Listen to the full episode :

In today’s episode of the Masters of Automation podcast, the host Alp Uguray speaks with Antti Karjalainen, the founder of Robocorp, a Python-based open-source RPA (Robotic-Process-Automation) platform. Antti shares his story about what led him to automation, why he chose to open-source his product, the benefits of open-source, such as collaboration and innovation, and how it aligns with Robocorp's mission to make automation more accessible. We talked about the Robocorp community, managing remote teams, integration of generative AI and RPA, as well as the ethics of AI.

Our conversation then shifts toward the future of the industry, and Antti shares his insights on how he thinks the industry will evolve. He talks about integrating ChatGPT-like solutions within the product stack for automatic code generation and the challenges that come with it. The ability to integrate different solutions and providers within the core-RPA automation stack allows them to be more flexible and focus on the core automation promise. Being open-sources and Python-based helps with that integration as well.

We also discuss the industries that are the most and the least resistant to change and automation based on their exposure. The podcast ends with a discussion on managing a remote team. Antti shares his experience of running a fully remote team and the challenges and benefits that come with it. He talks about how a remote model creates equal opportunities for people from all over the world to participate fully in the company and get ahead in their careers.

This episode provides insights into the world of automation and its impact on businesses and employees. Antti's experiences offer valuable lessons for entrepreneurs, startups, and anyone interested in automation and AI.

Here are some of the questions we discussed:

• What inspired you to delve into the world of automation and RPA? Was there a particular problem or challenge that motivated you to build solutions for it?

• As the creator of an open-source RPA platform, how do you balance the needs of both developers and business users within your community? Are there any challenges in catering to both groups?

• The field of automation and AI is constantly evolving. What exciting developments do you see on the horizon for the industry, and how do you plan to keep up with these changes?

• In your experience, which industries have been the most resistant to change and automation? How can we overcome this resistance and help businesses see the value in adopting these technologies?

• With the rise of remote work, how have you managed to build and maintain a successful remote team? What strategies have worked well for you, and what challenges have you faced?

Transcript

Alp Uguray, Creator & Host: Welcome to the Masters of Automation podcast series. In today's episode, we have Antti Karajalanen. Welcome Antti, thanks for joining the episode. 

Antti Karjalainen: Good to be here. 

Alp Uguray, Creator & Host: Antti is the founder and CEO of Robocorp, San Francisco and Helsinki based company that set out to change the licensing and delivery model of robotics process automation with open source technologies. We're very excited to speak to you today. We have a packed agenda around how the RPA world is going to evolve, how your personal story that led you to start Robocorp, and particularly emphasizing on the licensing model as well as being open sourced. And then we'll dive into a little bit of the broader industry topics, like where the RPA is going next and where the industry is going next, especially with the intersection of AI and the recent developments in the generative tech space. So to kick things off, Antti, what led you to RPA and automation? What led you to create Robocorp? 

Antti Karjalainen: Yeah, I mean, everyone has their unique story here. But I think nobody who's following the space when it took off couldn't notice how fast it was growing. So that's what draw my attention. I come from an engineering background myself. So first, I kind of look at RPA as, huh, that's weird. You're doing UI automation mixed with kind of deeper automation at the same time. What are the use cases? What's the value it's providing? And after investigating into that kind of more deeper, I realized that it's something that every large company will do at some level. And then with my background, I could see that I had some exposure to the QA test automation world. So I could see that going from these desktop use cases, it'll be the first way. But then we're going to see a wave into more complex back office use cases. And then started really thinking that, what should a developer tool look like if we build this for the automation professionals and approach it from that side of the world instead of starting with the simplistic use cases in mind, kind of record and playback type of things. So with that observation, I had some background in open source projects, namely RoboFramework, which is a Python-based keyword-driven framework. And I was able to connect some dots and thought that there would be something behind it. And I started exploring it, saw the cost of attraction with the solution, saw also what it actually takes to build a platform like this. It's not a trivial feat in any way, especially when you're serving the enterprise as your customer base. So kind of one thing led into another. And I was really compelled to start the company. I was never actually planning as a career move to start a startup company. That wasn't something that I was sort of, it's a great thing to do, obviously, but it wasn't something that I was like, I have to start a company. How could I start a company? It sort of came to me in a way that I couldn't just look away from it. 

Alp Uguray, Creator & Host: You've seen the opportunity as well as getting your background that led you to a way there could be something to build here and then address a pain problem that users have. And in terms of the community, it's interesting as well. So I like to ask, because it's an open sourced software, you'll be able to have anyone come in with Python background and whatnot and start building. So how is the community look like at Robocorp? I would assume they're mostly developers with some background in Python and be less like on the citizen development. 

Antti Karjalainen: You would be surprised how mixed it is. We have people with accounting background who came into the tech and started using it. So it's surprisingly not as sort of pro-developer heavy as you would imagine. So RPA obviously has the promise of being a rapid development tool. So the project should take days or weeks rather than months. So obviously, the emphasis is on ease of use and rapid iteration. So oftentimes, a person with in-depth software skills might overlook the stack and say, hey, I can build this with Python. Why should I use the Robocorp stack? And kind of not see all the things that we've built into it, which makes it actually super good for even the pro-developer. So we have people from mixed backgrounds all the way from the citizen developers to really experienced software engineers using it. And the more experienced people are actually using it in some really creative ways as well. And so the stuff that we have on the tech side is the core open source stack, which is a tool chain that manages your bot environments, your building and running the bot, how you set up the project, how you go about and operate the project. And then you have Python in there as sort of a natural base platform. And then as an abstraction layer, you have RoboFramework. So people can build on RoboFramework layer, which is like the easy human readable, very high abstraction layer. Or then you can build with Python as well. And so you can kind of choose your own level. That's what we have for core open source and then developer tools wrapped around it. So VS Code as an extension. And then we also have the low-code development platform automation studio that's actually now coming out of beta into general availability. But still very much in working progress. We keep adding stuff on that weekly basis. Really, the goal is to make RPA development with the Python-based stack as easy and fast as possible. 

Alp Uguray, Creator & Host: And we've seen, I think, in the early days of RPA when people were feeding the flowchart low-code, they were invoking a lot of Python or JavaScript or Java codes. And then we were trying to teach them, OK, we want to stick to the way the platform does it. And then I think similar way with the RoboFramework that you have at RoboCorp, make sure that people follow that structure and then leverage Python. And I mean, it's a very interesting space, like you mentioned. And it's a very good blend of business users adopting tech skills and techies adopting the RPA skill set and then cutting together and actually solving the business problems. So I think if we look back a few years, five years ago, the industry changed a lot. And right now, we've seen a lot of different vertical impact of the RPA as well as the product stack growing to become a platform approach as well, especially for an enterprise where they have a full automation lifecycle. Now that all of the companies are interested in how to solve their business problems based on each of the stages in that lifecycle, how do you see right now the industry evolving over time? What's it going to be like based on your experience interacting with the community, the users, and customers? Where do you think it's heading to right now? 

Antti Karjalainen: So what we saw a few years back, a lot of the first generation platforms started adding a lot of these sort of functional areas that are close to RPA itself. It started with the desktop automation, kind of moved to the back office automation, unattended side. You started getting the center of excellence in the enterprise being involved rather than most of the citizen developers in the early phase. And then now adding on top of that API automation, IDP, chatbot capabilities, these kind of things. And so as we speak to our enterprise customers, what they tell us is that there are space, for instance, in IDPs moving so rapidly that why stick with a single platform vendor when you can choose your best vendor in each of these categories and build the best of breed solution? So that's been our strategy all along that we want to just focus on building the best possible platform for RPA itself, which is a core capability, being able to deploy bots that can act as digital workers in an enterprise setting. And then having interfaces and APIs on the control room level, on the orchestration level that allow you to plug in API automation, and these other adjacent solutions. So I think sort of the monolithic platform is not going over as well as we thought a few years back. And we are seeing this kind of more mixed approach of, hey, there's a cool new technology that came, blew up two years ago, or one year ago. I want to use it here. How can I integrate that with these RPA capabilities? And really what RPA is allowing you to do is, in my perspective, it's kind of crossing the last mile of a use case in an enterprise. So you look at these new companies coming up, especially with the new AI wave. All of them are doing amazing stuff with the generative models, which is, I think, amazing. And we can talk more about that, how that applies to us. But then if you go to see the demos, they're all, OK, I have a browser extension. And I'm automating between Stripe and Zendesk and Shopify and all of this. Like, huh, what if your business doesn't operate in the browser? Like, yeah, those things actually exist. And so that's where it kind of breaks down. And that's where RPA comes in, even in the most modern AI-enabled technology, to cross the final mile of a use case that you need to complete. So, yeah. Yeah, that was a very interesting point, actually, because you're right. I think the business, first, there's the data security aspect of what you can transmit through the browser. And then the second, can it scale to the entire organization to use as an automation?

Alp Uguray, Creator & Host:I think we can touch on that because it's, I think with IDP, for example, and the ability to have plug and play approach so you'd be able to integrate any of the, maybe IDP, maybe like predictive analytics and so forth, to be able to integrate that into RPA platform and RPA is kind of the information transmitter and then the trigger to start those tasks and process automation. So how do you see terms of the scalability of Robocorp's integration in that case? Because if you think about it, especially tying to a little bit of generative tech, since the platform is based on Python, you'd be able to integrate with some GPT-3 or maybe stable diffusion and help with a low code interface that can maybe generate some code and whatnot and increase the speed to build. So what are some of the things that you're seeing? 

Antti Karjalainen: Yeah, so us being based on Python and also being code native, so when we talk about low code, what we do is we have a low code app or bot builder that generates code in the backend. So it's all in the sort of code model at the end of the day. So right now today, what we can do is tap into things like code generation models, copilot, all of that stuff, do things like, okay, hey, explain this piece of code to me or create documentation. I had an error, can you help me fix it? These kinds of things, but as there's this conveyor belt of innovation that's rolling really fast right now, I would say that in a year, I mean, it's hard to predict these timelines, but in a year or two, we should be able to see that you can pretty much draft a bot by prompting. This is my outline of a business task, help me build an outline. Okay, here you are. Okay, this function needs to do that. This needs to do that. Go and try it out. Didn't work. Okay, that's the error. Help me fix it. Stuff like that. And then you put it in an automated loop and all of a sudden you kind of like have the bot build itself and then you're running in production. Oh gee, I hit an error. What's that about? Go and explain it to me. How do I fix it? So in a short while, we actually might be in a situation where you can build Python-based bots without actually having to know how to code too much at all and doing it like in like 20% of the time that it will take you to do it manually. So that's pretty interesting outlook in our space. And then kind of jumping from the developer to going into the actual capabilities themselves what the automations can do. That's another topic. So by the way, when we talk about RPA, I think that's like interesting new transformer models coming out, especially it is trying kind of to build the action layer for these language models. So actually being able to take action on applications in digital work is not yet there. So we haven't seen anything come out, but in RPA it's sort of heavily rule-based. When you go into a bank and say that, okay, I want to deploy a digital worker that needs to complete this action, break it down into code, audit the code, make sure that it does every step of the way what you expect to do. And it does it the same way every time. You don't want to kind of let a model loose in your enterprise systems that you're not 100% sure what it's going to do. So I think the code is going to be there whether that's human written or not. That's my prediction, at least today, ask me tomorrow if I'm still thinking about the same way. But then on the capability side, we'll see rapid iteration of capabilities that we see the bots being able to do. Right now, it's good for these language models are good for summarizing, condensing information, maybe generating a certain type of answers. You would see that customer support work is kind of prime example of where we can apply it today. Legal work, case management, stuff like that. We have sort of these use cases where we see that today's models are good, but then you kind of start going into this like really interesting area of sort of close to like emergent capabilities when you start chaining together these models. So you might give it some capabilities, you might chain together answers from another model to another model and sort of have them play together as an agent. It's almost like watching something emergent happen there between that there's projects like LangChain that are enabling some of that work, which are all by the way, based on Python. So it's kind of playing in the native field where we are. So I think the ecosystem choice on our side is definitely helping here. 

Alp Uguray, Creator & Host: Yeah, absolutely. I mean, that's very interesting point. And one thing that stands out is also the ability of generative tech to not only create content, but also identify the pain problems in the process itself. Like maybe address some of the issues related to auto healing, or finding where the bugs are at and then fixing the bug by writing the code themselves. And then maybe it goes through some Q&A, through a developer's approval, but it can really accelerate that process, right? I've heard stories about auto-healing bots like so many times, basically retry loops. But I think we are actually there where we could see some of that come to reality not so far away in the future, where I've seen a language model create a Python code and run it, gives out an error, then says, okay, there's an error, fix that error, and then go modify the code, run again until it actually does what it has to do. And that's like today. Which is awesome. And I think it's the, like you said, I think auto healing has been an urban legend for a very long time. And I think now maybe with generative tech it makes it possible. And it's only one of the use cases like from your exposure, I think the core products stack because it's Python based, it's very integrable, right? It's very fluid and it can keep up with the new technologies that are coming up. As you guys are building and integrating, like what are some industries that are the most resistant to change, right? In terms of, they resist to the new idea, or they resist to that new idea, or they resist to the new idea, or they resist to that integration or that solution proposal. And obviously it's two-handed, right? They're most resistant one and they're the least resistant ones more open to change. Like how do you see that? 

Antti Karjalainen: Yeah, it's a good question. Well, when we talk about, let's say financial institutions, we have some great customers, banks doing really cutting edge stuff with our technology, really taking off in growth. At the same time, try going to a new vendor on boarding in a bank. It's like pulling teeth and it's for good reason, obviously there's compliance and process in place for good reason. I think there's so many customers and people who wanna talk to us and would love to use the tech, but then, oh gee, I need to go through procurement. I need to go through legal. I need to go through compliance. And well, that's our business at the end of the day as a tech company to make sure that every box is crossed and we have everything tight and controlled. So, unfortunately a big part of the value has to be compliance, security, robustness that we provide. There's no way to go about that. And coming back to the new stuff happening in the cutting edge of the new AI wave is, I'm just like waiting to be able to apply that to these companies, but then I know what the hurdle is to actually get in. So, I would just say that it's not necessarily the people in the companies, but the nature of the business itself and the rigorous compliance they're under is making them resistant to change. And it's sort of a good thing for new companies out there who are kind of imagining what will be possible in today's world is also realizing that when you actually wanna deliver it, you need to go through all of this. And so that's why I'm kind of excited about Robocorp as a sort of company who been here for a few years already. We have the enterprise customer base. We know how to operate there. We have all the checks in place. So, now actually being able to take new and emerging technologies, this company is super exciting. I think it's the adoption of a new technology takes time for a company. And then as the customers take their time, their competitors adopt the technology. And then it really gives that competitor the edge to probably capture certain market share. And then that's really after that, when they're like, okay, maybe we should start to think differently. And then there's this ChatGPT, maybe we should adopt it. 

Alp Uguray, Creator & Host:There's the high portion of the tech coming and then there's the competitive portions as well as actually solving the business pain problem and then addressing that on time portion, which I think the latter happens only for a few smart customers that really think forward. 

Antti Karjalainen: Yeah, you always need to have the lighthouse customers who's forward looking enough to take the leap and then we have customers who first conversation they said to us that, hey, this is a new thing, right? Yes, it's a new thing. We never do anything first, come back in a year. That's fine. We're happy to come back after a year as long as we get the first brave ones to do it.

Alp Uguray, Creator & Host: I'd like to take the discussion to a little bit of the like generative tech and Robocorp and also some touching on the regulations and maybe ethics of AI portion of things as well. Like I think, obviously within our space in RPA, like generative tech's benefits are are not as broad as like asking the chat GPT to write a blog post, but we can ask it to write an SDD or a PDD or generate a process map or use codex and then create some codes. Whereas in the marketing world or in the design thinking world you can actually create a compelling article and it can have bias, it can have you know certain conflicts of opinion or maybe not factual or getting the facts wrong. But like what are some of the things that if we were to apply a generative tech within RPA like beyond generating code and documentation, like some maybe regulative or like ethics of AI type of discussions that we may face that maybe we don't think about today? 

Antti Karjalainen: Interesting, you know there's the question of bias, how you prompt them out, what kind of bias it has, and also the learning data itself. And then by the way copyright issues is like you know that's a thing that has been in discussion, it's like you pulled all this data into this model and now it's making stuff up based on your creative output. Is that right? Are we actually allowed to even do that? And so that's the obvious thing. But then I would imagine that in the first stages of applying all of this, especially in our domain, it's going to be human-in-the-loop type of interactions when you cannot explain the model in any reasonable way or can't really know what it's going to come up with. There's no way to really give control over to it. And by the way, leaping from text generation to actual action in the real digital world, it's still something that hasn't been done to a really meaningful extent. And that's something that I think when we see come out of the box at some point, it'll be the next wave of transformation that's going to happen. Super interesting to see that. But right now you can kind of give skills to the models and they can kind of play against these skills and make stuff already happen. But doing that in a business context where you're actually liable for the actions, if you're not going to have any human-in-the-loop, it'll not go well most likely. We've seen this, the Google demo that they did was a barred phase live. And it's like, okay, you don't know what's going to happen. So it's in a way a scary business if you kind of just let it loose and you're kind of liable at the end of the day. It's not something that I think we're going to see this year, but next year. 

Alp Uguray, Creator & Host: I agree with that. Maybe the risks really overpass the rewards. If FBAR or ChatGPT says the wrong thing in the name of an enterprise, then it can be really problematic, detrimental for them. It's the human-in-the-loop is an interesting one in two angles. I think the first is the content needs to be moderated by a human before you push it forward. And then the second is maybe it can augment into the attended automation more. For example, in attended automation, running triggers for front office, for running maybe like a contact center agents going into their control center and then around me this automation to pull up the billing about this customer. Then maybe recognizing that from the text within the conversations or within maybe some text inputs by an agent itself, it can help the communication with robot and human better. Again, to what extent, maybe not this year, but it could be maybe one potential area as well. And the speech recognition stuff is also so powerful right now. You can transcribe every conversation, have them in the customer's file and then go through and when this next conversation summarize what's been already discussed in five bullet points and do that live while you're interacting with the person on the phone. So what was the first thing you wrote in ChatGPT? 

Antti Karjalainen:I asked it probably to write up a short automation script using Robocorp. I'm not surprised about that. Yeah, probably that was the thing. Then I've been using it prolifically to write up stuff on various... I haven't fully written a blog with it, but inspiration. And then I think I wrote the Christmas greeting for the company with that, but I still felt it wasn't the right tone. So I had to modify it. And oftentimes when you're out of ideas for something, you ask it to, hey, generate me five ideas for this and that. But sometimes it feels like lacking a bit of originality, especially when you're in a narrow domain. I can see that marketing usage is absolutely wonderful to have that kind of ghostwriter, which is going to be 99% better than most PR agencies will be able to do. 

Alp Uguray, Creator & Host: Yeah, I see your point. I think to have critical thinking about how the industry is going to shape or how it's going to address some of the pain problems, what are some of the gaps and then solutions to fix those gaps is like?  You guys have a global team right now. And then you mentioned earlier to me that you guys are fully remote, so what were some of the challenges as well as the benefits of running a fully remote company? And again, were you fully remote before COVID as well, or was it something that COVID triggered? Can you speak to that a little bit?

Antti Karjalainen: Yeah. We started out as a fully remote team between the founders because we weren't in the same cities, even though we were kind of close to each other, we weren't in the same city. So we started out as a fully remote operation anyway, and hired the first employees. At some point, people started asking for an office. I was like, oh, we need an office. Like, what do you do there? We got a small office in Helsinki where we had hired some developers. We had an office in SF, but then with COVID, we really didn't develop any office culture to begin with. It was just a few people would go in there and work. So it was never the full company in the same place in the first place. With COVID, we pulled out all the stops. I remember this decision. It was the first summer and I was driving somewhere and I was thinking that, man, this could take six months, 12 months, two years, who knows, like for a startup, that's going to be forever anyway. So we just need to think that this is going to be the way we operate forever. And it wasn't a hard change by any means, but it just allowed us to unconstrain where we hire. So we could start hiring from different countries, like for development work, we hire different countries in Europe, trying to keep them close with time zones. Then for go-to-market, that was natural to build that team in the US for a lot of our customer bases, but we still cover Europe as well with the people we have. And then now that we have been extending our partner operations, we have pretty extensive partner channel already. They will have a lot of people in India. So having people, boots on the ground, being able to get together and actually meet face-to-face there, it's a huge benefit. So just being close to where the customers are, where the partners are, where the development talent is, and being able to choose from those without being constrained to that office location, I wouldn't be able to pull together as many talented developers we have in Helsinki, for instance. And if I would pull that together in San Francisco, we wouldn't have any money left. So that's where we are at right now. And then kind of the disadvantages of it, it's when you have big changes you may need to make in, let's say, strategy or the direction of the company, it would be great to just pull everyone together and, hey, let's just talk about this. I have this thing, we want to change X, Y, and Z. Now you need to call an all-hand Zoom meeting and it doesn't really convey the energy or the same kind of message oftentimes. I would love to have everyone working in sales in the same office and just have the energy there, sharing all the interactions, stuff like that. But you need to compensate with having in-person interactions whenever possible, doing off-sites and team off-sites and those kind of things. And then going out and meeting customers in person, that's the thing that actually is really good. Not trying to do sales fully remote is a great thing to do that. But I would also add that what remote model actually allows you to do when you fully embrace it, when you don't have the central hub where, okay, this is where the management team sits and then people are remote. Don't do that because that'll imply that you'll have to be in the central location if you want to be close to where the decisions are made. So if the headquarter is online and everyone is committed to that, that sort of creates equal opportunities to people from everywhere they are around the world to participate fully in the company and really have an equal opportunity to also get promoted, get ahead in their career, even without being part of the headquarter somewhere. Because that would be the worst kind of setting. It's like you have the sort of the management team here, a nice office somewhere, and then higher developers wherever they are or higher sales reps wherever they are. That doesn't work. 

Alp Uguray, Creator & Host: Yeah. Yeah. I agree with that. I think in a fully remote company, the information transmittance from two person and all that information changes over time as it's been told to the fifth guy or the fifth person in the chain. And if it's retained as the same identical thing, then that's great. But if it slowly gets modified and then more perspectives add, more perspectives add, then it can really change that strategy perspective. So that said, we're almost at time. First, I'd like to thank you very much for taking the time and joining the call and sharing your perspectives and opinion. And I know it's a little late in Helsinki, so I also appreciate you taking that hour with me to speak about your experience about RPA, AI, the future, remote teams. It's been a pleasure to have you and speak to you. 

Antti Karjalainen: Yeah. Thank you so much.