Value Driven Data Science

[00:00:00] Dr Genevieve Hayes: Hello and welcome to Value Driven Data Science, where data professionals become strategic experts. I'm your host, Dr. Genevieve Hayes, and today I'm joined by Tim Dietrich. Tim is an independent software developer with over 25 years experience building business software for organizations ranging from startups to Fortune 50 companies, including Siemens and the Library of Congress.
[00:00:29] Recently he has become known for building a virtual team of AI specialists that allows him to operate with the output and breadth of a small firm while remaining a team of one. In this episode, we'll explore how data professionals can use AI agents to dramatically expand what they can deliver alone, and position themselves as more valuable rather than replaceable.
[00:00:57] Tim, welcome to the show.
[00:00:59] Tim Dietrich: Thank you for having me. It's an honor.
[00:01:02] Dr Genevieve Hayes: Each week, it seems like there's yet another announcement of technical workers losing their jobs to AI In Australia. For example, tech Giant, Atlassian recently laid off 1,600 workers, 10% of their global workforce to steer more spending into ai. Now granted, not every AI related job loss is necessarily as it first appears.
[00:01:28] Some experts point to AI washing. That is companies using AI as cover for restructuring decisions they would've made anyway. But regardless of the reason, the reality is that AI has fundamentally changed the nature of white collar work. And for technical workers, including data professionals, it raises the uncomfortable question.
[00:01:50] If AI can do what you do, why would anyone need you? But as with every other technological advancement before it, the solution isn't to compete with ai, but rather to harness AI as a force multiplier, expanding what you can deliver, taking on work you couldn't have touched before, and making yourself not just relevant, but indispensable and as technical professionals experienced in working with cutting edge software tools.
[00:02:18] We are better placed than most to do this. And Tim, that's exactly what you've done in your own work as a software developer, you could have tried to go head to head against AI coding tools , in which case we probably wouldn't be having this conversation. Instead, you've done something far more interesting.
[00:02:37] You've used Clare Code to build yourself an entire team of world class specialists available whenever you need them. Now in preparation for this episode, you told me that up until about nine months ago you were, what you described as the world's most negative person when it comes to ai.
[00:03:01] What caused this about face and how did you come to build your virtual AI team?
[00:03:08] Tim Dietrich: Yeah, that's a really good question. Like I had said to you before we started recording that, had we talked a year ago, first of all, you wouldn't have wanted to talk to me because I would have nothing to do with ai, but I really was just, I think, one of the most negative people about it.
[00:03:23] I was tired of hearing about it, quite honestly. I still am. But I thought it was overhyped, it was just too much. And I think a lot of that was kinda driven by fear. I thought that. If AI is as good as people were making it out to be that at some point it would, take my job, so to speak.
[00:03:44] And I kept thinking as long as I just make it to retirement age, I should be fine. 'cause I'm thinking, it's gonna be five years, 10 years out before AI can really code well. And boy was I wrong about that? So I was very negative about it, it, it just kept hearing about it constantly.
[00:04:04] And so it was getting close to about a year ago that I decided one day, okay, I'm gonna look into this. And I had used AI on and off to do little things but wasn't super impressed by it. And it really wasn't until I started to look at it through the eyes, I think of a programmer and start using AI through.
[00:04:25] The APIs that some of the providers make available that I realized that, there was something there and I still wasn't really sure, what I wanted to try to do with it. But I started experimenting with it and tried to find ways that I could use AI with the programming work that I do, which is I do a lot of work with the NetSuite ERP system.
[00:04:48] What I found and what I was really impressed by was its ability to do financial analysis. I couldn't believe how good it was at that point. And it's only gotten better since, so that was really, me just saying, okay I really should look into this. To be honest though, at that point I still thought as a developer I had time, it couldn't possibly do the kind of work that I do in terms of writing code.
[00:05:15] And it wasn't really until, I think I would say September, October of last year. I know Claude Code had been out for a while before that time, but it seemed like there was something that sort of snapped at around that time. And I think the whole. IT community sensed it. That cloud code had made this jump to being really good at software development.
[00:05:42] It wasn't perfect and it still isn't, but. It was so good that it felt okay, like that window of opportunity that I thought I had, if I could just eke out another five years in my career, I'll be fine. That I, no way. So that, I guess that's a long answer to that question but that's it.
[00:06:01] And it's been just crazy ever since.
[00:06:03] Dr Genevieve Hayes: You've built yourself a team of virtual employees. Why not just hire a human team instead?
[00:06:11] Tim Dietrich: I've always been a solopreneur one man shop, I don't like to manage people. I don't like to be managed. I'm the world's worst. Employee. But I've often wondered, if I were to take the real leap and make a quote unquote real company with employees would I be more successful?
[00:06:31] I think financially, maybe, but i'd love the work I do, and I don't think I would want to necessarily have to just manage people all day. So I've always avoided having employees. I guess it's a long way of saying that. With AI in building the team it never really came down to, should I build a virtual team or should I hire humans?
[00:06:53] I didn't have the funds to hire a data scientist or a marketing director. I'm one guy with a MacBook, I don't have buckets of money in land. So I guess what I'm getting at there is it wasn't really a choice between should I hire humans or build a team. Being able to use AI to tap into expertise and talent it's magical.
[00:07:18] For me it's oh, I can afford now to hire. Anyone that I want at any time, and I don't have to pay them. They work all the time. If I wake up at four o'clock in the morning and I need something done, I don't have to call somebody or page them or pay them over time but it's not about the money really.
[00:07:37] It's just about the fact that now I have all this expertise, all this talent and the ability to do things that I could never do before. So it makes the work that I had been doing even better, I think, and it expands what I can offer clients.
[00:07:54] Dr Genevieve Hayes: One thing I find, I also have a solo business. One of the things that AI really helps me with is for very small one. Off jobs because you can't have a full-time employee for something that's just one afternoon, once every three years. It just doesn't make sense. And even the effort involved in engaging someone for that might be significant because you've gotta find the person and then you've gotta book time and all that.
[00:08:23] So just being able to have an AI to work with makes more sense.
[00:08:29] Tim Dietrich: Yeah, it does, and I think depending on how you use ai. You can spin up, say an artist, graphic artist to work on something for you today. And then when you're done with it it's ephemeral. You're done.
[00:08:42] You've got the output. Maybe it's a new logo or something like that. And then that's it. Where I think it gets really interesting is the ability to have that virtual graphic artist, be available so that in the future when you need them, they remember, there's, history, memory of what they did for you before.
[00:09:01] And so if you say to them, the logo that was created was great. It looks good on screen, but I need to be able to print it. You don't have to go all the way back to the beginning. Teach it, about your business so that's not necessarily the best example, but I think you know what I mean, like there's some continuity there, there's some history that you can build up over time.
[00:09:22] And that's one of the things I think by the way, is like what made Claude code specifically just such a turning point. Just the fact that you could use it over an extended period of time and it feels like you are building. History with that agent.
[00:09:40] Dr Genevieve Hayes: Yeah, I find that with the Claude Chatbot, because I've been using it for so long, it understands the way. I think about various things, so it gives me answers with that context.
[00:09:52] Tim Dietrich: where I think that gets challenging though, is when you're trying to use AI on behalf of a client, there's times where if you're not careful, it will look at. The way that you've used it in the past and that memory will seep into the way it's trying to do something for you on behalf of the client.
[00:10:12] So I guess what I'm saying there is that there are times where it is good to be able to say, I need a fresh agent,
[00:10:19] Dr Genevieve Hayes: yeah, that's when I used the incognito windows.
[00:10:22] Tim Dietrich: Yeah, there you go. I don't want you to know anything about my past.
[00:10:26] Dr Genevieve Hayes: Yeah.
[00:10:27] Tim Dietrich: Yeah, no, I get it. It really is like having people that are working for you.
[00:10:32] Dr Genevieve Hayes: So you've got over a hundred virtual specialists in your team and counting. What was employee number one, and how did you come to create that one?
[00:10:42] Tim Dietrich: that's a good question. It probably wasn't a technical agent. I don't remember exactly which one it was. It probably was something like a business consultant or something like that. Because I think when I first started creating those agents, and by the way, it wasn't like I just sat down one day and I thought, I'm gonna write prompts or create agents.
[00:11:03] It just grew out of. The work I was doing in creating prompts to do a very specific thing. But that's how I had been using AI in the past, I think like most people do, where you go into Claude and you say, I need help with this.
[00:11:18] And you have that iterative sort of conversation with it, and eventually. You come out with an artifact, a document or something, that was your goal all along. And I think the way that I grew into building these bots was I would go through all that. And by the way, this is a good tip.
[00:11:38] Regardless, when I would come out with exactly what I wanted, I would sometimes ask Claude if I wanted to do this again, what would the prompt be to give you? To make this a faster process and it will write a prompt for you. So that was interesting because you could start to get repeatable prompts that Claude itself gave you.
[00:11:58] It doesn't get any better than that. And then over time, I realized that a lot of the things I was asking Claude to do were related. Some of them were technical, some were business oriented, and I remember, the light bulb moment of what if I just. Had Claude, act like that person that could do all those things.
[00:12:19] And what would they be like, what would their skills be, their experience, you know? And that's where the whole idea of building the bots came from. And then I just got better and better at it. And there may be a couple of days a week where I don't create a new one.
[00:12:35] But. Those are few and far between. Like when I need something now, that's where my mind goes. Okay, what would be the perfect bot for this?
[00:12:43] Dr Genevieve Hayes: Do you have a bot that determines which bot you should assign a particular task to?
[00:12:49] Tim Dietrich: I do. I actually have a process that I go through where when the new bot is available, I have the sort of onboarding process. I feel like I have an HR department. But really it is, it's something that sort of registers it so that when I do go in and say, who's the best agent for this? It can determine which one that I should use.
[00:13:11] Now, I also have, and I wrote about this on my blog at some point. Like a team that's specifically for web development where I have a project manager and then a developer and I think a designer and a u bot, and I only talk in that case with the project manager.
[00:13:29] And it, does it on its own to figure out, okay, I need the developer to be doing this, the designer to do that. And it manages that whole thing. It knows well, I'm still waiting on this from the developer. I'm waiting on that from the designer. And I can at any point ask the project manager like, where are we?
[00:13:51] And it will tell me, I'm waiting to hear back from the developer on this. But it'll figure out where it can run things in parallel. So it'll say to me that I'm waiting for the developer on this. The designer's working on this. So it's not like it's a sequential process that it has to go through.
[00:14:10] It's like having a real team. And the best part is I only have to talk to the project manager.
[00:14:15] Dr Genevieve Hayes: So how do you build one of these agents from scratch?
[00:14:19] Tim Dietrich: That's the magic, right? That's the secret sauce. When I first started doing it I did it the way that I described earlier where. I stumbled into asking Claude, what would the perfect person be for this? And then just refining that over time.
[00:14:36] So as boy, that's a really vague way answering it. But yeah, I just stumbled through it and got better and better at it. Now I've got so many of the bots, like you hit on it. Sometimes I don't know which one to use. Even after asking Claude itself which of my agents is the best one for this?
[00:14:55] And so I'll sometimes have to say to it, can you tell me what the difference is between these two? And by the way, another tip, you can take these bots. And this is also just a good, prompt engineering tip in general is you can have multiple instances of it and then compare their output.
[00:15:14] For example, if you had a prompt that was gonna evaluate something, like you've got a contract that you need somebody to get eyes on it and tell you, where are the problems with the contract?
[00:15:24] You can feed it to Claude. You can feed it to Jet GPT, get their combined results back, see where they're different. You can even ask them to grade each other's work. You can even use different models within the Anthropic suite, models but you can do the same thing, with either a prompt or with one of these agents.
[00:15:43] Dr Genevieve Hayes: Have you actually thought about building a bot to create more bots?
[00:15:47] Tim Dietrich: I actually have I tried that at one point and I wasn't happy with it, and then I was surprised at how unhappy I was. And I think that's because when I do create the bots, a lot of time goes into it. Like I'm trying to figure out how to encode that expertise into it. It's more than that though.
[00:16:09] And I've wrote about this recently on my blog. It's not just the expertise, it's trying to encode the way that kind of professional thinks. And that can be very difficult, but when you get it right, the results are amazing. Yeah, I think that when I had created a bot to create other bots, which is a frightening idea by the way.
[00:16:33] Like at what point does it just keep going? I think that it just didn't have the human qualities that knew what to bake into the bots. It's a really weird answer I gave you there, but I think, right? There's some human element to it.
[00:16:48] Dr Genevieve Hayes: And then we descend to having an army full of Agent Smiths from the Matrix.
[00:16:53] Tim Dietrich: Yeah, this agent creates that agent, which it can totally do that would probably just like my hard drive would fill up with the code for the, yeah. Agent Smith, number 1, 2, 3. Yeah.
[00:17:06] Dr Genevieve Hayes: so how do you decide what work to delegate to an agent versus what to keep for yourself?
[00:17:11] Tim Dietrich: So that's another thing I think is important. You hear a lot about the fact that people are just letting the agents just run wild like more autonomy than I think they should. I always try to keep myself somewhere in the middle, that human in the middle phrase that you sometimes hear.
[00:17:30] In the AI world where I'm not quite at a point where I feel comfortable enough delegating work that I'm not gonna review myself. So at some point, even if I decide I'm gonna delegate work to AI in general, I know I'm still gonna get eyes on it later, and I think the decision about whether I delegate something or not it's getting to the point now where there's not much, I don't delegate.
[00:17:58] There's still some work that I do that I cannot use AI on certain technical things that there's just no easy way for me to get clawed to see what I'm seeing. I think the important takeaway there is that even after all that I've done with it, I still feel like.
[00:18:15] I'm required, I'm still a company of one. When I get to a point where I'm not needed, and it's just this AI driven company where Tim doesn't matter anymore, that's the moment where i'm checking out. I'm done because I won't be needed.
[00:18:31] Dr Genevieve Hayes: You've mentioned several times the AI data scientist in your virtual team. What prompted that and what does that AI agent actually do?
[00:18:42] Tim Dietrich: So it's interesting, a lot of the work I do, again, is with NetSuite in the ERP space and what I've seen over time is that the longer that a business is on. Any system really, but it seems to be, especially with ERP systems, that the quality of the data, just over time, it starts to degrade. And the reason I think for that is that you end up with all these processes and subsystems that don't do things the way that the ERP used to manage it.
[00:19:11] So there's inconsistencies, there's duplicate data. And so I got to thinking about how could I help my clients that are running into that problem? There was one project I was working on in particular where that was becoming an issue. And so I thought, if I had to hire somebody to do that work, and again, I didn't have the budget to actually hire a human, who would it be?
[00:19:35] And I think I even used Claude to help me figure that out. And it's I think you need a data scientist. So I created a data scientist. And pointed it at what I was seeing. And it came back and it gave me advice on how to clean things up and how to figure out where those inconsistencies were, what was causing them, that sort of thing.
[00:19:57] And I think in an email to you, I told you that at one point I actually asked the data scientists what techniques did you use to give me that advice? I sent it to you, you maybe understood some of it for me. I'm not a data science. It was like, okay, that's pretty cool, and it worked.
[00:20:16] More recently I wrote about an issue Simpsons paradox. So I was running into that. I'm not gonna go into the definition of it, 'cause your listeners probably know what it is too, but I didn't, I ran into that while working on a project and I eventually figured out what the problem was, but I didn't have a name for it.
[00:20:38] And so we, client work got it done, moved on. And then later I thought I wonder if I could ask my data scientists what was that? And it knew right away it was like, oh, that's. Simpson's Paradox and here's what it is. And it was really interesting. Ended up again, writing a blog post about it, and it's just one of those things where now it has me wondering like, how many other times have I run into that and not known it.
[00:21:00] So I even had it help me figure out , what can I do with the prompts that I am creating to avoid that in the future. And it's, in the blog post, so if you're curious, just check it out.
[00:21:12] Dr Genevieve Hayes: When you sent me that email that you mentioned previously with all the qualifications of your data scientist, what struck me was, wow, it can. Do an awful lot of stuff that took me an awful lot of time to learn, I gotta say that like you, this does make me a little bit nervous about the prospect of an AI agent taking my job in the future.
[00:21:38] And I'm sure many of our listeners are probably thinking the same right now. How can someone like me or our listeners position ourselves so that we're seen as being more valuable than your AI data scientist, rather than replaceable by it?
[00:21:56] Tim Dietrich: Okay. Yeah, this is, this will take some time to unpack, but a friend of mine who's also in the NetSuite space, we used to do a podcast, and he used to say to me. It's not AI that's going to take your job. It's another developer that's using AI that's going to someday take your job and probably five or six other developers jobs with it.
[00:22:18] And he was right for sure. I feel just again, so empowered with what I'm able to do. But having said that, I do think that at some point. Even a developer that's using ai, he's gonna be in the cross hairs, right? At some point you really will, regardless of whether you have any development experience.
[00:22:40] And by the way, I think we're really close to being there. Can just sit down and say, I need an app that does this, whatever this is, and okay, give me a minute. Here it is. So the window of opportunity, I think, for being an expert and using AI to have that productivity boost, that force multiplier that you refer to in the intro.
[00:23:03] I think even the opportunity there the timeframe on that is probably pretty small.
[00:23:09] But having said all that, I do think that right now what's made my business stronger is the fact that I can do more than I ever could before and. I'm able to deliver a better product, a better service for what I have always done. So when I write code now with Claude I can have it creating documentation alongside the creation of the code.
[00:23:35] I can have it create documents on how to install the app, what to look for, a user guide I can build in, help systems and things like that. Things that would've been. So difficult for me to provide to clients in the past, just simply because I would've had to charge more. It would've taken longer, and I probably never would've hit the level of quality that Claude can do in creating those documents or that type of help system.
[00:24:03] And I've been doing this for a really long time, so I thought I was pretty good at it. Now, I like wonder, right? So I don't know if that applies. To the data science professions, but I'm guessing that it probably does, in addition to doing the work that you were hired to do, there are bonuses that you can provide whether you're charging for them or not.
[00:24:26] Yeah, and then there's just the other ways I'm using AI in my business I'll give you a great example. I had a meeting with a client today around one o'clock my time. I got revised requirements on a project and was able with Claude to take the new requirements, really have it, help me think through them, and then create the revised proposal, and it took about 30 minutes, and in the past I would've been all over the place.
[00:24:54] I'll get that to you early next week
[00:24:57] Dr Genevieve Hayes: so if a data scientist wanted to start building their own virtual empire tomorrow, where should they begin?
[00:25:04] Tim Dietrich: I would start by building one agent. I would pick something that isn't part of your core too, like not the work you love to do. A good example might be create an agent that's gonna help you give a quote on a project that. A client has asked you for a quote on a proposal and that's how I would do it.
[00:25:24] I would probably start by giving it a proposal that you have really created in the past so that it has an idea of your style. Maybe even ask it are there things I should be including in the proposal that I'm not, and then I would give it a real task okay, based on what you saw from my previous proposal.
[00:25:44] Here's a project that I need to quote on, create one for me, and then just go through that process I mentioned earlier where at some point you say to it, if I wanted you to do this for me again in the future, how would I do it? And it's more than just giving you that prompt
[00:25:59] at some point you have built up prompts to do things like create a quote or evaluate a contract, and at some point you can meld all those together to just be your. I don't know what to call that person. , By the way, you can use Claude to help you figure out what the title of these things are, but like your business manager, right?
[00:26:18] So now your business manager can help you with quotes, can help you with reviewing contracts, and then at some point you might think I'm doing enough of that kind of legal type work where maybe I need a lawyer, and that gets into the whole. You gotta be careful with, how you're using these things.
[00:26:38] But at the same time, yeah, I have a legal assistant, if you will. And again, I couldn't afford that. I couldn't afford to hire a lawyer for one hour, a real one, yeah. So going back to your question, I would build one. Get comfortable with it. Learn the process and then think of okay, what's another one I can do?
[00:26:58] And eventually at some point I think you will be tempted to say, okay, how could I create one that actually does the work that I do? And for me, like that one was, a sweet script developer, which is the language that we use to customize NetSuite. I think, that was one of the first technical ones that I built, and
[00:27:16] it was insane. I really thought there's no way it's gonna write code that I could feel good about. And it did.
[00:27:23] Dr Genevieve Hayes: For listeners who wanna get in contact with you, Tim, what can they do?
[00:27:27] Tim Dietrich: So I have my own website and my blog is on there. It's at timdietrich.me. Me and I also am pretty active on LinkedIn. So if you look me up on LinkedIn, that's. A good way to follow along with what I'm doing. Sometimes I'll post stuff to LinkedIn and eventually post it to my blog.
[00:27:46] Sometimes I do it the other way, but that's really the only way to get me off social, is to look at LinkedIn.
[00:27:53] Dr Genevieve Hayes: Okay. That's it for today's episode of Value Driven Data Science. But if you want more from Tim next week, you can catch our Value Boost episode where we explore how to build your AI advantage without becoming AI dependent. And if you found today's episode useful and think others could benefit, please leave us a rating and review on your favorite podcast platform.
[00:28:16] That way we'll be able to reach more data scientists. Just like you. Thanks for joining us today, Tim,
[00:28:23] Tim Dietrich: Thanks for having me. This was a lot of fun.
[00:28:26] Dr Genevieve Hayes: and for those in the audience. Thanks for listening. I'm Dr. Genevieve Hayes, and this has been Value-Driven Data Science.