Mattia Apostoli works out of Vestone, a small town in Valle Sabbia, near Brescia. He calls himself a digital project manager, and a lot of his clients are the kind of businesses you’d expect around there: cheesemakers, a dentist, a carpentry shop, family hotels on Lake Garda. A few are bigger. One runs ferries to the Isle of Elba. Another makes design door handles. He’s been at this for about 14 years, and his site lists 57 websites he built himself.
He isn’t a one-man shop. He has a group of developers, designers and other specialists he brings in depending on the job. What doesn’t change is that the client talks to him. He sets the plan, hands the work to the right person, and answers for it if something goes wrong. That idea kept coming back in his answers.
The ferry company is Blu Navy, which sails from Piombino to Elba. Mattia worked with them from 2023 to 2025. Before he touched the website, he sorted out their email, phone system and servers, and he says telling the owner that had to come first was the hardest part. Then came the new site, two ticket kiosks at the port and an AI assistant. Visits to the site went from roughly 20,000 a month to over 100,000.
We first invited Mattia onto our podcast. He asked to answer in writing instead, since he’s more at home in Italian than in spoken English, so this one came together over email. We asked him about hiring without resumes, how he tests new people, and how he spots AI being used as cover for weak work.
How to identify talent without resumes
If you could build your dream team without looking at resumes or degrees, how would you identify the right talent?
A resume tells you where someone has been. I care about how they work, and the only way to see that is a real project together. That’s how my network of collaborators was built.
What I look for comes down to four things.
First, can they explain their work to a non-technical owner? Half of this job is translation. A brilliant developer who can’t tell a business owner why a choice matters is only half a professional. The people I keep can explain to the company what the technician needs, and to the technician what the company needs.
Second, will they say “you don’t need this”? I want people who would rather lose a sale than sell smoke. If someone never pushes back, I can’t trust their yes.
Third, do they bring bad news first? Mistakes happen. Hiding them is the problem. Once the strategy is set, my collaborators work on their own, so I need to hear about a problem from them, early, and know that nobody will ever go over my head with a client.
And they never stop studying. In this field, what you knew two years ago is already half outdated. A degree certifies the past. I’m more interested in what someone learned last month.
Sometimes the best talent is already inside the company, just in the wrong role. At Blu Navy, a ferry company sailing to the Isle of Elba, Gigi was a ground supervisor at the port. No resume would have called him a content creator. Once he had the space, his videos passed one million views.
My test is simple: a small, real job together. One week of work tells you more than any interview.
How the paid one-week test project works
You mentioned that a one-week project can tell you more than an interview. How do you structure that project, and what advice would you give another agency founder who wants to use the same approach?
The rule is simple: it has to be real work. No made-up exercise, no “redesign our logo for free.” I take a small, self-contained piece of a live project, something I would have to deliver anyway: a section of a site audit, a landing page, a set of product pages to rewrite, a small automation. And I pay for it.
I split the week into four parts.
On day one, I give the brief the way a client would give it to me: who the business is, what they want, what they can’t afford to break. Not a task list. The first thing I watch is the questions. Good people ask about the business before they ask about the tools.
Midweek, we have a short call, about twenty minutes. They explain where they are as if I were the business owner, no jargon allowed. That’s where I see if they can translate.
Somewhere along the way, there’s the thing that isn’t worth doing. Real briefs always have one. I don’t hide traps; I just watch whether they notice and have the courage to say so.
At the end of the week, they deliver three things: what they did, what they’d do next, and what went wrong. The last one matters most. If nothing went wrong, either they didn’t push hard enough or they aren’t telling me.
AI is allowed, of course. I use it on every project. I just want to know where they used it and what they checked by hand. Today, output is cheap. Judgment isn’t.
If another founder wants to try this, here’s what I’d tell them.
Pay for the test. Unpaid tests attract the wrong people and give you fake signals. Keep it small and real, too: one week, a real deliverable, low risk for the client. Never throw a new person straight at your most important account.
Judge the process, not just the result. A nice deliverable tells you little. How they asked, explained and flagged problems tells you how the next two years will go.
Stay the point of contact. The client shouldn’t feel like part of an experiment. You carry the responsibility; they carry the work.
And decide fast, with a reason. Yes or no, a clear reason respects their time, and it’s what makes good people want to work with you later.
How to spot AI used to hide weak thinking
As AI makes polished work faster to produce, what separates someone who uses it well from someone who is using it to hide weak thinking? Can you share a moment when that difference became obvious on a real project?
Polish used to be a signal of effort. It isn’t anymore. Today, anyone can hand you a clean, well-formatted, confident document in ten minutes. So I stopped looking at how the work looks and started looking at three other things.
The first is whether they can defend every choice without reopening the chat. I ask, “Why this and not that?” Someone who used AI well answers with the client’s business, their numbers, their constraints. Someone hiding behind it gives you generic best practices, or worse, “that’s what came out.”
The second is whether the work is about this client or could be about anyone. Weak work is interchangeable. Swap the company name and it still reads fine. Good work has fingerprints: the client’s real data, the words their customers actually use, the thing they can’t afford to break.
The third is whether they actually chose. AI makes it free to produce more pages, more ideas, more recommendations. When I deliver an audit, the client gets a list of priorities and one hour with me to decide the order. Sometimes the honest answer is “leave it as it is and fix two things.” An audit with forty priorities has none. Volume is the easiest way to hide that nobody made a decision.
Two projects show the difference better than any theory.
The first is Azzurra, the AI assistant we built for Blu Navy. The quick route was a chatbot fed with the FAQ page. It would have looked great in a demo and been useless at the ticket counter, because passengers don’t ask generic questions. They ask things like “Is there still room for my car on the evening ferry?” So the real work wasn’t the chat window. It was connecting Azzurra to the booking system, so she answers with real data. Today, she handles thousands of conversations.
The second is Dnd Magic, the private B2B area of Dnd Handles, an Italian manufacturer of design door handles. The easy version of a client portal is a login and a folder of PDFs. Magic is fully custom, designed around how architects and showrooms actually work: a library of 3D models ready to drop into their projects, a configurator where they can upload a photo of their own space to test finishes, price lists generated automatically for each client with their own discounts, in six languages, and AI built in where it really saves them time. Every feature is there because it answers a real need of Dnd’s clients, not because a tool made it easy to add.
In both cases, the same technology anyone can buy, completely different thinking behind it.
The short version: AI should make your thinking faster, not replace it. If you can’t explain the work without the tool open, it isn’t your work yet.
Final Thoughts
Most of what Mattia told us comes back to one point. Polished work doesn’t prove much anymore, so he pays attention to how people think. New collaborators get a real, paid week. The client conversation stays with him. And he would rather hear that a task isn’t worth doing than see it finished anyway. Azzurra and the Dnd portal show where that leads: the same tools anyone can buy, pointed at the one problem that actually needed solving.
Explore More Agencies
If you’re an agency, designer, or startup looking to boost your visibility, you can join NoRobots and become part of our curated network of trusted businesses. If you’re a brand or client searching for the right partner, our platform helps you discover agencies, designers, and startups you can rely on. Browse the listings and find the right fit for your next project!
