Innovating with AI, a Journey from CTO to Entrepreneur

Dave:
Welcome to another episode of Emerging Tech on Do the Woo. I’m your host, Dave Lockie. With me today is Aaron Edwards from docsbot.ai. Hi, how are you doing?

Aaron:
Hey, how are you doing, David?

Dave:
It’s been a while, isn’t it? Yeah, it has. We had some collaboration back with the WAPUU/NFT stuff and the Web3 WP, which I guess is like three, four years ago now, something like that.

Aaron:
Yeah, I think so. Yeah, probably about three years.

Dave:
Wow. Time’s flown. And I guess we’ll go back and do kind of a quick intro to you, but, so we’d crossed paths and collaborated before all this ChatGPT stuff really came into the mainstream, and you obviously followed that advice of if you’re in crypto, pivot to AI. So I’m curious to hear that part of the story, but maybe you could just give us kind of an intro to you and how you came to be hit by.

Aaron:
Yeah, sure. Well, my name is Aaron Edwards, as you said, and I’ve been in kind of the WordPress world for, gosh, that’s like 2008, 2007, 2008, back in WordPress 2.6, I think the first one that looked a little bit pretty. And I kind of started there just building my own multi-site. Back then it was called WordPress MU, Multi-user, now multi-site, and was hosting my own there and started learning how to do plugin development and things like that. And then became a developer at WPMU Dev for a little while, and eventually rose up to CTO and served there for, gosh, a long time, like 13 years or something as CTO at WPMU Dev. So lots of experience in the WordPress world and building things, but ultimately my favorite thing is always building with new technology, whatever the new thing is, I love figuring it out, how to do practical things with it, how to transform that into products that users love.

And so even Web3, I’m still fascinated by Web3. I know kind of the shine and luster and stuff is off it right now at least, but I still believe in it. But yeah, learning how to do smart contract development and things like that, and experimenting with doing projects that could use Web3 with WordPress, and that’s where we first met. Yeah, so as AI stuff was starting to go, I started actually learning about that back when it was deep learning and neural networks were the first things and learning how to do stuff with that. And then this whole generative AI thing, as that came out, I was like, I’m not missing a new technology wave. I’m going to be at the forefront and be experimenting, be trying the new stuff, seeing what I can build with it. And that’s what I’ve been doing. I built a bunch of experiments and one of them especially took off, and that’s Docsbot.ai. And now that’s my full-time job as an entrepreneur running that and growing that and building new features and things for that.

Dave:
Thank you for that. And I love that intersection. I think there’s this thing that people in technology, this behavior that I find weird, which is that if you’re working in technology, then really the onus is on you to keep up with what technology is doing. The nature of working in technology is working with change. And so when I see people who are either resistant to change or seem entrenched in the technology world as it is today, it always confuses me. And so yeah, I definitely share your passion of trying to figure out what’s coming and marrying that to whatever technology one cares about, which I think in our case, it’s both WordPress. That’s what it feels like code for us.

Aaron:
Yeah, it is funny how people can be so pessimistic. It’s kind of like the thing they say—the best music for any person is what they listened to in high school. It doesn’t change. I think it’s kind of the same with technology. Maybe it’s part of human nature; we just don’t change or are uncomfortable with change. But I always tell my kids this all the time—you need to learn how to learn. That’s the most important thing to survive in the world because things are always changing, and they’re going to be changing faster and faster. And the most important skill that you can learn is how to learn, how to constantly be learning. For me, I’m lucky that’s fun for me, and I love learning by getting my hands dirty and actually building things and trying out some of this new stuff and seeing what can be done. And so far, that’s worked out for me well in my career and things like that.

Dave:
So tell me, how many experiments roughly did you run? So presumably this is whilst you were still CTO-ing at WPMU Dev. How many experiments at least got to some sort of public release? Did you run before you hit Docsbot?

Aaron:
Yeah. Well, the first thing I built was I just love cloud technology. And so we built a lot of products that were for WordPress that were cloud-hosted at WPMU Dev, like our Smush image compression plugin or the Hub for managing lots of WordPress sites, all kinds of different things like that. So I got to scratch that itch there as CTO and, of course, managed hosting and things like that. The first kind of project that I built with a friend of mine, Josh Daley, was Infinite Uploads, which is a cloud storage plugin for WordPress. And I still actually own that. So it was kind of like the first foray. But as this whole AI movement came out, I was following it very closely. And so when GPT-3, they first started giving—it’s funny, people don’t remember this because it’s just such—it’s moving so fast now, but OpenAI was so restrictive about giving access to these models. They were so afraid of the damage it could do and things like that. So you had to know someone in Silicon Valley to get access to the API for GPT-3. It was hard.

And so I had actually looked at it. I was like, I’m going to make the first copywriting plugin for WordPress because I see where this is going. And back then it wasn’t like these models—they weren’t instruction-tuned like ChatGPT—you had to do some really careful prompt crafting because you would input a string and then it would continue that string. So it definitely took a lot more skill to use these APIs and to get good output. And so I was really trying for that and actually started building the plugin, all this stuff, but I could not get access. No matter who I pinged on Twitter, I could not get access. And then probably right when I did finally get access to the API, then Bertha came out. I don’t know if you’re familiar with that WordPress plugin for AI. Andrew, he’s a good guy, and Vito, I think, is co-founder of that.

Dave:
You weren’t the only one that saw this.

Aaron:
Yeah, now I’m glad that I’m not in that space. Gosh, it has gotten crowded since ChatGPT and stuff became so widespread to everyone. So my first thing was—and then the other thing—I was like, well, images. When DALL-E was first announced from OpenAI, and that was the same thing with them—they were very restrictive. They were afraid of the damage it could do. And so I was like, I’m just going to be ready. As soon as that becomes available, I want to be there to build a product with that or whatever.

And so I was actually following very closely the kind of secret Reddit societies that were following Stable Diffusion, which was kind of the first open-source image generative AI model, and following that really closely and things like that. And so I was there, I was ready the day the weights became available for that to be like, alright, I’m going to launch something. And so I created a WordPress plugin and first just shared a demo on Twitter, how you could generate images in the WordPress admin. It was actually a Gutenberg block that I created to do that. And that kind of went viral on Twitter. That was the first time generative AI image was available anywhere, really—not just WordPress. So that was exciting. So I went to ground for a week or two to turn that into a mini product and get the plugin pushed and released and stuff like that. So that was the start of Imagine.ai, which was kind of the first AI WordPress thing that I had built. Unfortunately, it turns out people don’t want to pay for that.

Dave:
And so this is still while you’re at WPMU Dev, and they were just cool about you going off and doing this sort of plugin? How did that dynamic work?

Aaron:
Yeah, at first it was hard. I mean, James Farmer, the owner there, he was super cool, super supportive. He has a bad-boy public personality from the past, but really he was always super good to me and supportive. Even as I was trying some of these things out, it’s like, yeah, go for it. You still have a place here. So I was able to experiment and iterate and ultimately, over the course of six months, eight different kind of AI projects just for fun. And it was the last one, Docsbot, that really took off and found a real good niche and use case for businesses.

Dave:
Strike rate, eight to one. It’s definitely better than a lot of the entrepreneurs that I see. But perhaps all of your historic product stuff gave you a bit of an edge here in terms of being able to get stuff to market.

Aaron:
Yeah, I think so. I mean, having so many years of experience building products already that do millions of dollars in revenue, I think that definitely helped in learning things. But still, I was a developer at heart, so actually running a business and growing a business and learning marketing and sales and all that stuff, the last few years have been a huge learning curve for me beyond just technology. So it’s been a challenge. I’ve enjoyed it, though.

Dave:
Let’s try and line up a bit of structure now. We kind of caught up to here. So I’d definitely like to give you the opportunity to introduce Docsbot, what it does, why people pay for it, what you hope to do with it down the road. I’d like to then talk about that journey that you just talked about there—from an employee, even if a senior one, through to the solopreneur culture and journey. And I’ve watched a few people go through that. I definitely find it exciting and compelling and daunting. So I’d love to hear what that was like for you. And then I’d like to hear what you are excited about next. Is it just total focus on Docsbot and what you want to do with that? Are you starting to look around, seeing other technologies come online, or is it all AI all the time for you? So yeah, I guess any of those in any order, but I think we’ve talked about Docsbot a bunch. That’d be good if you just introduce the product and what it does and why people should use it.

Aaron:
Yeah, it’s late 2022 when ChatGPT was released, and that was kind of mind-blowing to everyone. And a lot of people had the immediate thought, how do I use this for business? The first thing everyone does is, oh, write me a poem or a rap or something. Content creation, which is, I think, the most basic use case for this generative AI technology. But I started learning about something called RAG, which is retrieval-augmented generation. So the first thing people realized was that, yeah, they’ll ask it about what’s the best WordPress plugin or tell me about how do I change the setting in WooCommerce? And often ChatGPT had some knowledge in its training data, but it’s not grounded in any kind of fact. So these models will happily hallucinate—that’s what it’s called. They just confidently make up answers because they want to make you happy, because they’re not really thinking necessarily internally about what you’re really asking.

So this new technology was starting to—some people were experimenting with it—called retrieval-augmented generation. And basically what that does is, imagine that you paste—you take a document about how to change that WooCommerce setting, and you paste it into ChatGPT, and then you ask a question, how do I change this setting? And instead of reading the whole document, the AI is actually able to ground itself in that text that it just read to summarize or provide a custom or succinct answer that’s grounded in fact. And so I was thinking about, oh, this is really cool. This could be super useful for customer support or things like that for businesses. And of course, the whole chat interface—that was completely new. Everyone was excited about that, being able to chat back and forth in a conversation style with a bot. So I kind of experimented with some of that.

And I ended up building—it’s called ChatWP, and that’s still online. And basically, I just crawled all wordpress.org, all the support documentation and developer documentation. And then I imported that all into a vector database and kind of created a front end and launched that as just kind of a free tool for the community, so where they can ask questions about WordPress that are grounded in all the support documents. And then I had the forethought to put right underneath it, is this something you’re interested in for your business? And that just collected an email address for a waitlist, and I made a little survey that’s like, okay, here are some potential features. What would be the most important for you to have? So that kind of went—I wouldn’t say viral, but it was kind of like the first use case of WordPress and that kind of ChatGPT chat-type model stuff.

Dave:
So you went very lean.

Aaron:
Yeah.

Dave:
Here’s a prototype, tell me what you’d like. And then you went off and built it.

Aaron:
Yeah, exactly. So once I saw the popularity of that and that there’s a lot of—I mean, it was mainly WordPress business owners that were adding to the waitlist and things like that, that it’s like, alright, I think this is somewhat validated as maybe one of the first practical use cases for a B2B generative AI. And I spent probably three or four weeks building an MVP and launching it and productionizing it into something that—and that’s not easy. This technology was new. Even vector databases and things like that, they were not designed for that kind of scale where you have multiple libraries of training data for lots of different bots instead of just one giant one for your business. So definitely a lot of learning, a lot of back and forth, bug fixing, all that kind of stuff. But yeah, it kind of took off.

Aaron:
At first, it was slow. I had the waitlist, I sent it out to them, and a number of them were trying it out and signing up. And then I got really lucky. Somehow, some Japanese Twitter influencer found it—I don’t know, from my tweets or something—and tried it out. And he just recorded a video of creating a chatbot in Japanese and showing that this technology can work for languages other than English. And that went hugely viral in Japan, Japanese tech Twitter. And so I just started getting—I woke up in the morning, and I was like, what is going on? My servers were crashed, and I had a hundred support tickets in Japanese.

Dave:
Amazing.

Aaron:
I was like, oh man, pinch yourself, what is happening? So yeah, that week was—

Dave:
Make sure you’re not dreaming.

Aaron:
Exactly. So that week was crazy, just figuring out all that, getting things stable, adding the basics—because this was a raw MVP. There was no way to delete your data, no way to change your email address. There were all those annoying little things that you need in a product that I just held off on because I was trying to do as lean of an MVP as I could. Although I probably could have made it even leaner. I think each product I built, I learned more and more, like, oh, this could be leaner, this could be leaner. And that’s very important. But so that was huge and also made me kind of focus on a new niche where I had to build first-class tools for multilingual support. And a lot of my competitors—there’s just a few that came out within a matter of weeks of me too—because when this was released, it was literally days after the ChatGPT API became available.

Dave:
Yeah, you’re very fast on this.

Aaron:
So actually, when I was building the MVP, I was still using the old GPT-3 chat API, which was—people don’t remember this—but it was literally a hundred times more expensive. So that was exciting. And since then, just been growing and doing that. And I eventually transitioned to doing this full-time in the spring, last spring.

Dave:
And unlike a lot of—I follow a bunch of the solopreneur Peter Levels types on Twitter and living around the world doing cost of living arbitrage and doing all this stuff—you’re in quite a different place from what I understand. Certainly, last time we caught up, you’re a pastor, you’re deeply embedded in the community. You have kids, you have a house, you’re not packing everything you own into a rucksack and setting off to Bangkok for three weeks to go and build something. This was, I guess, a different sort of transition than young, free, single, sleep-on-a-sofa kind of solopreneur that you see a lot of. So that must have taken a bit of courage—that conversation with your wife and your boss and everything else. I imagine you handled it well, but it must’ve been non-trivial for you.

Aaron:
Sure. Yeah, and I joke about that too. Most people like the startup life. They think, okay, you are going to burn away your twenties, maybe your thirties, grinding for the VCs or whatever and trying to build something. And maybe if you’re super lucky and the 1%, you become a billionaire or whatever, and then eventually settle down and get married and have kids and all that kind of stuff. I kind of did it the opposite way. I got married at 19, had kids young—my kids are all, I got two in college and one’s a senior in high school. So now I’m able to put that time in and that grind in on the other end, I guess, which is good.

But definitely, yeah, it’s a huge mindset shift. I was probably your best employee—faithful, loyal. I would never complain. That person that’s always there seeing a problem and figuring out how to fix it and solve it and being self-directed and that kind of thing—that was kind of my role as CTO. And I know that my former boss, James Farmer, really appreciated the time and stuff that I put in there, and he was so supportive. I mean, it was scary to kind of take that step and say, here’s what I’m thinking. I feel like this is kind of the next phase of my life. I don’t know how it’s going to work financially or anything. I just want to try some stuff and see if it takes off. But he was super supportive. And even just my transition away from there was very gradual. I had a good amount of time to train up a number-two person that could work alongside me. And then even after that, when it was like, okay, I am going to step down from CTO, they still had me on as an advisory role, part-time for almost a year just to make sure that transition was smooth and everything. And that was a huge blessing as well. So nothing but good things to say there, but definitely it is a huge step. It’s a very different stress, like owning a business, running a business versus being an employee.

Dave:
And you’ve done the kind of build-in-the-open thing. So I feel comfortable asking you about the financial side of it that you touched on there. How is Docsbot going? What are the revenues? What are you planning to do with the business? Any of those questions—obviously, don’t answer if you’re not comfortable—but just curious for your journey.

Aaron:
Yeah, well, I’ve never been super excited about the VC route or whatever. I don’t know, coming from that—being kind of the employee to being an owner—the biggest benefits of me wanting to do that was to have ownership, to have full ownership, to not have bosses anymore, kind of have that freedom. How hard I want to grind—if I’m going to be dedicating every hour of the day, or I just want to work. I mean, even with this startup going on, over the summer and stuff, there were periods where I literally took almost a month off and just like, okay, I’m just going to relax. I’ve got some employees that are helping with support or covering bugs, that kind of thing. But I’m okay taking a month off just to be with my family and chill, even if that means that maybe I’m losing some growth to competitors or because I’m not doing every single thing I can.

So I prefer that. I prefer to work—they say that we’re designed to work more like a lion than a sheep. Sheep graze all day, and what you classically do—a nine-to-five job or whatever—you’re just constantly plugging away. And ultimately, I think that’s less productive than if you can have focused times, one to three months where you’re just hardcore all-in, building things, making stuff happen. And then you’re taking a month off to just go to the beach or whatever. And that’s kind of how I’ve lived my life—making sure I have that purposeful time as family and kids, especially as my kids were younger. So I wanted to have the freedom to be able to do that and continue to do that even as an entrepreneur. So it’s kind of a conscious choice. We’re fully bootstrapped. We don’t have any investors currently. When we first launched, and even to this day, we still get a lot of investor interest. This field being the new technology is on fire.

But I’m not necessarily opposed to that if it’s a strategic thing to where—not just money—but some kind of thing where, like, I mean, if I can sell 10-20% of the company and this partnership is going to make it grow 3x, I mean, it’d be dumb not to do that. But I haven’t found that opportunity yet and am just bootstrapped and staying profitable and that kind of thing.

Dave:
I think that also leaves you with the freedom to make those sorts of decisions. And maybe that’s a sensible follow-on question. There’s been a lot of conversation about what’s a kind of durable moat for an AI-powered business, and what is just a ChatGPT wrap that’s eventually going to get rolled up into a more capable central product from the OpenAI team or Perplexity or… How do you think about Docsbot in that sense? Are you just taking things day by day? It definitely doesn’t seem like you’re in the mindset of ‘get out while the money’s on the table.’ So how do you—do you have an alternative master plan, or is it just keeping on things and iterating away? How do you strike that balance, I guess, between tactics and strategy now that you’ve found PMF?

Aaron:
Yeah. Well, it’s definitely been, I guess, an underlying fear. No one, even the best VCs, they have no idea how this is all going to shake out. But I think we’re starting to see some patterns, and I think the patterns that I’m seeing are that the AI wrapper companies are actually going to be the only winners in all this, which is funny because it’s not really the underlying technology—it’s the execution of it for a specific business niche. That’s where the value is. That’s where any kind of moat that you can build in AI, that’s where it’s going to be—applying the technology to the niche for that specific business. Then also tied into that is marketing distribution.

So yeah, there are definitely products out there that are better than mine. You have OpenAI or Anthropic or Google or whatever releasing things all the time that may cut into my features, and that’s scary, obviously. I’m like, oh, is it just going to die tomorrow because this becomes free or whatever, or easy? But from what I’ve seen, there’s power in the niches. And maybe it’s not going to be like a billion-dollar company or anything like that, but I think that it can definitely be a decent lifestyle company, at least for the foreseeable future, by building the specific tools that businesses need—or consumers need—even for a specific niche.

Dave:
Where I last caught you was maybe it’s not going to be a billion-dollar company, but who wants the stress of running a billion-dollar company anyway?

Aaron:
Right? Yeah, maybe it won’t be a billion-dollar company, but I definitely think there’s going to be an untold number of hundred-million-dollar companies that are doing this kind of thing. And often lifestyle. Often you’re going to see profit margins that are unheard of. You’re going to have a team of two to five people that are running a hundred-million-dollar company, massive amounts of revenue per employee because so much can be automated now—all that kind of thing.

Dave:
Well, maybe this is a good segue then. AI… I stayed with crypto and Web3 as my primary interest. I do my best to keep up with what’s happening with AI, and I’ve done some build stuff, but more just for my own learning and certainly no product stuff. But one of the things that’s happening at the moment in crypto land is the kind of AI agent meme coin thesis, where you’re going to end up with an AI agent. In fact, I mean, we’re sort of seeing prototypes of this already, but the thesis is, like, somebody is going to spin up an agent and it has a directive to go and generate a token, and their only job is to maximize the value of that token. There’s a couple that are kind of like that but have a human in the loop to some degree or not.

But that intersection between an agent having, I guess, enough intelligence—loosely defined—to go and pursue objectives, to consume comments and feedback, and to respond to those and to have a dialogue with a community, and then the ability for it to have a different type of agency in terms of having a digital wallet that has funds that it has sole control over has been a kind of interesting one to watch over the last few weeks, I guess. And so where this segues is, I can imagine there’s going to be teams of two to five people running hundred-million-dollar companies. Are there also going to be AI agents who end up running these sorts of businesses? Because you don’t have to do much in terms of nesting and recursion to imagine that there’s a CEO bot who has a CTO bot who has a bunch of developer bots. And actually, when they capitalize and have the ability to pay for infrastructure through that digital wallet, then sure, there are some advances that one might need to make in terms of capability and everything else. But do you think that’s plausible, that we’re going to see entirely autonomous businesses succeed?

Aaron:
That’s a good question. I have plenty of experience with these agents and things like that. It’s hard for me to imagine them getting to that level right now. There’s a lot of excitement around it. And even with the latest 01-type models, we’re just very far away from that full autonomy. Now, you can build agents that are tuned to a specific task and that work pretty well. The more you can guide them and have a set path for them to follow, then the more reliable they can be. So definitely, you can do some amazing automation stuff. No one imagined being possible before, and we’re on the cutting edge of that.

As a business owner, I’m trying to make our business AI-first. Instead of thinking about who can I hire for this, I think about what can I build to do this for me. I try to have that mindset up front because there is so much potential. There’s so much that you can do, especially in marketing. And I’ve been doing the AI coding thing for years. But as far as a free agent that has that level of autonomy, I think we have quite a ways to go without the human in the loop.

Dave:
I don’t feel like we’re ready for that world, so I hope it is a bit further away. But it’s been interesting to watch at least the first realizations or the first proto projects that are starting to show what could be possible down the line, for sure. Alright, so if that’s a bit too wild and a bit too far off, then tell me maybe what you are interested in terms of where you see this technology going. Are you just, as I say, sort of tactically iterating on Docsbot? What are the things that you’re playing with?

Aaron:
Yeah. Well, first off, I don’t think I ever described what Docsbot does. So let me start there. Basically, we allow businesses to build and train a custom chatbot, and they can use that for customer support, like tier-one customer support on the frontline, either through a chat widget that they can embed on their website or by connecting it to their ticketing system, like Help Scout, to where it could write automatic replies, or at least write a draft that your human employees can either approve and send or slightly edit if they need to. And then we have a ton of businesses that use it for internal knowledge access. So there’s a huge use case—you have these huge standard operating procedures, big PDF documents, or intranet sites, all this data, and it’s really hard to access it or to get real answers to questions.

So a lot of our customers are actually just using this internally for support or as a tool for their own employees. And then we have people using it just for research in general, like personal uses. They’ll upload their directory of PDFs, and they want to be able to ask questions of their documents, things like that, to extract information. And then, of course, there’s content generation. People are used to using that on ChatGPT or platforms like that. But with this, you can have it trained. We have one client that’s like an ad agency in the medical drug space, and so they have—I don’t even know how many—50 clients, and they train a custom chatbot on all the information about each of those clients. And so they use that internally, their marketing team, to generate ad copy, generate copy for websites, social media, all that kind of stuff.

So a ton of practical use cases for business. And the biggest thing that I’m always working on and always building is we have to build the internal tools that businesses need to use this stuff. Yeah, you could build a custom GPT or whatever on ChatGPT to do some of this stuff, but you need both the tools and APIs to be able to deploy that easily to your users or to your customers. So that’s one side—deploying it in a way that’s useful to them. And if you’re doing it for customer support, you need that seamless handoff to human support. If it detects that they’re getting frustrated or can’t answer, you want to make sure that there’s no gap in there between going from the AI to human.

And then the other part is all the management tools on the backend. So you want to track chat logs of what your users are asking, figure out topics, get reports about how well the bot is doing, how well it’s able to resolve customer queries—all that kind of stuff. Day to day, I’m often working on those kinds of features, just adding onto that the tools that users need.

Dave:
Well, I think that’s a good way to wrap up. I appreciate the deep dive into your journey and everything you’ve built with Docsbot. It’s fascinating, and I wish you all the best!

Aaron:
Thanks so much, Dave. Always a pleasure chatting.

Dave:
Likewise. Take care!