Deciding Which Roles Go To AI or To Humans, According to Russell Anthony

Check out Kingsmen Digital Ventures on Clutch. Read client reviews to find the perfect partner for your upcoming project.

One of the Kingsmen: Introducing Russell Anthony

Russell, tell me more about Kingsman Digital Ventures and how it got started and your role there.

Russell: Awesome. So we started Kingsman actually 10 years ago in 2016. And at the time I was an engineering manager for an electronic health record and medical billing company out here. Before that, my career was all startups, zero to one type thing. And what I learned working at that company was just the difference in quality of software that it takes to manage billions of insurance claims, both performatively and securely, and the difference in architecture that it takes to build that kind of real quality software compared to the scrappy little startups I had done before. And the thesis at Kingsman was that there were a lot of companies that weren't traditional software companies that could benefit from that type of software. And our first client was a local healthcare organization where they had personal trainers, nutritionists, chiropractors, doctors all under one roof. And they wanted to build software that could integrate their different systems with the different compliance requirements to streamline their internal operations as well as the customer experience. Things like even being able to track testosterone levels against strength versus supplement purchases.

AI Won't Replace Your Team, But It Can Free Them

How have you transitioned some of those roles that were done by staffers now with AI, but also making sure that the quality, especially for those doctors you mentioned, stays at par?

Russell (02:12): I do pride myself on as long as we've been in business, our systems, they've never gone down, they've never been hacked. And that to me is table stakes. That's not going to change. So I would say I haven't transitioned any roles to AI, and I don't plan on it anytime in the near future. I don't believe what we're calling AI nowadays, the LLM is ever going to achieve AGI. What I have transitioned and where I do see opportunity is transitioning responsibilities. Even things like writing detailed Jira tickets from requirements and Figma designs, you can save a lot of leadership time by automating that part of the process. Code reviews, the initial review. It's one thing to just set up GitHub Copilot and say, "Hey, this code looks okay," versus actually setting up MCP in context so that it can help review against the requirements. And really to me, what we're talking about is gaining efficiencies, not replacing people.

How to Experiment With AI Without Burning Your Clients

So these engineering decisions then become more of effective tooling. And you talked about code review, you talked about maybe some product requirement documentation. Are there any other things that you've locked in that your engineering team really uses or are you allowing them to experiment on their own?

Russell (03:27): So for me, again, the quality of what we release is necessary. So any kind of experimentation that we're doing, I do it at our cost. When there are gains, obviously we want our clients to benefit from them. But if I mess around with, let's just say I'm going to try to use cursor, divide code an entire app, and then I look at it, I'm like, okay, this is not the quality we expect. I'm not going to, A, impact the client's timeline so I can experiment with and say, "Hey, we messed up," but I'm also not going to build up for that. So it is very experimental and really my approach is finding workflows that actually make a difference and then rolling them out to teams where they get leveraged from those tools.

Ownership is Everything: Treating AI Like a Junior Engineer

How have you layered in the accountability in your team that when pull requests are made, is it certain levels and who's managing that layer when something breaks?

Russell (04:34): Great. So I'm of the opinion that managing AI is kind of like managing a junior engineer. AI, for example, there's automated dependency upgrades you might see pop up on GitHub. I mean, you could even set up OpenClaw to be doing some of this stuff for you and saying, "Hey, I implemented this because I think it's a good idea." And it's just like hiring a junior. They might be very ambitious and they may sometimes have brilliant ideas that actually matter, and sometimes they might have ideas that are a little misguided. And management's job is to manage that so that what comes out of it is effective and good for the customer, for the company, and also so that the engineer doesn't get burned in the process. So to me, whether a junior fresh out of college puts up a PR that breaks something or AI does, the team that owns that system is responsible for what happens when it gets released.

Now, you can automate and streamline a lot of that with good test coverage. Really, you should have good automated signals that tell you if something's going to break before it goes to production. You never want to find out and prod that something was broken because we didn't test it enough.

But really, at the end of the day, one of the most important principles you can have at any company is ownership. You don't just merge the PR because someone told you to. It's like, no, I'm responsible, you're responsible, we're all responsible because this is our project, this is our baby, and if we're going to release some code, we're going to make sure it works.

So do you make them understand the markdown files as they're doing it or have people gotten spoiled a little bit with, "Hey, AI understands the system. I'm just going to prompt it to do some of the work?

Russell (06:15):  Again, I think I like to compare it to managing a junior engineer. A good lead will give very clear expectations of, here's what I want, here's how to do it, here's how to implement it. Not here's how to write the six lines of code in this function, but generally speaking, here's the structure of the deliverable, and here's what I want, here are the acceptance criteria, and here's how I know it's working. How they write the function, whether it's OEvent or OvantSquared, that can all be improved and that's usually not going to be a critical issue. But what does matter is that the team can take ownership of design of what they're going to release and then be accountable to building things the way that they're comfortable maintaining and owning. So it doesn't matter who writes the line of code, it's got to be something that the team wants to own and is comfortable maintaining and can work with.

The AI Layoff Headlines: What's Really Going On?

You just said something that I think about a night and day maintaining and owning, and there's a lot of, you see it all the time. AI causes job layoffs and engineers are laid off, but at the end of the day, AI cannot critically think the way you're talking about it. It cannot own and maintain. What's your opinion on all of the headlines and are those real? Is it AI actually automated or where are we in this lifecycle?

Russell (07:35): So what's interesting is that if you use Claude, I'll always say thinking LLMs, really they're just a prediction sheet. They're choosing the best Next Word based on everything it's seen before. And part of it now is that it's choosing to make it look like it's thinking, but it's not really thinking. Where it can be very useful and very effective is research. Even putting together a scaffolding of a product requirements document or suggesting an architecture for a feature. It can move really fast. It can save you a lot of time, but LLMs are never going to think. They're never going to reason and they cannot replace creative roles. Now, I am an optimist. I believe we will achieve AGI. I don't think it's going to be LLM. So what's happening now is you see companies, large companies, not going to name-drop anyone, but we see these headlines of 10,000 jobs reduced at company X because of AI.

And there's two things that I see that I believe are happening. One is these are public companies and every quarter they got to show numbers and those numbers have to look good and then their numbers go up and everyone's happy when the numbers go up. So you want the top line to go down or go up, you want the expenses to go down. When you invest $200 billion into AI, how do you not make your expenses go up by $200 billion? For a company of that size, you just have to get rid of people and say it's because of AI and everyone gives each other high fives and they're like, yes, we automated the jobs. But the reality is I believe they're largely just managing optics. It's, hey, we're holding on by a thread with a skeleton crew and we're claiming AI gave us efficiencies, but really we invested into AI and now we have to cover that expense by reducing other expenses.

The other thing that I've seen that, and this is somewhat of a conspiracy, but I see it as somewhat of a group negotiation tool. And what I mean by that is that if enough large companies come out and say, "Hey, you, Russell, the engineer, your job is becoming irrelevant because AI can do it." And I hear that over and over and over again, and then I get laid off and then they come back to me a year later and they're like, "Ah, I guess we can use you. Here's 60% of your prior comp, take it or leave it." And in my head, I believe that my job has become less relevant. Perhaps there's an angle where if we lay off enough people, you might be able to reduce the expectations of engineering salaries. And obviously the public companies would love that just like they would love having less people.

But overall, a lot of what we're seeing with these AI investments is coming from the top down. It's not engineers on the ground saying, "Hey, this thing makes me so much better. We have to use it." It's the CFOs, the CEOs saying, "Hey, we could save billions of dollars. Let's hope this works.

The 90% Problem: Where AI Helps and Where It Falls Apart

So talking about engineers and what they're feeling and what they're seeing, is there anything you are comfortable with removing from them so that they can do the other parts better or do more in your company? Where is that shift coming?

Russell (11:16): So again, the way I like to think of AI, and it's probably been said too many times, it's a typing speed enhancer. So as long as I stay engaged and I'm directing where to go, as far as that zero to 90%, that can get a lot faster. But we also at Kingsman, we see this all the time. The last 10% is the hardest. The last 10% can take 50% of the timeline. You can get 90% of the way there with AI, just like you can get 90% of the way there by hiring inexperienced offshore engineers often. It can get you somewhere, but actually finishing it to completion in the way that the client likes, that the user likes, that the business is happy maintaining, that part still needs, we call it the adult in the room. There needs to be the babysitter, the owner, the person that says, yes, I am not going to sleep until this works correctly.

And not only am I saying that, I know how to back that up because I have the experience. So as far as decisions that I'll remove, it's like, cool. Which order you write CSS class names in? I couldn't care less. I already couldn't care less. There's stuff that I was already very comfortable delegating to juniors and the skilled engineers. The architecture, how are we going to handle 50,000 TPS with HIPAA compliance? I'm not going to ask Claude how to ... It might give me some ideas. It might help me pick a better encryption algorithm, but the overall structure, the architecture, how it's going to work, not only do I need to choose that because I don't believe it can, even if it can choose it for me and I get lazy and just let it happen, I've seen this happen to myself and other engineers.

It's like when you just try to vibe code and you say, "Hey, Claude, build this app and it one shots it for you or build this feature even." And I tried this a year and a half ago. I was like, "Wow, codex is amazing. I have to use this." And I let it build a feature and I review it and I was like, "Yeah, this looks mostly right." And then the next feature I'm working on, I have to touch some code that had my name on it that I wrote and I don't understand it. And I'm like, I don't remember writing that. And it's kind of like your brain just kind of turns to mush. It's very interesting, the sensation of being responsible for code that you didn't write and wouldn't write. So again, I think I'm repeating myself a little bit, but it's very important to stay engaged and to be responsible for the outputs regardless of who's writing the code.cIt's just a seniority thing that every engineer learns to do.

Why So Many AI Projects Are Failing Right Now

I want to dig into something you said earlier when you said AGI is coming, but not through LLMs. What do you mean by that?

Russell (14:35): When I think of an LLM, and I try to simplify it in my mind, because let's be honest, it's a very complex machine and it's trying to actually understand every detail of how it works. I don't think anyone actually understands it. It's more of a black box.

But I mean, I started doing AI in college in the early 2000s, and when we were learning about machine learning back then, you think of it as -- there was a great book I read actually in my early 20s called On Intelligence, and it was talking about the future of AI and how this works. And the core principles haven't changed. Really, you have a bunch of vectors, which are points and space of time and then weights that basically like, here's the path from A to B. And we've gotten better and better at training these paths so the AI can say stuff that sounds reasonable. But at the end of the day, all we're doing is we're lighting up a path that says, after I say the word hello, I should say the person's name, do I know the person's name or should I say hello, what's your name?

We've gotten very good at training a machine to do that, but that's all it's doing. It's repeating what has already been seen. AGI, advanced general intelligence means a machine that can actually think. It's not a prediction machine where actually what humans can do, there's two parts to the way the brain works. There's the whole memory and pattern recognition and repeating the thing, and that's what AI can do. What it can't do is wonder and be curious and be creative. And that is where I don't know what it's going to take to get there. I don't know if we're five years away, I don't know if we're 200 years away. Either way, I believe it will come. It's not ChatGPT. It never will be. Maybe OpenAI will create it, but it's not going to be another GPT model. It's going to be something completely different.

Claude vs. ChatGPT And More Fun Topics

Who do you think's winning right now, ChatGPT or Claude, OpenAI?

Russell (16:33): I think that Claude is winning, and I believe it's because they focused on the enterprise first. All the tools like talking to ChatGPT, making videos with Sora, all that stuff, it's fun and people get excited about it, but at the end of the day, it's kind of gimmicky. And I do believe there's going to be some amazing consumer technology that comes out. For example, I can't wait to wear a pin that transcribes every conversation I have and then schedules all the follow-ups and works as the extreme executive assistant for me where I can just stop thinking about what do I need to do next and something's managing that for me. That will be amazing in my personal life and in my business life. But even if you look at, what is it, Suna, the music generation tool. It's really cool. When it first came out, I was making all kinds of songs about my dog, about Kingsman, about my wife.

But now if you look at what they're doing, they're trying to shift entirely towards marketing to actual musicians and record studios and things like that because that's who can actually benefit from it. And at the end of the day to make money and charging users $20 a month for ChatGPT when it's subsidized and costing you and your investors $200 a month, there is no light at the end of the tunnel. But then as soon as you switch to the enterprise edition and, wait, I went over my credits in 20 minutes and now I have to start paying $200 a day extra or $2,000 a day extra or whatever it is they want me to pay, now there's actually a path to profitability and no business, regardless of how great they could be, regardless of their potential, without profit, they're going to fail. They need the ability to sustain their own operations.

And that's why I believe Anthropic's doing better because they've targeted businesses from day one, and that's what's going to give them the runway to innovate more because they don't have to keep asking investors for more money once they can become profitable.

What non-work-related things do you use AI for? Is there anything in your life that you're like, "Man, I do this everyday?"

Russell (19:10): Not every day, but one of my hobbies is I'm a collector. I collect everything, shoes, watches, and my biggest passion ever since I was a kid is Pokemon cards. And I use a lot of deep research into market trends to try to decide which cards and sets to buy and hold. And I'm regularly using that. I haven't gone so far as setting up an open clock to start bidding on eBay for me, but I'm on the verge of doing that because I'm looking to buy things at what I believe is a good price to hold for a decade or more. And it's a lot of fun just having something that can crunch that much number, that many numbers and that much history as.

Would you trust an agent or OpenClaw to help with some sort of business like a collector's business? Or what's your view on OpenClaw in general, maybe inside of your company?

Russell (20:38): Oh, inside the company, it's on its own isolated network. Zero trust. And again, I'm experimenting with these things, but step one is to create its own network. It cannot talk to anything else at our office. No, I don't trust it.

So I've been experimenting with using it for marketing, giving it campaigns, giving it very low limit credit cards and just seeing what it can do. Hey, if I give it a project like, "Hey, we want to target customers that bid X and I don't want to give away what we're doing. Let's see what kind of traction we can get if it's commenting on Reddit on my behalf." And then cool, you have $1,000. Do you want to spend it on Google or LinkedIn and what keywords? And letting it play around a little bit to see what kind of results we can get. So far, and again, I'm very, very, very at the ... I'm just breaching this. It's new for me. What I've seen is it's very cool to watch it work. Haven't seen any good results yet, but at the same time, I know people that are non-technical people that are a little bit more reckless, I guess, where it's like, yeah, I have seven Mac minis running in my shed on my network and they're doing all my research and I've replaced six employees on my team with them and they do everything. And I'm just like ...

And they've also opened seven mortgages accidentally. 

Russell (21:58): Yeah. Exactly. It's like it has access to all my bank accounts, everything.

Yeah, it can email me, but I think there's a lot to be said there. Again, especially when we're talking about research, it is very, very good at researching and it can really speed you up. What it can't do is that creative side of, hey, even just the thesis of, hey, I think a great angle if we target X, what are some ways we can do that? I think LLM's idea of a great thesis is an average thesis because it's been trained on everyone on the internet saying this is a great idea, whereas the people that truly have great ideas, there's so few of them, it's not enough data that it's going to train the model to think like that.

One last question. If you were to leave one piece of advice for CEOs or other CTOs who are trying to change their org, what are they getting wrong with AI today that you think that whether it's them reading the headlines incorrectly or to your point at the beginning where people are just getting advice because these people are telling them to do it because they need more money, is there anything you would leave somebody with?

Russell (23:16): I would argue not to think of it as a cost savings. At the end of the day, I believe best case scenario right now, your investment is going to increase. Unless you've already overhired and you're going to reduce some roles anyway, I don't think we're in a place where you can just start laying people off because you have AI, but you can increase some metrics, whether that's quality, speed. There are ways to get more out of the team that you have, but I would just say start conservative. Try to get a three or a 5% boost before assuming you can get 50 and go from there and learn what works. And when the engineers tell you, here's what's happening, listen to them. Don't say, "Hey, here's what I heard from the CEO of Nvidia. Listen to them. Did it actually help you?" And I will tell you that I personally, I've spent a lot of time over the past, I want to say three or four years now using Agentic workflows and trying to get AI to help me.

And I am still on the fence of personally, am I getting more out of it than if I just sat down and write the code myself? I'm on the fence. I can't tell you for certain whether or not it's helping me. And I have every incentive in the world for it to work for me and to find a way to make it work. And I'm still a little hesitant. I don't see a reason for me to tell my juniors to start using ClaudeCode instead of doing it themselves. I think that's the equivalent of giving someone a nail gun before they learn how to use a hammer and saying, "Frame my house." It's just not.

Russell, where can people find you? Shameless plugs are completely okay. 

Russell (25:17):  Well, we are usually writing code and not posting on social media too much. We are a boutique agency, so we don't have a massive marketing firm putting everything on social media and on socials, but our website is kingsmandv.com, DV for Digital Ventures. You can find us on LinkedIn as Kingsman Digital Ventures. I'm on there. My business partner, Brandon's on there. He's super friendly, always available and ready to talk. And then any kind of tactical questions or ideas you have, I'm always happy to chop it up with as well. 

About Russell Anthony, the CTO at Kingsman Digital Ventures

Russell Anthony, is the CTO of Kingsman Digital Ventures, an agency that partners with businesses to build tailored web and mobile applications, integrate enterprise IT systems, and design data analytics platforms, with a strong focus on highly regulated industries like medical aesthetics

Interviewed by: Sergei Dubograev, the SVP of Development at Clutch

About the Author

Sergei Dubograev VP of Development at Clutch

Sergei Dubograev is the VP of Development at Clutch, the leading global marketplace of B2B service providers. With a proven track record of driving innovation and delivering exceptional results, Sergei has led cross-functional teams to create and launch products successfully in-market. His strategic mindset, coupled with his ability to identify industry trends and customer needs, has consistently propelled organizations to new heights of success. At Clutch, he oversees the strategic planning, execution, and optimization of the company's development portfolio, ensuring alignment with business objectives. 

See full profile