transcript:

See Beyond the Bot Episode 11 for the first half of the conversation.

Steven King: Other questions on that manufacturing side of things?

Audience Member: Can I just ask one deeper question on what you were just describing? The large language models had a lot of trouble explaining what they were doing, right? And that seems to be improving. Can you ask the system to show you what it’s thinking about as it tries to figure out how to pick up the object?

Steven: Pattern recognition—and part of AI—has been around since the 1960s. We’re leveraging those techniques to detect patterns in imagery. Yes, we can output what patterns the system detected. We can output yaw, pitch, XYZ coordinates. It’s not easily understood by humans, but we can provide the raw numbers. Basically, when the system sees an object at a specific angle, those numbers tell us, “it should go to that angle.”

In most cases we don’t analyze those deep numbers manually, but we can output them to understand why it made a particular decision. If a robot crashes, the first thing we do is ask: how did it get there? Why didn’t a safety bound stop it? We’ll dig into those numbers for that scenario.

Anthony DeHart: One exciting aspect of reinforcement learning: while the system can’t always explain in words why it made a particular move, we can run it hundreds of thousands of times in simulation and observe aggregate behavior. That gives insight into how it perceives and reacts to its environment.

Steven: This is powerful—you can crash the robot a thousand times in simulation before ever deploying it in real-world environments.

Audience Member: When you’re doing that video portion, roughly how many iterations are you doing? And is it a derivative of a large language model you’re working with? How many times—hundreds, thousands, millions?

Steven: It’s closer to 100 iterations. It depends on the precision needed. If you’re picking up a standard ball, maybe 20 to 30 runs suffice. But if it’s an oddly shaped part—like something from a Honda motor—we might need 100 to 200 runs. We’re not talking millions.

Tony: That’s where mechanical design complements AI: for example, our end‑effectors are designed with a softer middle and firmer edges. That creates flexibility—you don’t have to be precise down to the micrometer to grasp an object. You could pick up an egg or flip a switch with the same gripper.

We’ve covered a lot of pitfalls and best practices using AI. Steven, here’s a question many face: should they implement this themselves, or partner with a teammate or vendor?

Steven: First, make sure the process is solved by humans before automating. Humans are great at repetitive decision‑making. Once you’ve figured out how to do it manually, then you can automate it.

Next: assess your internal capabilities. Do you have mechanical, electrical, and software/vision engineers? If yes, you may be able to program a robot on your own. If not, consider buying a robot system and partnering with a vendor who can integrate it into your warehouse or factory, train your staff, and provide ongoing support.

One key recommendation: work with providers who have skin in the game. You pay them for uptime. If the system goes down, they should have financial incentives to fix it quickly—not just say, “good luck.” Ensure your vendor is aligned with your goals.

Tony: And how does that translate to knowledge‑type systems like chat tools?

Steven: We recently talked with a company with a homegrown financial system—built by one person who left. They’re now paying someone big fees to maintain it. Nobody wants to be held hostage like that.

The same applies to AI: you must understand and manage it internally. You need to ask the right questions about data, servers, and control. As you design your systems, think about long‑term ownership. Don’t just rely on a vendor who solves your problem today—think about who will be there tomorrow.

Tony: So, should someone go get a PhD in data science and then use ChatGPT?

Steven: Ha. No—not that sequence. Start by solving real problems. Sometimes a data scientist helps with data analytics and insights. Other times, the front‑line employee using the tool every day is the best person to deploy it.

Remember: time in the market beats timing the market. Get hands‑on—even if you’re not an expert. Learn the terminology and how it works. That’s more valuable than a degree alone.

Tony: Totally. For example, we have cost‑effective robotic arms that you can clamp down on a desk and start the same day. Our software is no‑code—drag, drop, and go. You can start simple: automate one task. Once that works, expand sequentially across the line.

Steven, a common question we get is: if someone wants to upskill in AI without becoming an expert, where do they start—other than Beyond the Bot?

Steven: Attend events like this, and watch Beyond the Bot, of course. I have a 15‑year‑old and 13‑year‑old, and in our house we say: use technology to learn and grow, not to escape.

For example, I wrote the syllabus for our AI tools class by asking ChatGPT: “What are the most important tools a media professional needs today?” It gave me tools I hadn’t known. So I subscribed, explored them, and now I’m building AI‑driven ads.

Be curious. Use the technology to improve yourself or your business.

Tony: I’d add: beyond textbooks, your best training is on LinkedIn, tech meetups, and Innovate Carolina. You meet people at similar experience levels and share best practices.

Audience Member: I use Copilot for code coverage. We aim for 95% coverage—and it takes two hours instead of two days. The thing is, when reviewing candidate GitHub repos, you can’t always tell if the code is theirs or generated by AI like ChatGPT or Copilot. But humans show their mistakes—if someone doesn’t understand architecture, you’ll spot poor implementation.

Tony: Exactly. Copilot is a great example: it’s fantastic if you understand architecture and processes. It doesn’t replace problem‑solving skills.

Audience Member: I’ve started using data‑process automation with ChatGPT, fuzzy logic, and first‑principles thinking… I also wanted to ask about “live coding” or “vibe coding”—what’s its impact in the next six months for non‑developers or when hiring?

Tony: Live coding can be amazing for prototyping: someone might generate a mobile app very quickly. It can even achieve market success—100,000 users, for instance. But the code may not scale or be architected well, so hiring and production decisions depend on business needs.

Here’s a story: during early-stage UI prototyping, one of our senior developers copied a 10‑minute meeting transcript into a tool that generated three clickable UIs in 10 minutes. Sure, none is production‑ready, but having live options saved months and allowed fast A/B testing with users. That’s a killer use‑case.

Audience Member: Any cool examples of ROI or standout use cases?

Steven: I love this one: Space Force asked if we could use our vision system to detect space junk and jettison it into the atmosphere. Super cool—but unfortunately the project just wasn't a good fit for where we were at that time as a company.

One real ROI case: three months after deployment, a customer reported using 70% less paint than humans—getting ROI in just three months! They saved money and reduced hazardous waste.

Tony: Another fun idea: experiential robotics.

Steven: Yeah, we had a company that wanted us to make a huge crane game. Like, a 20 × 20 × 20 ft arcade crane. It would have been a world record! They walked away due to cost, but it worked as specified—arcade mode or using computer vision to get a perfect pick for every user.

Audience Member: Reading about AI agents—like systems that not only book your vacation but also execute tasks—how does that relate to robotics?

Steven: Think of physical agents (robots) like physical manifestations of AI. There’s discussion around humanoid robots because they fit human environments— Actually, I'm gonna stop for a minute and let's talk about humanoid robots.

The reason we have humanoid robots and they've grown so quickly is because, if you remember the nuclear explosion in Japan, the robots could not get up the stairs and they couldn't open doors. So the push for humanoid robots was to be able to work in a human environment that was unsafe. That is a great, valuable use case.

The majority of humanoid robot use cases we're seeing today could be done by a task oriented robot much easier. But we're not there. We haven't got to that realized place, but I think there is a place for the humanoid robot.

So your question is around this idea of an agent- can a physical agent act on my behalf? The real question is, do you want agents that just give insights, or ones that act? Autonomous vehicles act on recommendations—for safety, they’ll hit the brakes, not just warn. Do you want a robot to “push the button” in the physical world? Ethics and testing are critical.

Booking dinner is low stakes. But in physical tasks, mistakes can be severe. We advocate extensive closed‑environment testing.

Tony: We also differentiate automation wisely. Human interaction matters, so just because it can be automated doesn't mean it should be. And that’s something we think about a lot—especially when we’re working with clients in spaces like hospitality. We always ask, what are the parts of your process you can automate in the back of the house, so you can be more present and engaging in the front of the house? That’s the kind of balance we’re looking for.

Steven: We don’t want to automate the things that create those memorable guest experiences. We want people to still feel that human connection. Personally, I’m not a fan of putting robots out there facing the public in hospitality settings. But if automation can help you do your job better behind the scenes—if it gives you more time and energy to serve your guests—that’s where it really makes sense.

Audience Member: Have you found that people are resistant to AI in those spaces? Are they worried about transparency or how it’s being applied?

Steven: Yeah, absolutely. That’s something we take seriously. For example, we went to visit a company just this past Friday, and we intentionally didn’t wear any branded gear. Because there's this very real fear: “Are these people here to replace my job?” So we try to be sensitive to that right from the start.

When we work with a company to deploy robots, we talk with them about how they’re going to communicate that to their staff. We had one client who hadn’t even signed the contract yet, but they went ahead and mentioned it in their internal newsletter—just, “Hey, we’re switching to robots.” And people freaked out. That kind of rollout doesn't work. The boss was excited, sure, but it caused real concern among the team.

So we really try to work with each organization to understand where their stakeholders are coming from—and that includes the line workers who will be working side-by-side with these robots. They need the right training, and more than that, they need to feel comfortable with what’s happening.

Tony: One small but really effective thing we do: we give the robots names. Like, actual human names. We’ve even added googly eyes to some of them. It sounds silly, but it changes the way people relate to the machines. It reframes the robot from being a tool that’s replacing you to being a coworker who’s helping you.

Steven: And it has a measurable impact on productivity. People are more likely to help the robot if it’s struggling to reach something, or to work with it more intuitively. It’s those little micro-decisions that add up. That kind of buy-in matters.

Tony: And to your point about transparency—honestly, a lot of people think automation means giving up control or visibility. But I’d argue it’s the opposite. When you automate a process, you get this incredibly detailed audit trail. It’s like having someone there constantly documenting every input, every decision, every outcome. You can go back and see exactly what happened, from beginning to end.

Steven: And when you’re writing the program, you’re front-loading all the control. You’re deciding how the system behaves, and then you can be confident that it will behave that way—every time. That kind of consistency is incredibly valuable. You’re not leaving it to human variability. You're designing it once and knowing it’ll execute the same way again and again.

Steven: Tony, I'm sure you had a good closing point you wanted to end with. So maybe we wrap up with that?

Tony: Yeah, thanks for teeing that up. You know, we’ve talked about a lot today—there’s so much happening in this space. But how would you answer the question “How do I future-proof my business?”

Steven: Honestly, I don’t think you can future-proof anything. I hate that term.

What you can do is prepare for an unknown future. I’ve mentioned my kids before—I don’t know what jobs they’ll have after high school or college. I wish I could tell you. But I can’t.

What I can do is make sure they’re good problem-solvers. That they know how to communicate, tell stories, write well, and interact with people in meaningful ways. That’s the stuff that lasts.

And it’s the same with our company. I want us to be flexible, to be able to move quickly when things change. To be able to recognize a trend and adapt fast. That’s where smart automation helps us—when you use AI to make yourself more nimble, not less.

So I don’t say we’re future-proofing—I say we’re trying to build for an adaptable future. And to do that, you need to train your team, engage them, and build systems that can flex with whatever’s coming.

Tony: And I’ll just add—on a personal level, this is all about maintaining curiosity. Attending events like this, reading, experimenting, talking to people—that’s how you stay sharp. It’s not about having all the answers right away. It’s about staying in a learning posture and seeing this technology as an opportunity, not a threat.

And if you want to keep up with that kind of mindset, following things like Beyond the Bot helps—LinkedIn, Facebook, YouTube… I know, shameless plug.

Steven: Nice plug!

Tony: Thanks! But seriously, we’re so grateful to everyone who joined us. We’ll be around afterward if you have more questions. And if you want to connect later, you can find us on LinkedIn. Happy to help however we can. Thanks again, everybody.