Where AI ends
A personal case study on where to put the line between human and AI
I‘ve wanted to write this case study for months, because it’s the clearest answer I have to the only question that matters right now: where exactly does the human end and the AI begin?
It’s a personal story — my (Peter) personal project. Which means no NDA gymnastics, and I know every detail by heart.
Quick context first. Over the last year, AI has absorbed most of the quantifiable work in marketing. Generic copywriting, the wood-chopping level of design work, the basics of UX, almost the entire performance suite with advertising, analytics, and reporting, competitor analysis, anything that involves reports, summaries, and briefs, and the basics of strategy creation. Faster, cheaper, often better.
You might argue that your team, your super-unique specialism, is the exception. More likely, you just haven’t switched yet and are working manually out of necessity. Either way: AI is taking over your junior layer and much of your mid-management work, at a fraction of the time and cost. If a skill is quantifiable, AI is better at it — or the next model will be. That’s a race humans can’t win, so there’s no point entering it.
We have argued for a while that this final frontier will be taste. Human taste — the ability to select what’s unique, what’s worth keeping. Not the artist, but the gallery owner. Let AI create, let humans curate.
That thesis was incomplete.
First, I don’t want to hurt anyone’s feelings, but taste is not really needed sometimes, and most marketers don’t have any anyway. Marketing cares more about operations than creativity. Creativity, itself, is kinda itchy and funky. We don’t know much about it, but one thing sure we do know that it doesn’t fit on quarterly reports. It is not quantifiable. And because our clients, our bosses, and even us are obsessed with tracking, numbers, lists, and graphs, we simply trained out to care about creative things. Add the brainrot of social media to this and what kind of content we consume day-by-day, and we end up with a generation without any taste whatsoever.
So… why bother with taste?
But more importantly, what we underestimated is how fast the models would absorb the execution side of taste. Ask Claude today to build a website using 3-4 colors you provide, a short outline of what you like, and a few sentences about your brand. It will one-shot a site in minutes that most designers and developers would take two weeks to deliver — and I’ve personally watched designers sit on projects for weeks and produce worse. Taste as selection is being commoditized. Since taste is not really needed, we don’t really have it, and generic baseline of it works and does the job 95% of the time, and AI can do that automated, then, again… why bother?
But what’s left then?
What survives sits upstream of selection: the idea. The point of view that generates the selection criteria in the first place.
The idea is personal. It comes from your worldview and it points at the future: “I think this should exist, and I want to make it happen.” People call this agency, but agency without an idea is a hamster on a wheel — maxed-out willingness, nothing to aim it at, which describes most startup hustlers. The creation has to come from you.
Here’s how that works in practice.
The idea
I’m at the stage in my life where I can do whatever the f—ck I want. And for many years now, I wanted to completely stop working for others. And there is not enough money on the entire planet that is enough to pay me to sit in and manage a team, again. I loved doing this content business since it can be done from anywhere, anytime, and it brings enough cash that is reasonable.
So first, we toyed around the idea with Torsten to expand this AI-Ready brand. Start some sister brands around sales, customer experience, and creatives. Kinda holding the entire customer buying cycle. Certainly doable, but since I gave up client work a year ago, I realized something. I simply didn’t care at all. Not to mention that we are experts on marketing, but not experts on CX or sales. And since, I don’t care, I don’t want to be.
So I realized, what I want really is a passion project, a personal one. Where I combine my expertise (marketing + AI), my agency (I work like a monster if I care), and my passion together. The money? Yeah, it will take care of it itself, I’ll play the longgame. Again, I’m at the stage in my life where I can do this (takes years of saving and buying boring ETFs! :D).
From there, it was simple. I live in Italy, a country I chose. I wasn’t born here. I’ve been a foreigner here since 2021, but I know more and more about this place and its people, and I have a very solid point of view on how to enjoy this country as a foreigner or traveler.
So, Anywhere Italy was born. It’s a publication about the hidden Italy, 1,000 specific towns to be precise, offbeat experiences, local culture, and regional food. Written by a local expat (myself) for slow travelers and people who are interested in moving here.
From day 1, I committed to one rule: I don’t write about anything I haven’t personally experienced. The towns and restaurants must be visited, the experiences experienced, the food cooked or tasted by me, and me only.
AI’s role
Let’s start with AI, because that’s what we want anyway, don’t we? From day 1, I built an AI project around this idea. The AI has four roles.
The researcher.
There’s a very specific way the AI does research for this project. First, it ignores English-language sources, because most of them are shallow garbage about this country — yes, even the big names. Only Italian sources play. Maybe French or German.
Second, everything I produce should be grounded in research, because I can experience anything, but I can’t know everything. What I do with the research is up to me. I decide whether The Story about the not-so-famous Medici prince makes it into my coverage of a Tuscan hilltop town. I studied history and sociology, so I can take the findings and make them shine with my own point of view.
Bottom line: I don’t do research. AI does, based on my pre-defined needs. It saves me an enormous amount of time.
The advisor.
AI acts as my business advisor. It handled the brand strategy, the initial research, the entire business setup, the marketing strategy, and the full content architecture. All of it interactive — I brought well-defined goals and needs — but how would I know how travel media make money and promote their content? Let the bot know that and advise me.
It’s also an ongoing consultant. A business strategy is not set-and-forget. It’s an evolution. Whenever I have questions, AI consults the existing data and helps me decide.
For this role, the AI has access to everything. And I mean everything: the brand book, the written strategy, the tone of voice, the full content archive, all performance numbers, all design elements. The more information it has, the better the advice.
The builder.
The AI builds every digital asset. This project launched 2 months ago — during which I was fully offline for 2 weeks for personal reasons — and it already has a site, an app, and a database of 1,000 towns. Anywhere Italy Towns is the database is built on key filters: tourism factors, location, population, publicly available content, and more. I hand-selected the 1,000 towns from Italy's roughly 8,000 communes. AI collected, filtered, built, and made the whole thing searchable and presentable in 3 workdays. The database is interactive, either via the site or an app, letting you plan your slow travel based on your preferences and needs. It's free to use, and its main job is SEO and programmatic SEO for long-tail discoverability. Every plan generated is a page for SEO, forever, and it’s generated by users, not by me.
Both the site and content production run through an automated agent I named Marco. Marco knows every asset and the entire brand, lives on a VPS with access to everything, and runs around the clock. He maintains and refreshes the site, generates SEO content on it, connects the human-written content to the site, scans the competitive landscape, and reports on all of it.
Some workflows are fully autonomous — Marco runs them without me asking. Others are semi-autonomous: I trigger them with a single request, and the flow runs to completion on its own.
The distributor.
Marco runs almost all social channels for this brand. X, Bluesky, Threads, Pinterest, and Facebook are fully autonomous, driven by a social media handbook and using human-written content. I don’t touch them. I don’t care.
Instagram is a prime channel and requires human handling, so Marco stays out. YouTube is fully human too.
Marco also sends me a weekly report on what my competitors have written. I turn that into a Reading Table post, quoting and linking their pieces. It doubles as my RSS reader, so I actually read them all, which is fun — they write about Italy just like I do. That post is a key growth lever, because it’s effectively a collaboration with multiple authors.
And Marco is my PR agency. He pitches podcasts, PR articles, and bloggers outside of Substack for collaborations. He has email capabilities and can be triggered by my email. We built this capability first for our AI Ready CMO agent, Jenny — she does the same, plus sponsorship outreach. It worked, so switching the brand over was a no-brainer.
The entire cost of AI usage:
~$100/month on the Claude Max plan + tokens
$10/month on PostForMe, the scheduling app Marco uses for social
$20/month on Hetzner, the VPS where all sites, assets, and Marco live
A rounding error on any marketing department’s budget.
Now, let’s see where I am in the picture.
The human’s role
The “product” is not the planner or the towns page. The product is the content and the list itself. Monetization happens on both ends: you pay to read some content, or you buy items from the list. The latter isn’t activated yet — we’re in the growth phase — but it will be once the list is large enough.
None of it works if the content isn’t good enough. And how do you write about Italy, one of the most-covered topics on earth, without producing another 10-things-to-do-in-Florence piece? Only by writing from your own point of view. I live here. I experience the country the way only I can, and I have very strong opinions on some things and zero interest in others — which is a strong opinion in itself. AI has no opinion and no experience. It can fake a point of view if you hand it one, but it reads as generated within seconds.
So the human has four roles too.
The writer.
Let me be precise, because it matters. There is plenty of AI-written content on the site. The towns pages and their copy? Almost all AI. It exists for distribution, it carries no personal touch, and I can’t scale it without AI. That’s fine.
Content published under my name is 95% written by me. The other 5% is the manuals — practical pieces that exist to help people do X. There, AI research is the core, AI writes most of it, and I’m essentially the editor adding my point of view.
But the Features about towns and experiences stay 100% human. Otherwise, what’s the point?
The prime social media.
Instagram is our core growth channel. It’s Italy — people love the visuals. Our brand is slow travel, so instead of quick-cut, dopamine-fueled “44 things you must do in Italy this summer,” we make slow videos that calm you down. All done by hand — mostly by my wife, who runs that channel.
YouTube is second, also 100% human, same idea: slow, ASMR-like content about Italy. Usually me reading a Feature article out loud, plus a 10-20 minute video I recorded — a piazza, camera standing in one place the whole time.
LinkedIn is template-driven. AI helped with research and the templates, but due to platform limitations, I post manually, mostly written by me. I have 25K+ followers on LinkedIn, by the way. I stopped sharing AI+marketing content a couple of days ago, not because I don’t want to, but because it is so crowded and insane that it just doesn’t make sense anymore. However, I started sharing content about Italy. Oddly, it works. People love seeing stuff that simply just doesn’t belong to the professional spaces like LinkedIn. We are all humans afterall.
A sidenote, because it’s a real model limitation: the audio-video part is the hardest to scale. I have to sit down and record and edit. I tried cloning my voice with AI, and it did an impressive job on short formats, but it can’t do long ones. A 15-minute reading in my cloned voice doesn’t sound like me as a human. I make mistakes, I breathe and pause at unusual points, and I have a hard-to-replicate English-Hungarian-Italian accent. The output was artificial, so I stopped. And I didn’t go cheap — I was on the maxed-out ElevenLabs plan. Maybe voice models evolve. For now, this is a clear limit.
The networking.
All networking is me. I network with other writers and expats covering Italy. I write my own Substack Notes, and it’s me in the Substack chats and comments. It doesn’t feel like work, because this is my passion — I couldn’t do it for marketing anymore, sorry.
It’s also me in Facebook Groups and Reddit threads. Honestly, I’d hand this one to a bot if I could — it’s a distribution channel that currently can’t be fully automated, and the limitation is purely technical. If you’ve figured this out, tell me. I’d pay for it.
And obviously, when podcasts, videos, and live sessions happen, it will be me. That can’t be faked — for now — but I wouldn’t fake it for this brand anyway.
Improvisation.
Remember: no strategy is fixed. It’s an evolution, and the evolution of ideas runs on improvisation. In boring business speech, we call it pivoting. AI can’t pivot. Humans can.
Too early to tell — this project is a couple of months old — but I’m sure I’ll pivot later. It happened with AI-Ready CMO. It’ll probably happen here too.
So, what’s the learning here?
This case study applies well beyond passion projects — it maps onto marketing campaigns just as cleanly. As of today, we run on 5 rules for where AI ends.
AI can’t have original ideas. It produces reactive ideas — advising, conversing, brainstorming. If you’re looking for a truly unique idea, you won’t prompt your way to it.
AI can’t have a point of view. It can fake one, like it fakes opinions, experiences, and feelings. Only individuals have points of view. If you want a unique perspective, search for it within yourself.
Putting AI at the core of your brand is risky. Brand damage is real and PR crises are real. Where personal connection, touch, or networking matters, AI shouldn’t have any space.
AI can analyze but can’t get inspired. Agents improve within the feedback loops you build for them, but they only learn inside their own space. A human still has to look at the reports and decide what they mean.
AI won’t know where the limits are. Ask it, and it will generate the entire brand, all the content, all the strategy — relentlessly and endlessly. You decide the limits. You put up the barriers.
If you're curious about the project itself, join the ride here:
Share with a colleague. Earn paid membership months.
It seems experience is the new language…