Most AI Tools Won't Survive
Damian Ślimak on twenty years of automation before it had a name, why clients never ask whether AI is involved, and the one skill still worth training.

Twenty Years of Automation Before It Had a Name
In the sixteenth episode of the CMS Conf Podcast, Damian Ślimak makes a claim that sounds like tool skepticism and is really a statement about where durable value sits in a system. He has spent close to twenty years around online stores, and for most of that time the work was importing products from a manufacturer into a shop, pushing orders back out, and converting files between formats that refused to agree with each other. None of it was called automation then, and none of it was called AI. The job did not change when the vocabulary did, and he has watched the interface to automation get replaced three or four times while the problem underneath stayed exactly where it was.
If you run an online store and suspect that a measurable part of your team’s week disappears into moving data between systems by hand, Damian’s company Nawigator.ai is the direct route to fixing it. The work sits on the unglamorous layer of e-commerce: product feeds, order handling, and the repetitive operations that quietly eat staff hours. Check what they do at nawigator.ai and bring your worst manual process to the first conversation.
The Toolbox Is Deliberately Small
Damian’s working set is short. Gemini, Claude Code, Obsidian, and a browser. His argument is that the entry cost of any one of them is a few hours while real depth takes months, so breadth buys very little and depth in a single product costs everything else. Obsidian holds his files as plain text on disk. He avoids tools that turn an entire meeting into a document, because what he wants out of a meeting is two or three lines. The browser is Arc, no longer developed, kept because its profile switching still beats the alternatives: university work, one client, private matters, one toggle, an entirely different configuration. Anything he uses once or twice a week gets tested against free alternatives and written up on his site so other people can skip the testing.
The Constraint File Is the Real Artifact
The clearest illustration is a snail. Damian’s mascot, Wolniś, lives in an Obsidian folder alongside a written universe: a neon-lit technology city called Neonowo and a village called Hamakowo, described in plain, consistently formatted text files. Generation runs through Gemini. For about a month the images came out random, because the model kept giving the snail a shell even though the character is humanoid, wears a hoodie, and has headphones on. The fix was a single line at the top of the most important file instructing the model never to add a shell. After that, “Wolniś gives an interview” became a complete prompt. During the recording he typed exactly that and got three images back: same character, same clothes, same startled pose at the computer, same city.
The style was chosen the same way. Rather than asking for a snail that looks great, he generated a list of roughly thirty of the best known comic styles and rendered the same character in each. A 3D version came out looking like Paw Patrol and a Simpsons-styled one sat too close to Homer, so most of the thirty were discarded and one fit. The portable part is the ratio: fifteen minutes gets you a working result, and five or six months goes into the constraints that make the output yours. Damian ties that to something visible across the web right now, which is that most newly published sites and event posters look interchangeable, drawn from three or four repeating layouts. The constraint text survives a model swap. The model does not.
The Adoption Problem Is Not Technical
Damian works mostly with companies of ten to twenty people, and describes implementation as two separate arguments. The first is with the owner and it is easy, because licences run around twenty dollars per person per month. The second is with the staff, and that is where everything stalls. His example is a company where three employees spend entire days preparing tender documents, which is largely clicking. Two tenders in eight hours becomes roughly one hour of generation plus verification. Those three can do that arithmetic as fast as anyone, and what they see at the end of it is one remaining seat.
He does not dismiss the fear so much as redirect it, arguing that the aim is more output and higher pay rather than a smaller team. But the resistance he describes as harder to move is quieter than job anxiety. Someone whose day has been identical for three years, two hours of email, three hours of something else, breakfast, done, is being asked to rebuild the routine from scratch. Teaching the tools is not the hard part, because the tools are simple now. Getting people to want to learn is.
On outright replacement he refuses to forecast. Repetitive preparation can go: research for a podcast interview, reviewing what a guest has said before, drafting a question skeleton. Conducting the interview cannot. The same applies to engineering, where the number of people writing basic functions can fall while the architects stay.
Clients Never Ask Whether It Is AI
Asked whether his e-commerce clients understand how much of their business is already automated, Damian answers that they do not care, and that this is the correct state of affairs. You own a phone and have never wondered how its processor schedules work or how files get written to its storage. Someone who buys Canva wants the background gone from an image. Whether that happens through a model or a generic algorithm assembled out of ifs and switches makes no difference to them, because what they paid for is the outcome.
That leads to a harder claim. In most places where AI is currently being inserted, it is either unnecessary or actively in the way, added because it is fashionable rather than because the product needed it. Calling AI a feature is a marketing decision; the accurate description is an implementation detail.
Obsidian Is a CMS, and the Site Runs on Astro
Asked for a favourite CMS, Damian answers Obsidian, and he means it literally, since a content management system manages content. For client work he reaches for WordPress most often while openly disliking it, calling it a heavy machine with too much clicking, and he keeps the skill because a few jobs land every year. PrestaShop is where he specializes and where most of his store work happens.
His own site runs on Astro, and the reasoning is architectural rather than aesthetic. He builds locally and uploads compiled files, so the site is fast, nothing is running on the server to break into, and the output is HTML. That is what lets him say without embarrassment that he has not written a line of its code and does not read it. He wanted a tag system where one tag rendered green instead of blue, so he dictated that into the computer, pressed enter, made coffee, came back to a wrong colour and buttons that were too large, and corrected it. Several sessions of that produced the site. The position holds only because a static build has far fewer ways to break than a database-backed application carrying plugins.
The Skill Worth Training
Asked what the industry should be training so it is still relevant in three to five years, Damian answers with one word. Thinking. Connecting information, designing the flow, and verifying the result stay with the person, and everything else is substitution. WordPress basics take about an hour, expertise takes years, and WordPress could be displaced next week by something faster and better looking. Anyone who invested in the tool is left holding a tool. Anyone who invested in the process moves it onto whatever replaces it.
The incentive is worth naming, since Damian sells process training and runs workshops, and a position that makes tool expertise perishable happens to make his product durable. It is also the position his own twenty years support, because the feed importer, the format converter, and the current constraint file are the same job wearing different clothes.
It is consistent with how he plans to spend CMS Conf, which is in the corridors rather than the talks, on the grounds that an eighteen to thirty minute session can outline a topic and no more while thirty minutes with the speaker over coffee gets the specifics. He flags that as a preference rather than advice, and he does sit through workshops, where the longer format has room for detail.
Damian is building a free Gemini course on his site right now, four lessons published and at least ten planned, all of the introductory ones free, structured around a freelancer running five clients who needs a daily graphic idea and a weekly marketing plan for each of them. It is the compressed version of what he will run as a full workshop on the Polish track at CMS Conf 2026 this November, where the same pipeline gets built properly, from a configured environment through to publishable content. Start the course now to warm up, then book your workshop seat before the track fills, because hands-on rooms are the first thing to sell out. Tickets and the full agenda are at cmsconf.com.
Tags
Keep Reading

Or maybe a warm-up before Gdynia? Join us at CMS Conf Local in Vilnius
Before we meet at the main CMS Conf 2026 in Gdynia on November 12–14, we’ll see you on October 24 at CMS Conf Local: Vilnius.

From Browsers to Buyers
Katarzyna Janoska on why generic copy kills conversion faster than slow load times, and why reassurance is the real job of a small business website.

SEO After the CMS Absorbed the Technical Layer
Aleksandra Kołodziejska of Top Online on GEO, the 90/10 content split, and why real experience became the strongest ranking signal for local business sites.