Introduction
Andrzej Kinastowski (AK): Konnichiwa! Welcome to the AI automation Dojo where we stir into the abyss of the modern workplace and the abyss stirs back with a loading screen. Today we’re diving head first into the glorious and sometimes bewildering state of automation in 2025. We’ll be tackling the cyclic history of automation. I’m your host Andrzej Kinastowski, one of the founders of Office Samurai, the company that dares to ask: “What if business consulting could be done without the PowerPoint presentations that clearly violate the Geneva Convention?”. Today we have a special guest Dominik Jaskulski. Dominik, welcome to the podcast.
Dominik Jaskulski (DJ): Yes, hello everyone, konnichiwa. It’s nice to be here.
The cyclical nature of automation hype
AK: Dominik, you have been in the world of automation for a very long time. Tell the people who are watching us: why are you here, and most importantly, when is the last time you wrote a VBA macro?
DJ: I guess the main reason is that we run business together. Prior to Office Samurai, I was working as automation manager, responsible for scaling up automation programs at GBSs of big enterprises. I guess the last time I wrote a macro was around eight years ago. I have 16 years of professional experience, and one of my first tasks when I was still an intern was actually writing a VBA macro. Then, for many years, when people were mentioning office automation, very often in 80% cases they actually meant VBA.
AK: Do you miss it sometimes? I also started with macros. Today we want to talk about the state of automation in the year 2025. What do you think we should be talking about?
DJ: I would actually start with what we will not be talking about, because definitely we are not the guys who are good with industrial automation, with going to production company talking about Industry 4.0 IoT. If you are interested in learning how automation in office environment in services is being done, then please stay with us.
AK: You started with the VBA macros, but a lot has happened since then, right?
DJ: Definitely. In 2014-2015, everyone thought RPA is coming, it will replace all type of this traditional macro-based automations. The hopes were super inflated. When we were speaking with some heads of GBS organizations, they actually thought: “70% of accountant work will be automated”. Since then, my suspicion is that actually the number of accountants grew, it didn’t fall. I have a feeling we are in very similar moment with AI also. One or two years ago, after GPT 3.0 has been published by OpenAI, everyone expected AI will automate everything, AI will take your jobs. Have you heard about Clara?
AK: No, Clara, no. I heard you mention it, but I don’t know the story itself.
DJ: The Clara CEO one and a half year ago said: “AI is already so good in answering our customer queries” they decided to lay off 700 people, planning to lay off a total of 3,000. Their valuation was also inflated a lot – I think $46 billion, then after one year after those layoffs, it was seven billions. They announced recently that they will be hiring again. They are saying that it’s more about quality that they want to provide a better quality, which cannot be really done only with AI.
AK: But why didn’t it work? RPA is now business as usual, and if you turn it off, the company will collapse. When it comes to AI, it looks like we’re going through the same thing: inflated expectations. Why doesn’t it work? Everybody’s showing off their labs where everything works, but somehow when you try to implement it in big scale, the market doesn’t seem to be able to make it happen.
DJ: Actually, I don’t agree here with you. I think that RPA proved to be useful. We know examples of companies where they have automated 15, 20, even 25% of workforce. I really like this quote that we overestimate what a technology can do for us in a short term, but we underestimate the influence it can have on our lives in a longer term. It was the same with the internet. If you are thinking now about a 30-year period, it really changed how the economy looks like, and what we do as human beings.
AK: We did have the dot-com bubble. Do you think GenAI will change the world in such a fundamental way as the internet did?
DJ: Probably in longer term, yes. A lot of people are thinking about those simple chat bots, which is more like replacing Wikipedia or replacing Google. But the actual power of LLMs is that you can combine it with knowledge of your organization. This AI can not only answer you, but even suggest some approaches, help you do some planning. And then you have agentic automation. AI solutions where you don’t really need to see the AI as a human being; it can be somewhere on a server working in the background for you, helping your company to move goods from place A to place B.
Zealots, deniers, and the cost of waiting
AK: I think it’s a good way of thinking about GenAI and how it’s going to be implemented kind of below the surface. You actually have zealots on both sides. There are people who think that AI will change everything in like one or two years, that will make unemployment rate skyrocket. And on the other side, you also have those companies that prefer to build a road map, hire McKenzie consultants, and they are already behind.
DJ: We are often recommending to be somewhere between them; not to be over-optimistic, but we also cannot pretend that those technologies don’t exist. We should actually start checking how we can implement those technology within our business processes, how AI can support our customers, order management, HR functions. The sooner your organization will start piloting, the quicker it will learn where it makes sense to use it and where it doesn’t.
AK: Waiting doesn’t help you much if you don’t have firsthand experience. We know those companies that basically have nothing when it comes to GenAI. Everything is blocked with a genuine fear of some data leaks. Not even things that are easy and safe to implement, like if you’re on Microsoft, you can get a Copilot. But a lot of companies are still refusing to do that because they’re still planning and building road maps.
DJ: I would actually say that there are a lot of companies who are on the leash of compliance and IT security. It will take them the next three, four, five years to start anything with AI. There are also a lot of stupid ideas sometimes even from board members or CEOs because they don’t understand what AI actually is. I was speaking with one of our customers, where the CEO of the whole 10,000-person company came and said: “I want you to spend 200,000 on Copilot licenses”. Companies are buying them to show that we gave everyone access to AI, but very often they have not trained those people.
AK: If you’re buying a technology that is so fresh, people are not yet used to using it. We know organizations where we have been doing a lot of trainings for them.
DJ: Maybe Copilot will not make sense on every position in your company, but for sure you should play with those technologies. There are areas in your organization where writing emails, doing notes from meetings, preparing presentations, where this type of software can improve the productivity of your employees. But if you are paying several million dollars for this kind of software and only 5-10% of your workforce is using it, it doesn’t really make sense.
AK: We have those AI zealots and AI deniers, and it’s kind of coming back to the RPA hype. We see companies starting with RPA even today.
DJ: They actually potentially lost quite a lot of money because usually at the beginning when you start with those technologies, you have a lot of low-hanging fruits. Very obvious processes where you can use a certain technology on a bigger scale which have a higher return on investment.
AK: Do you think companies know how to use AI in their business processes?
DJ: I have actually seen quite a lot of companies that invested in an enterprise AI platform, and since 9 months we are keep hearing from them: “We are still preparing the implementation, we are still looking for first use cases”. I have a feeling we are repeating the cycle. I’m wondering what will be the next cycle that will come after AI – quantum computing?
AK: Quantum computing could also mess around with a lot of things, especially in the security area. We still have companies trying to implement very large AI enterprise platforms, and very often I see these are tools that were created 10-15 years ago, and someone just added access to LLM on those platforms. I’m really surprised that companies don’t do much of a pilots or stage implementations.
DJ: You can already see some hate on AI on LinkedIn, that “we were promised X and AI didn’t deliver”. We have this case of an AI assistant implementation in Microsoft Teams where the solution cannot reply in different languages.
AK: Languages are something that GenAI is really, really good at. When you’re taking a consultancy to build a tool for you, you would expect them to know those things and to be able to test it properly.
DJ: It’s very similar to RPA. The big consultings have been building huge programs, giving you 20, 30, or even 50 developers, and they actually have been hiring people that just graduated from universities and that were learning how to do RPA on your projects. I have a strange feeling that in many situations it might be similar right now.
AK: I think we need to study RPA more to have a better grasp of what’s going to happen next with AI. I like this idea of the cycles going on and on.
DJ: The majority of the market usually stabilizes like four or five years after the first hype, so probably 2027-2028 we should see a similar standardization. I think this one has much bigger amplitude. But the speed of change, how fast all players on the market are implementing new AI solutions, it’s mind-blowing. It’s also a matter of specialization. We might find that there are more LLMs specialized in a certain area.
The best approach: agile experimentation
AK: Looking at the AI technologies, how unstable it is and how fast it changes, I think this agile way, try something small, do a test, do a lab, make up your own mind – this is probably the best thing that you could do.
DJ: In big organizations, you have this Safe framework, and I’ve seen a meme that it’s already 800 pages. How can you be agile if you have an 800-page manual? It doesn’t sound agile anymore.
AK: We have one of our values at Office Samurai: No Bullshit. When people see that you are not trying to sell them something but generally help them solve the problem, it’s a totally different discussion.
DJ: It kind of circles back to the zealots discussion. You cannot be a zealot.
AK: What gets you most excited these days?
DJ: I can quote Daniel Lines here that the future is agentic. This is the most hot topic right now: AI agents, how to build them, how to test them, how to orchestrate the processes where you might have multiple AI agents.
AK: We are now talking about big frameworks: agentic automation frameworks when you can have multiple integrations, multiple RPA robots, multiple queries to SQL databases, and then somewhere between, AI agents doing the work or even orchestrating other AI agents.
DJ: You can compare it to GenAI models that are generating videos or pictures. I guess when we will see a new category when it comes to Oscars – AI-generated movies – then you will be able to say: “This technology is mature enough”.
AK: Just don’t expect them to be easy and kind of out of the box working for you.
DJ: There are those analysts that claim that RPA is dead. I heard the same story in 2015. With AI this is potentially possible, but again it will take some time.
AK: It will also be an overkill in most cases. If you have a process that your automations need to perform thousands of times every day, then doing it the RPA way makes the most sense. Sending the whole screen to an LLM is going to take a lot of energy, a lot of tokens, and a lot of money. Still most of our customers are somewhere using S400 so old IBM mainframe systems.
Starting smart: advice for companies
DJ: For companies that are thinking about what do I do now, what would be your advice?
AK: One thing is making sure your organization has access to some kind of LLMs. Another thing is personal productivity, where you might want to use Copilot, but you also need to invest in education. Then you have areas related to knowledge management, where you could implement assistants. It’s best to use existing fronts like Slack as a front end to your AI assistants or agents. And then you have enterprise agentic automation. It’s more like a puzzles that you need to have.
DJ: Andrzej, maybe I will ask you what’s for you most exciting when it comes to AI?
AK: I think with AI automation for me the most exciting thing is to see all of the failures. The ones that are performed by the biggest players on the market are going to be the most spectacular because if you’re overpromising heavily on the AI front, it’s going to lead to some spectacular failures. We had a few pilots that we had to end on pilot stage, for example, translating what the consultants are talking about with the customers in real time. We need to show both the things that work and the things that do not, because people need to establish a baseline of what’s real and what isn’t yet. Dominik, thank you so much for being here with us today.
Conclusion
AK (Andrzej Kinastowski): And they all automated happily ever after… or did they? That’s the end of this particular chapter in the slightly unsettling fairy tale of automation 2025. Big Arigato for lending us your ears. Huge thanks to our guest Dominik Jaskulski and to Anna Cubal, our producer. Recorded as always at the legendary Wodzu Beats Studio. Until next time, keep your passwords complex and your expectations for a fully automated utopia moderate. Mata ne.
