I started designing Serata Perfetta from a simple idea: make it easier to find something to do for the evening. We launched in Rome and Milan, but the project is meant to keep growing city by city, not stop there: Florence and other cities are already on the roadmap. A drink, a concert, a movie nearby, without spending half an hour checking five different sites, wherever you live.
I'd been working with vibe coding for a few months, and like anyone who goes through the first weeks of it, I ran straight into the same doubt. The code shipped fast, clients were happy, but I didn't feel entirely comfortable charging for work that, for the most part, a language model had written in my place. It's a feeling with a name, impostor syndrome, and among people who work with AI every day, especially on code and websites, it's more common than people admit. What I was actually doing, though, was orchestrating: understanding what was needed, translating it into instructions the AI could execute, and stepping in personally on the code where automation wasn't enough. Orchestrating, fittingly, is the word we'll come back to shortly.
The process before, the process after
Before, the workflow was fragmented. I'd talk to Claude to define an idea, copy the result into Cursor to turn it into code, have a summary file extracted in Markdown, and go back to Claude for the next step. It worked, but every handoff between different tools cost time and lost context.
I met VibeCoded Tools along the way, and with them Martino Cesaratto, the project's founder. What convinced me wasn't a promise that "the AI does everything on its own," but the opposite: a system that makes the orchestration itself faster and more reliable, while keeping the developer at the center of the decisions.
With the VibeCoded Orchestrator (VCO), the workflow changes radically. A local knowledge graph keeps persistent memory of the project across sessions, a code graph automatically indexes functions and components so the AI can search the code instead of re-reading it every time, and a set of specialized agents covers planning, writing code, testing, and review without having to rebuild context from scratch every time. Everything runs locally: no project data leaves the machine unless I explicitly decide it should.
With this tool, and with Martino's constant updates (he builds it first and foremost to use it himself every day, so every improvement comes from a real problem), the learning curve for me and my clients climbed fast. It wasn't a sudden revolution so much as friction that quietly disappeared: less time spent re-explaining context, more time deciding what to actually build.
What I actually built on Serata Perfetta
I'm not speaking secondhand here: these are live features on the site, today.
Structured SEO, not improvised. Serata Perfetta dynamically generates dozens of dedicated pages by city and category (tonight's events, concerts, restaurants, clubs, cinemas, theater, for Rome and Milan), with a sitemap that updates automatically every hour. Every event carries Schema.org structured data in JSON-LD format, so Google can read an event's date, venue, and price without having to guess from the page text.
GEO, optimization for generative engines. Beyond classic SEO, the site exposes an llms.txt file designed to be read and cited directly by ChatGPT, Perplexity, or Gemini when someone asks "what to do tonight in Rome." The robots.txt explicitly declares Content-Signal directives: the content can be used as a source in an AI assistant's answers, but not for training models.
An always-current events database. Behind the scenes, a database collects and keeps events, venues, cinema listings, and editorial guides up to date, so users always find fresh information without anyone having to enter it by hand one at a time.
A real sustainability model, through transparent affiliate links and direct partnerships, so the project can grow without asking for a subscription from people who are just looking for something to do tonight.
The result: first-page Google rankings on several local searches within months of launch, and growth is just getting started.
What I'm taking away from this
Vibe coding doesn't strip value from the work of people who practice it seriously. It shifts it. I no longer write every line by hand, but I decide the architecture, verify the data is correct, and step in where human judgment is needed, not just execution speed. Having a tool like VCO support that work, instead of replacing it, is what took away my initial doubt: I'm not selling AI-generated work passed off as my own. I'm selling a method, and the results it produces.
If you're evaluating vibe coding for a project of your own, or want to understand how an orchestrator works in practice, the VibeCoded Orchestrator is free and open source (AGPL-3.0), available on GitHub. Install the free Orchestrator or explore the full catalog.