Who is Robert Niechciał?
- Who he is
- Robert Niechciał, professionally involved in SEO for about 20 years (he jokingly calls himself an SEO dinosaur), CTO at the Vestigio agency.
- What he does
- He rolls out artificial intelligence at the agency and, together with Damian Sałkowski, runs Sensei Academy, an AI training company that trained 1,500 people in Poland within a year and a half.
- What we talk about
- What an AI native agency is, how AI changes an agency's margin and value, why IP matters, whether a human will always be needed in the process, how to overcome skepticism in a team, which models to choose, what "wydajemisizm" is, and how to prioritize rollouts with the ICE method.
- Where to find him
- LinkedIn (Robert Niechciał) and SEO conferences in Poland.
Best quotes
„For me, AI native is not an extension of our operation with artificial intelligence in the form of a chatbot, but a rebuilding of what we do every day in our companies and organizations around artificial intelligence.”
„AI at the end of the day is not responsible for anything de facto, the human is responsible.”
„If you want to roll out AI in your company, stick to one provider.”
„What will win in the world of artificial intelligence in the long run is critical thinking.”
Recommended materials and sources
Guest’s pick
Mentioned in the episode
Related conversations
A Winner’s Mindset in Business: “Competition Is My Fuel” A Conversation with Michał RochwergerMichał RochwergerShared topics: SEO
How to Build Brand Visibility in AI - Maciej ChmurkowskiMaciej ChmurkowskiShared topics: SEO
Working in Teal: How to Build an Effective and Close-Knit SEO Agency - Ewelina Podrez-SiamaEwelina Podrez-SiamaTheme: International growth
The SaaS Market, Tool Consolidation and the Impact of AI According to Łukasz GorącyŁukasz GorącyTheme: AI-native work
Transcript
Host: Rafał Moszkowców. Guest: Robert Niechciał.
Introduction
Rafał: Good day everyone, and welcome to another episode of Globalne Horyzonty. Today with me is Robert Niechciał. A star of the Polish CEO scene, and in fact not only the Polish one. Hi Robert, welcome to Globalne Horyzonty.
Robert: Hi Rafał, it is an immense pleasure to be here, and I am very glad that you wrote me that comment on LinkedIn, because it triggered an immediate response from me and two days later we are here, so we have certainly improved communication.
Rafał: Great, it definitely started going well for us. Could you say a few words about yourself? Where are you from and where do you come from?
Robert: Yes, I am Robert. I have been in the SEO industry almost forever. I sometimes mockingly call myself an SEO dinosaur, because I have been doing SEO professionally for about 20 years. Day to day I mainly focus on the agency Vestigio, where I am CTO and where I am currently touching and rolling out artificial intelligence from every side, but not only that. Together with Damian Sałkowski we also run a project called Sensei Academy, an educational company in the area of artificial intelligence, and I will brag that over the last year and a half we managed to train 1,500 people in Poland, so that is also what we mainly do.
Rafał: Thanks very much. You do many things. I follow you and I will not hide that we have known each other for a while and met on various occasions, and I am really full of admiration, because there are very many things. And about Sensei Academy: one, you trained my team and that is still remembered very well, and two, I hear from people around that it is the only good and sensible AI training in Poland. I think you are doing a great job there.
Robert: You know, Rafał, what characterizes us with Damian, and not only, because we have a wonderful army of mentors who work with us, is that we are not trainers per se, we hold high roles in quite large companies and we do these things for real, day in, day out. And I think that is the difference. Of course sometimes this causes certain problems, because we may have less time to take care of educational material or feedback, since we are responsible for large teams and do a huge number of rollouts and projects, but that guarantees the value and quality of the material we show, because it comes up to date from the front line.
Rafał: You can feel it. When someone knows a bit what it is about and then reads that material or works with you, you can see that you know what this is about, that it is not pure talk. I think it is the only sensible way to run training at all. I wanted to talk about a few things today. I will surprise everyone, and you too for sure: I wanted to talk about artificial intelligence.
Robert: Oh.
Rafał: Such a novelty, right? There are loads of things here that I am wondering about. We cross paths on 150 fronts. I am seriously curious about your point of view. I will refer to your presentation from Kulturalnie o SEO, which some of the listeners have probably seen, heard, and which made some noise in the industry, as usual: Robert walked on stage and made noise. There are a few things I will want to touch on, and later we might go through some issues connected a bit with the future of our business. I am simply curious how you see it.
Robert: Sure. You call it noise, I called it a linguistic dynamizer that allowed that noise to be triggered, but I think you know it was needed.
Rafał: Fun fact: I used the title of your presentation internally, with our team. Meaning: what the hell are we doing. And I sent it saying they should watch it. It triggered an avalanche: Rafał, stop, why in such a way? Listen to it first, we will talk later. Okay.
What does it mean for an agency to be AI native?
Rafał: I would like to start with the AI Native Agency, meaning an agency based, if I understand it at all correctly, on AI. Could you say more about it, how you understand it?
Robert: Sure. We have to introduce a few perspectives on running an agency. Let us start with the classic perspective, meaning the one from a few years ago. What did we do in the agency? We had all sorts of departments and everyone operated on some deliverables, on the client's success, on their own KPI in a given area. And often it was a somewhat blurry operation. What do I mean? Everyone did it their own way, on their various tools. One preferred, say, Senuto, another Ahrefs, another Semrush. Someone used something else and finally it landed in a spreadsheet, and everyone had their own separate spreadsheet. And at a better or worse standard, products were delivered to clients.
Robert: What I mainly meant by drawing this operation: there were many perspectives, much expertise, every perspective and expertise was a bit different, and finally, in a more orderly way or not, the operation worked. How do I understand AI native in this case? I understand it as processes, things that happen in organizations, being standardized around artificial intelligence, where artificial intelligence is not an augmentation of a process, meaning an add-on to a process, but is directly the core of the processes we work in. The simplest use case from an SEO agency is, for example, content planning. There are many paths, but AI in data processing and data classification can help us wonderfully, just not when it is an add-on in the form of ChatGPT or Claude where, again, everyone will have their own, their own configuration, and everyone will get a different result. The point is for it to help in a structured way, consistent across the whole organization, on one hand increasing the quality of the outputs produced, but also the speed of the operation and the operating standard. I understand AI native as an absolute rebuilding of a company, organization, agency or department to the degree that everything becomes arranged around artificial intelligence, which is the core of the processes we carry out.
Robert: We both hold C-level positions, so it is worth looking at technological thinking, meaning how technologically it can help us increase quality. That is one perspective. But another is the human perspective, meaning the number of contexts a human must take on to produce, for example, a content brief: keyword analysis, then analysis of search results, analysis of the competition and many other windows that one has to switch between. I call it alt tab. We alt tab fast, so we change context. It turns out that there can be a great many of these contexts in our organizations, starting from the context of project management, of statuses, of client meetings, communication, through the tool context, meaning Search Console, Google Analytics, Senuto and many other data sources we process to produce a given thing. And that stops being effective. AI is able to connect all these contexts into one coherent picture and let us produce better things. That is how I think of AI native: not about adding Claude Desktop to someone's computer, here, install it, this is your account and off you go, but about centralizing the operation around artificial intelligence, centralizing data and centralizing the whole context in one place. And suddenly it turns out the operation becomes smooth, coherent, standardized. For me, AI native is not an extension of our operation with artificial intelligence in the form of a chatbot, but a rebuilding of what we do every day in our companies and organizations around artificial intelligence. It is an absolute rebuild at every level, both technological and operational, in terms of efficiency and awareness. Everything changes.
How do margin and IP affect an agency's value?
Rafał: Yes. It seems to me that the simplest way I would understand it for myself: if we rebuild an organization into a process that is repeatable, that follows a certain pattern, then we always reduce friction, reduce losses, so we generate a higher margin. And a higher margin automatically translates into the value of the organization and so on. I know that our margin is around 42 percent and I think that, going along a similar path, we will be able to reach 60 percent, because we have those losses already calculated, we see the friction we have because we do not have that central process connected with artificial intelligence. And that is a nice case, maybe someone will be listening, in the context of content, especially in SEO, GEO, or whatever people call it now. I simply call it SEO on steroids. We managed to take the margin we had on content, barely 11 percent two years ago, to over 90 percent.
Robert: That is a huge success on the margin side. I consider it from one more point of view. Look at a classic agency, whatever it does: what is its market, investment, acquisition value? Potentially you have EBITDA multiplied by a dozen or several dozen months, and potentially what constitutes your value in the team, meaning the HR asset, that you have exceptional individuals. The next asset is your processes, which can be replicated in larger organizations. And that momentary revenue you have here and now. At the end of the day it turns out that agencies are poorly valued on the market by investors of all kinds. But if all of this is defined as IP, meaning intellectual property, suddenly the game changes. An agency that until now ground out, I used such a word at the presentation, a time and material model, in which you sat on a chair and sold hours, suddenly becomes a technology company. And there the multiples and valuations are completely different. That is also what AI native is de facto: on one hand rebuilding our organization so that AI becomes the core of processes, standards, increasing margins, increasing EBITDA, but at the same time building intellectual value that works on completely different investment multiples.
Rafał: Absolutely, IP is key. I am glad we are all starting to speak the same language: that we do not build complicated contracts that clients find hard to leave for 19.5 months, and when they leave they have to hand over their Passat in settlement, because we constructed the contract that way. Instead we give actual value, intellectual property, that the client can use and simply will not want to leave. And we know that when we want to sell a stake in our agency, we will sell it through the prism not only of revenue, not only of EBITDA, but above all of IP. I take a deep breath that finally this is how it is being talked about.
Should agencies price by the hour or by value?
Rafał: But I am curious whether the next step after building an AI native agency, and that does not happen overnight, is a change in pricing. Originally we priced in hours, in "ass-hours", let us put it nicely, that is also a beautiful word. I admit that we still price some things that way today, but because we do not yet know how better, for various reasons, organizational and a few others. Only we price it while teaching clients the weight of tokens: our expertise is X hours needed to do this for you, plus tokens. That is interesting, because the first conversations we had with clients about tokens were: Mr. Rafał, man, what are you talking about? An invitation to an internal webinar, we will explain it to you now, and they understand the tokens, some have their own keys, use our IP that is with us, our workflows. But what I was getting at: do you think we will move to value based pricing?
Robert: You know, that will be an autonomous decision of each organization. Why? Because the market is used to the time and material model, to pricing by the hour, if I may say so, so you limit the need to educate the market and explain, you sell in the same model, and everything you managed to optimize is your win. And in support of that I have a very important sentence that Damian Sałkowski once told me. I ask him: Damian, so how much shall we charge for this data, this report? Because it is half an hour of work, so I do not know, a thousand zloty? And Damian says: Robert, but for us to have the infrastructure and technology, an investment of several million zloty went in so that we can spit out this data in an hour. So it cannot be priced at a thousand zloty, but at ten, because over the years we made huge investments and developer resources to build the technology that lets us deliver this in an hour. And I think that is a good key, because the client at the end is interested in the result and what you deliver, and how you arrange your processes internally is simply your win. So there are two paths. Either we are more efficient operationally and cost-wise, so we simply earn more, we continue the current model, because there are also old contracts from years ago that will probably be hard to renegotiate and the market is used to such a model, and that is okay. Or value based, absolutely, then the client pays for the concrete result of that technology, and at the end of the day we will meet at similar finances de facto, right? Maybe.
Rafał: Maybe. I am exploring this value based approach in various ways and I am getting ready for it, which is why I asked, because I think it is a natural consequence of being AI native. Gosh, exactly what Damian told you: thanks to our twenty-year investment, because I have also been in this industry almost that long, many sleepless nights and read scientific paper documentation, we are able to do something in half an hour. I even had such an example this morning: optimization on a huge scale using agents, of very many stores, platforms. And it will take us a week. Normally it would take about two years. Leaving aside the fact that we already know how to do it, the client would have to pay five, six people for two years to do it. So I am still hesitating how to dress it, how to bite into it. It is a bit new and I am also looking for some consulting, but I feel that for agencies that will truly be AI native, that is the direction.
Robert: Definitely, only that is probably for the future, from today to the next contracts, and you have the baggage of existing ones, so you have to weigh it very wisely and calculate. But of course value based today is the best model of all.
Rafał: Yes, no point going crazy, we are not going to plow through all the projects we have now, because that would end in huge churn.
Robert: Exactly.
What is human in the loop and how to deal with AI sceptics?
Rafał: I am still wondering about one more thing, because when people listen to this, an AI native organization, there is a feeling that it is soulless, that there is practically no human there. At least I feel that way, and when I talk to a few people around me, they have such feelings too. And you say there has to be a human in the loop, meaning that a human has to be in the process. Do you think this is a transitional stage and in a short while we will reach a stage where that human will not be in the process, or will there always be one?
Robert: As of today I am an advocate of there always being a human in the process. Of course I do not want to predict what will happen in the future, because I believe that in two or three years we will be talking about completely different agents, completely different language models and completely different agent memory, so a lot will probably change. But I think about it mainly in the context of responsibility, because thoughtlessly spitting out any product from artificial intelligence can lead to a crisis situation in business, because bad recommendations went out or something went wrong enough to cause financial turbulence on the client's side. I think that AI at the end of the day is not responsible for anything de facto, the human is responsible. And a human takes responsibility for the effects of a given process and signs off on them. I believe a human will always be needed in the process, if only to give feedback on the process. Because if we have any linear process, step by step, it is nice to have so-called feedback. It is good for that feedback to come also from our organization and not only from the client, in the sense that the result of the work is of low quality, because that can happen. That feedback in the form of a human will improve the process and take responsibility for the process and its result. So I think long term about making work pleasant for my people, but at the same time increasing their capabilities and developing them thanks to artificial intelligence, and not replacing them with yet another agent that will scrape something out.
Rafał: That is an interesting perspective, because I wonder how you deal with the mentality of people who approach solutions related to artificial intelligence with skepticism, to put it nicely. I do not know whether you have not met such people around you.
Robert: Of course, very many.
Rafał: It happens to me constantly. I do not like to be skeptical about anything at all. Honestly, I prefer to assume it will work out rather than not, but how to fight it, that is the wrong word, how to root it out, because it seems to me it may be our biggest blocker.
Robert: Yes, it is one of the biggest blockers, because we as visionaries already see it, but in organizations there may be many people who are simply afraid, and they are afraid because they do not understand anything, so it is a bit of a fear of the unknown. Organizations that actually want to adopt artificial intelligence, it does not have to be AI native, it can even be AI augmented, in the sense that a process is supplemented by artificial intelligence, may collide with the problem of people who will push it away out of fear. How to solve it? First of all, train, show how it works. And the second thing I did: I ran a company-wide feedback round where I asked people about the processes that take them the most time and cause the greatest frustration. Such weekly, monthly processes, for example project management or preparing for meetings. And what is the level of frustration? I think that if we approach adoption and breaking down the barriers people may have in such a way that at the very beginning we, as company boards, deliver instant gratification, meaning they quickly feel that things are simply better and easier, meaning we solve their smallest frustration, let it be management in Trello or Monday. And they feel here and now that it works.
Robert: And a second time they feel some other instant gratification, for example that they do not have to prepare for meetings with clients for 2-3 hours, but get weekly briefs about what is going on and what is planned. That is another instant gratification and step by step they will simply become convinced that this is not a replacement, meaning something that will potentially take their job or something they must fear because they do not understand it, but that it will start rewarding them, and the more rewards, the more desire for more rewards. So okay, now let us optimize this. That is the path I took. But a very important thing: it has to actually happen. It is not that I come back from somewhere in the world, I am pumped because I rested or because I sat on a plane for 15 hours and thought about it, and I come to the team, announce that from today we are doing this, and disappear. Once there is a promise that we are optimizing processes, eliminating frustration, it has to happen. That is the first step.
Robert: The second step: you treat a head of department differently, you treat the board of a company differently. When something comes from above, well, we eliminate frustration and there is a plan for it, but when my colleague, my partner, with whom I work every day in my branch, already uses it and has nice results, then I want it too. So it is important to have evangelists in the company, colleagues who already do it and share their results. To create in organizations, I call it the AI Forum at my place, a space for people where it is not me bringing: hey, I did this, but where they show each other: hey, I actually enjoy managing meeting notes thanks to this and it pulls out summaries, I wanted to show you because I built it for myself and it works nicely. That starts to function differently at a peer level, where a colleague shows another, inspires and also breaks the ice. You have to create a space for people so they can talk to each other, because we are in such a chase after the deadline, the month, the deliverables we have to deliver, the KPI, that sometimes we cannot stop for a moment and talk. It is worth making such a space. And step by step supporting each other, that is certainly a good key.
Robert: But a situation can occur where someone will be very resistant and will not want to. We know how that gets handled then, right?
Rafał: Euphemistically.
Robert: But still not on the basis that we do not give chances, do not inspire, do not show, do not make space and mutual challenge, hackathons, brainstorms, meetings and solving frustrations. That is the basis, you have to start there. But a situation may occur where someone will actually be resistant, and then the decision will simply be made differently.
Rafał: We do not want to scare anyone.
Robert: No, God forbid, it is not about fear. I believe some people may have various emotions and one should support, fight, show, but life is life.
Rafał: I agree with making that space. I think I also have a lot of things I do not use in terms of actually creating space for people. Sometimes it seems to me, and I admit it right away, that someone who is in this industry should pull toward it on their own: they may not know, may need training, may need support, but they must have their own drive toward the goal. That is my belief, which at best has been outdated for a while now, but sometimes I think: gosh, how can you not want to do this? Especially in the context of the SEO industry or software development, because there is that nice film "Don't Look Up" with Leonardo DiCaprio. I do not know if you know it.
Robert: I know, I know. Idiocracy, right?
Rafał: I think so. I overuse that comparison, but for me the changes in our industry, or in software development, content marketing, in marketing generally, AI is a kind of tsunami coming at us, or a huge asteroid, however we want to imagine the threat. Some people simply decide they do not see it, or the glaciers are melting and they decide they do not see it. I wonder what else we can do. Do you think it is worth investing more in human skills in organizations? Most often in SEO organizations, in my experience, there are no communication specialists. Is that perhaps a way to invest more in that human side, to have a communication specialist who will try to manage this in such a way that people want to try and do not give up?
Robert: Is that not called HR?
Rafał: I think not, honestly. Perhaps it can be called HR, but it seems to me HR does not do it, even if we can call it that: they generally do not improve communication, do not build those communication bridges, do not unify naming in the company and many such things that are segmented, looked after in a fragmentary way. At least that is my experience.
Robert: I think it is a fantastic idea. And if you are an organization of 20-30 people, I would guess you do not have the resources. Someone in R&D or in another position will always be more needed at that moment than such a communication specialist. Unless it forms organically inside, you will have that evangelist or leader who will organize it, then absolutely. It is probably a challenge for large organizations and there I see high justification for such a unit, because such organizations will face the biggest problem and potentially the biggest cuts will be there too, if their boards decide so. But in the end: create space, support, help, solve frustrations, show at every step how it works. If we are able to describe it with some efficiency metric, then definitely yes. And I believe sooner or later all the ice will break, because concerns, apart from those strictly about AI taking our jobs, may also be that the technology is not with us yet, that models still hallucinate, that the quality is not what we expect, and those are legitimate concerns. But you see yourself how quickly it is developing, as your listeners probably also observe this market, and those quality concerns are being resolved day by day.
Which AI tool should a team choose: ChatGPT, Claude or Gemini?
Rafał: I would like to sum up that I also understand these concerns, although sometimes I think it is enough to get informed to see what is happening on the market and those concerns start to be much smaller. But I understand that then FOMO appears, because that is what I mostly have: I wake up in the morning and already feel I am late. I do not know if you feel that.
Robert: I definitely have FOMO, though it is probably expressed more as I could do more, faster, harder, and agents can work while I sleep. I try to cut myself off from communication like X, it is a bit toxic from the FOMO perspective. But it all drives that you see another second brain, another solution, another model, another amazing technology, and also some people who circulate AI news, because such a group has formed: everything is amazing, everything is game changing and outstanding. And it happens every day, meaning every day is game changing. So gosh, where will we be in a week, right? So in all this you need common sense, and what will win in the world of artificial intelligence in the long run is critical thinking. I try to stay in the middle. The FOMO pulls me a bit, but I also build for myself not a to-do list but a list of things I do not do, a not-to-do list, and that is how I manage it. But my excitement is enormous.
Rafał: I recently heard from Radek at my company, because since I started using Claude permanently two years ago, I use Claude for everything. He says: God, I admire you so much. You find a tool, start using it and keep using it, while he has very many tools and constantly fixes something there and moves around. So I think that in the context of using artificial intelligence it is worth sometimes welding yourself to one tool, which goes forward by itself anyway, because then you will know how to use it better.
Robert: Especially when you want to transform an organization. Every one of your listeners who in any way wants to roll this kind of technology out to a team has to decide whether it will be ChatGPT, or Claude, or Gemini, or anything else, because once you roll it out to the team and suddenly something new comes out or you change your mind, you probably start from the beginning, right? From the perspective of Radek, a technical person, your business partner, and mine, we have to search, connect dots, even draw inspiration from various solutions. From the perspective of the organization and the team it is a long-term decision. Can you imagine rolling out, say, HubSpot in your organization, working on it for a year and suddenly deciding: not now, now we are doing something else? See how painful that can be even at the level of migrating data, not to mention having to learn to work in it. It is a very important decision, so once more, really: if you want to roll out AI in your company, stick to one provider.
Robert: And what I can reveal to you, I think this can once again also break the ice and it seems I managed to: it is not that we buy one account in ChatGPT or in Claude or in whatever tool you decide on, rather everyone must have their own. So that they have that personal touch, that it is theirs, it is safe, their data is there, their memory, their information, their conversation history, nobody has access and it is private and only theirs. That gives a completely different feeling toward using artificial intelligence than in a company where there is one account for 15 or 20 people. There it is a mill and it certainly will not work that way. I know it is a correspondingly larger investment, but it is worth it if we want to approach this adoption seriously.
Rafał: I think that is nice and interesting advice. As you were talking, I started thinking that we have a shared account somewhere on ChatGPT, while everyone has their own Gemini, and I behave completely differently with Gemini and completely differently with the chat where I know someone may see it later.
Robert: Exactly. And there are also aspects of memory, which is developing wonderfully when it comes to assistants and agents, where with each iteration the environment you work in remembers your previous iterations, so potentially it gets better and answers you better. It is yours. It learns you, how you consume answers and what you expect. Once again it is a personal touch and that also lets you work with it better.
Frontier, open source and Chinese models: what about data security?
Rafał: Robert, tell me, because we have already started talking about specific models, Gemini, Claude, GPT: do you have an opinion on Chinese models, open source models, is that the future of this kind of use, meaning not paying only for a license?
Robert: Let us say that the environments you named, meaning Gemini, ChatGPT and Anthropic as Claude, are so-called frontier models, meaning absolutely the best solutions, the best models, and they will always, at least in the near future, be better than Chinese models, although China is pushing in very nicely and catching up. But the frontier-class models do have that advantage, so for our colleagues, for our team, I would rather suggest working on these three environments because of their high sophistication and the fact that they are always, at least as of today, ahead of the Chinese models. Because then a person feels it really works, rather than, for example, it drifts, hallucinates or suddenly some squiggle appears. But I do find huge use for Chinese models, or open source models in general, in processes that a human does not see, meaning for example systems we build in our company where there is some kind of, I am guessing, classification, summarization, or things needed in a specific business use case. There open source models have exceptionally good use. On both the reasoning side and the context window side, meaning the size of data we can put into these open source models, it is a race after frontier-class models, but the quality is still somewhat lower as of today. I say somewhat, so they can safely be used for backend processes.
Robert: What remains is the matter of infrastructure. If we buy tokens or subscriptions, we no longer care, we have the infrastructure and it simply works. Of course we pay for that comfort that it works and is scalable. As for open source models, we either have to host them somewhere, meaning go into hardware investments, where we see there is a problem now even buying a Mac Studio with 512 gigabytes of RAM, as of today it is probably hard to buy, or look for alternative infrastructure providers. There are such providers, but their performance is correspondingly different from frontier-class models. And what does it come from? The biggest companies are so overinvested financially, they have infrastructure that is not even theirs de facto, for example Amazon supplies gigantic resources as an investment, so they can sell those tokens very cheaply. Open source models do not have that benefit of artificial infrastructure, I say artificial from the perspective of artificial dollars that finance it. So those infrastructures are correspondingly worse. But if we can find a suitable provider who will host it for us and a model that meets our business use case, then why not. I can recommend OpenRouter, you can find it on Google. There are practically all the open source models that matter available there, plus infrastructure providers you can use. It is also worth saying: if we have very complex or mass processes in our organizations, for example processing a million products in our online store daily or weekly, Chinese models can effectively optimize monthly costs, which is not insignificant at large scale.
Rafał: There is one more issue here that you have probably heard about more than once, data security. There are organizations, there are people who are afraid of sending any information, for example, to ChatGPT. I think rightly so, sending contracts, documents, you probably should not do that. And I often hear about using open source in the context of: download, install, lock it down and use.
Robert: Definitely, if an organization requires a high level of data security, then that is probably the only option, because at the end of the day, if you use external infrastructure, even open source, you are sending it somewhere. The only option is to host it yourself, unless you anonymize the data in such a way that it does not matter where we send it, but I do not feel that could work. Open source is definitely a solution.
Rafał: I think so, but we also keep it locally, because there are various ISO certificates, data security ones, that de facto make it impossible to fully use external language models, and I think that is a matter of good business, meaning implementing solutions inside medium and large companies that will simply be secure.
What is wydajemisizm and can AI solve it?
Rafał: Okay, but let us go on. I wanted to press you a bit more on your, my favorite word, namely wydajemisizm. Robert, you are subconsciously a linguist or a Polish philologist at the same time. You create words that stick in my memory.
Robert: Another one already, it seems. Lately I walk the dog a lot and have various strange ideas, and I think that one came about recently.
Rafał: It is a good word, polysemous, so semantically I think usable. But seriously: do you think AI is the solution to wydajemisizm? Today, and you also talked about it at Kulturalnie o SEO, but it is not new either, people evaluate content. We as SEO specialists, our teams, meet this every day: I like this content, I do not like this content. The reasons for this liking-ism or wydajemisizm are most often arbitrary. I wonder what you really think. Is it that properly prepared AI is able to eliminate this wydajemisizm?
Robert: It seems to me the problem appears at the level of initial education and of why we are doing this. Because as I joked at Kulturalnie o SEO, the people who evaluate it are often neither Polish philologists nor linguists and we need to start from the goal function: why are we doing this? Of course I understand the goal function, which is marketing and sales, and then we feel the vibe and it has to be pleasant, nice, tidy, and I allow wydajemisizm here, because everyone can have a different tone of voice, different expectations. But if the goal function is directly exposure in AI search, or topical coverage, or high positions in Google, we have to pay attention that besides our vibe there are algorithms, there are content processing procedures that follow specific rules. I think that is the problem: I understand, I assume you also understand how it works, but the people evaluating content may simply have no idea, for example, that a sentence has to be constructed in this way because it is unambiguous and saturated with information so that it has a chance of being cited. So if AI makes us a construct that can be counted, checked, that it has the right to be cited by a search engine or chatbot, and the evaluators know why it is so and what it results from, then it seems to me wydajemisizm will disappear. Plus there is the matter of several case studies and presenting the results of this approach. I encourage everyone to think about it and not to judge content in advance in your own way, especially content prepared for AI systems or AI syntheses, but to ask why, why it looks like this.
Robert: If someone can argue what, for example, entity salience is, meaning drawing attention to a specific entity, the exposure of the entity in this text will explain why it looks this way, or what semantic role labeling is and why a sentence has to be constructed that way for a search engine, then wydajemisizm becomes smaller and smaller. And these are hard things and this is the place where artificial intelligence can explain a lot, because to get through knowledge, especially linguistic, semantic, there is a set of concepts that are hard to digest. Based on my course, a colleague made a semantic encyclopedia, there are 280 technical concepts. So I believe it is quite hard to understand at all what this is about, but we cannot be offended that engineering knowledge is engineering knowledge. AI is able to explain it in a very simple way. I recently watched a podcast with Andrej Karpathy, who said that this is how he builds his education and knowledge for his collaborators, so that every agent can explain what it is about. Every AI agent will understand what a semantic frame is. It is an old linguistic concept from probably a hundred years ago, but it is hard to explain to someone directly, and when you throw it into any AI agent, it surely knows it. I had an example: once I wanted to push the concept of a semantic frame into some process and started explaining to the agent that a semantic frame is this, this and this. It says: man, I know what it is, you do not have to explain it to me. Five minutes later it was implemented in the solution. That is how it is easiest to reduce wydajemisizm, but it has to be cooperation from both sides.
Robert: Since we have mentioned the semantic frame, I will explain what it is. A semantic frame is something such that at the linguistic and language level, if we have the word "bank", everyone in our brain immediately connected bank with money, with an ATM, with a bank branch, with a loan and with a few other things that bank is linked to in our brain or at the language level. It has high significance from the perspective of building content, from keyword research, or building the structure of a site, solely on the basis of knowledge from a hundred years ago. And once you explain it to someone, it is already a bit better, right?
Rafał: Definitely, so education, education and once more education.
Robert: But it takes two to tango.
Rafał: I agree. Some people do not want to be educated, we also have to say that honestly and then you have to let go, because there are such people too. You reminded me, speaking about that frame, of a quote I wrote down: "You shall know a word by the company that it keeps." That is the quote for this linguistic concept, that really you can know a word by what surrounds it.
Robert: But think about it, Rafał's listeners, how you learned English. You had, for example, some context, a blank in a sentence and you had to insert a word. You thought, based on the left side and the right side, which word fits there. You certainly did such exercises. And another exercise was, for example, the word "dog", which you had to explain to your classmates using other words, meaning circling around the dog so they guess the word "dog". That is how English is learned. If you look at it from an algorithmic perspective, from the perspective of LLMs, LLMs were trained in the same way. All those who surround themselves with wydajemisizm will open up a bit to such simple concepts they had in high school or at university when they learned English, and you can explain to them what this results from.
Rafał: Great reference, great comparison. I would not have thought of that. You gave me another argument for when I talk to someone who suffers from wydajemisizm. So de facto we can say that what is happening now is a return to fundamentals, in the linguistic context.
Robert: Wait, I will derive for you what it is a return to. Artificial intelligence models were trained on language corpora. Language corpora in every language are constructed so that words occur next to each other in some context. You say, for example, "hot", the next one with high probability will be "dog", as "hot dog". Even in the example of this English lesson, where one word is missing in a sentence, the model guesses what the word is, meaning it predicts the nearest token with the nearest probability. If language models were built on a language corpus, meaning on our language, then our answer, if we want to optimize, is identical methods, because then we align with what that model expects, because that is how it was built. It was built on the semantics of a language corpus. If we approach the problem that way and explain why this word has to be used in that context, or why the sentence structure is built that way, because for example some retrieval score is very high thanks to it, then we have a strong argument for further discussion and for education. It is really very interesting if you think about how it came about, how it was built and how we can respond to that particular architecture.
Rafał: Thanks for that, a very good explanation. Some episode of the Acquired podcast started rattling around in my head. I do not know if you listen to that podcast.
Robert: No. Lately I look for inspiration elsewhere, but if you have something to recommend, I will gladly take it.
Rafał: It has the history of Google among others, although lately I listened to the history of Ferrari and that is great too, but the history of Google in two episodes is about six hours, and since you walk the dog for a long time, maybe you will find the space. It is a great story from the very beginning, with the history of OpenAI woven in, but really it was Google that invented today's LLMs, and it is simply a really well-researched story. Maybe I will link it, and I will also send it to you later. I really had a great time.
Robert: Since you have told about it, I think you should link it under this podcast for everyone.
Rafał: Okay, I will do that, because it really is worth it.
How to prioritise AI rollouts: ICE and instant gratification?
Rafał: Slowly heading to the end, I wanted to ask about one more thing. We said a lot about the agency, about what we are, what we are not, what we would like to be, how we work with people, how to introduce this. We touched on history, on semantics. And we get to how to set priorities for rolling out these innovations, if we can call AI rollouts innovations. I know you know and use, if I understood correctly, ICE, meaning Impact, Confidence, Ease, which is unpopular in the States.
Robert: Rather popular in the States.
Rafał: Can you say more about how it works for you at Vestigio?
Robert: The problem is that we have to manage what I already said: not shiny object but instant gratification, so that our team quickly gets a reward from rolling out artificial intelligence and keeps following the innovation. There can be many initiatives to optimize in your companies or teams, starting from project management, through meeting notes, content briefs, product descriptions, or whatever it is. I recommend asking in the team: what do we even do? And then: what of that do we optimize? And suddenly it will turn out you have a list of 20 or 30 things to optimize, but not each is important enough to prioritize, because resources are limited. This ICE framework, meaning Impact, Confidence, Ease, helps manage that. Impact, meaning how a given process will directly affect our organization. Confidence, meaning how sure we are that it will help us. And Ease, how easy it is to do. We describe, for example, the challenge of optimizing the project management process: impact will be high, say nine, because it is the foundation of how the organization functions. Confidence we are sure, because it is high, say nine. And Ease is not the simplest in the world, so let us give it, for example, seven. Now we multiply all those numbers together and it gives us a final value.
Robert: I also suggest describing it by the frustration involved and whether a developer is required, or any big software investment, or whether we can automate it, for example, with skills in Claude or n8n, depending on what technological resources we have in the company. In this way, describing in a table all the initiatives reported by our team, priorities suddenly line up: what makes the most sense at a given moment and in what order we should roll it out. Thanks to that we are sure that instant gratification will happen. Otherwise, if we wanted to prioritize as a business any innovations, we would probably get a list of what our team would like to do, we would calculate what gives us the biggest return on investment fastest, where we earn fastest, and we would prioritize that, so it would probably be marketing and sales. But that does not build instant gratification, the team will not feel it immediately and perhaps it is better to do it in a framework such as ICE, there are probably more frameworks. It is generally about prioritization and about closed cycles, to measure whether we have actually developed and done some rollout to our team and whether it works, whether it actually functions. And in closed cycles, if we step off the roadmap, simply asking for more, iterating over the process and constantly setting that ICE in, for example, quarterly cycles: what next, what next, what next. That is an idea for how to achieve success in adoption and prioritization.
Rafał: I think it is a quite simple method, but it can be very effective, because indeed we start, and I am guilty here, rolling out where there is the greatest potential for ROI, for savings or for opportunity. As you were talking I started to see that there are places in the organization that frustrate even me, and I do them and think: Jesus Christ, I really do not want to do this, and I never made it a priority to automate it or at least strongly support it with a process managed, for example, by an agent. So it is worth bending over this.
Robert: Definitely, especially since, once again, as we already talked about creating space for the team and for our colleagues, that space has been created. We have a board for it, the interested people at the moment, in our team that is two team leaders, one colleague who helps me in R&D, me and the operations director, we work on it and set subsequent sprints. At the next AI Forum or the next status it is clearly visible what we are clearing, what already functions in the organization, what does not, what is next, and everyone knows it is simply moving. It is not a shiny object or some CTO project where I came up with something, but a living ecosystem of rolling out innovation, and for now it functions somehow. I estimate we may have the breakthrough rather at the end of the second quarter, because when you start adopting artificial intelligence you have a transitional period in which you have to build a lot to feel it. We are already coming out of that period in which we built a lot, and we will only start to feel it in the coming months, but that is normal. In agencies the cycles are monthly, so if I did something yesterday, it may only be operationalized next month and the mechanism will start to drive itself, the effects will start to show up like a snowball.
Rafał: Have you thought about measuring it? Or maybe you measure it somehow?
Robert: There is too much of all this. I focus on development, not on building a metric for the metric.
Rafał: You know what I mean, because in the context of eliminating frustration I second that idea. I think a frustration index should also be some method that you develop and sell, because it is well thought out, especially with such large amounts of change. But I also think globally that it would be nice to have some eNPS or something similar to simply measure employee satisfaction. Because at Vestigio there are several dozen people, if I remember correctly, at my place there are 20, but there are organizations with 300, 400, 500 thousand people, so they have to somehow measure it vertically, horizontally, and eNPS seems a nice solution here.
Robert: A great idea, I have not thought about it in my specific case. But certainly at the next ICE cycle we will be able to do a summary of what we have already managed to do, how we are working and where we go next, and with every iteration it will be more mature.
Where will we be in a year or two? Contact Robert
Rafał: I am mega curious, Robert, where we will be in a year, in two years. Really, when I hear what we are talking about today, I am full of excitement. Where will we be?
Robert: I cannot wait and I am, as they say, very bullish on growth in the coming months and quarters. I cannot wait for what is next and for the effects of this work. This is a beautiful time for me and I really wake up. Having been in this industry for so very long, at some point there was a lot of routine in the operation, and at the moment I feel every day that I am doing fantastic things and I cannot even afford to waste a day, for example on the proverbial hangover, because you have to push.
Rafał: So healthy habits produced by artificial intelligence. So we do not drink, because it is a waste of time.
Robert: We do not drink, because it is a waste of time. We sleep, because you have to sleep, and we go to the gym, because then the head works better.
Rafał: Definitely. I am going to the gym in an hour. Robert, if we wanted to find you, or our listeners wanted to find you, where is best?
Robert: The best form of communication with me is LinkedIn, of course. If someone prefers Facebook, that is okay, but I would prefer LinkedIn. And of course I will gladly meet your listeners if they visit SEO conferences in Poland, any, there is a high probability I will be there and I will gladly talk in the corridors.
Recommendations and closing
Rafał: At the very end I would like to ask you for one or two recommendations, whether a book or a podcast, because we said a bit about that, but I would like, like an LLM, to synthesize a bit and give listeners two, three suggestions at the end.
Robert: Recently, on a walk with my dog, I listened to a fantastic podcast by Andrej Karpathy, whose title I think is "We are the bottleneck", where he talks about how we are the bottleneck in the context of agent orchestration. It opens your eyes nicely to the broader view of an engineer who stands directly behind OpenAI and to what AI is today, and introduces many interesting concepts, for example that we are in an AI loop world, where AI develops AI itself and autonomous processes. A very interesting, somewhat engineering, but opening podcast. And the last book I read and which I think is worth recommending, especially on the Polish market, is "The 5" by Tomasz Karwatka, who nicely tells his story with Divante and beyond, and also introduces these concepts of building intellectual value. A very interesting title. I ground through it fairly quickly on a plane, the whole book cracked open in both directions to and from Vietnam, so I think it is recommended for people who are building an organization, especially now.
Rafał: A hundred percent. Tomek Karwatka should kick something back to me, because whoever I talk to, everyone says: gosh, in the end you have to read this book. Everyone who is in business, or SEO, or in any technology. I recommend this book too. I read it literally two and a half times and decided it was too much and I had to let go for a while, because I first read it on Christmas Eve, before the borscht, because I got it the day before and simply could not resist. It really reads well. I had the impression it was a bit of a story from my life, only I did not yet know many things that Tomek knows, and he tells them wonderfully.
Robert: From the Polish scene in this area of interest, a very recommended title. I enjoyed reading it and what fascinated me was that some of my thoughts that I had had earlier I found answers to in this book, for example on intellectual value, or how to build the value of an organization through intellectual value. When I was building the strategy for this year for my companies I had not read this book, I only read it a month ago, and I did the strategy in November, and somehow it all started to mesh. Nice, I recommend it.
Rafał: Definitely. Let me add one more word from myself: I had the book ready, started reading it over the holidays, could not stop thinking about it, and a consequence of reading this book, among other things, was that, in the context of the NON.agency agency, I decided that first we focus mainly on large publishers and e-commerce, because there we can actually do large scale and large solutions, and second that we are giving up all the side services, performance, webdev and such things. We threw it all out and after the first quarter I no longer have any of those additional services and our processes have simplified so much that I can already see how easy it is for us to build IP, because we build that IP in one direction. I really recommend it: one, the book, and two, to then start implementing it, especially since it is written so that I felt I had thought like that before, but I was missing the kind of confirmation that it is well thought out.
Robert: And once again, focus, and what we said today on this podcast is a not-to-do list.
Rafał: Exactly, so as not to go crazy.
Robert: So as not to go crazy.
Rafał: Robert, thank you enormously. As always it was a huge pleasure.
Robert: Thank you, thank you, Rafał. I hope we will meet face to face soon and talk longer about our topics.
Rafał: Over tea, of course.
Robert: Of course, it is a waste of time to be hungover.
Rafał: Sure, all the best. Bye.
Robert: Bye, thank you.
Listen and watch
The same conversation in your favourite app.
The full edited transcript as a Markdown file, with the guest’s details and quotes.
Back to top

