Honestly, until a few months ago I had no idea this role even existed. But now I am fascinated by it.

So, what exactly does a Prompt Designer do? They are basically translators between humans and AI. They know how to "talk" to artificial intelligence to get the best results — turning our messy, creative ideas into clear instructions that AI can actually understand and work with.

Why am I so excited about this? Because every time we use ChatGPT, Claude, or any AI tool, the quality of our results depends entirely on how we ask. And that is a skill that can be learned.

The AUTOMAT Framework

AUTOMAT is an acronym I have been using to structure my prompts more intentionally:

• A — Act as a particular persona. Who is the AI impersonating?
• U — User Persona and Audience. Who is it talking to?
• T — Targeted Action. What do you want it to do?
• O — Output Definition. How should the response be structured?
• M — Mode, Tonality, Style. How should it communicate?
• A — Atypical Cases. Are there edge cases to handle differently?
• T — Topic Whitelisting. What relevant topics can it discuss?

The framework sounds technical but it is really just common sense made explicit. When you ask a colleague for something, you naturally give them context, specify the format you need, and tell them who it is for. We forget to do this with AI — and then wonder why the output is generic.

I am still learning this. But the more I practice, the more I see prompt design as a form of communication design — and that feels very close to what I have always cared about.

Today, AI helps me study, explore, and go deeper. Many fear that using AI will weaken our ability to think. Part of me agrees. But the real question is not whether we use AI — it is how, and how consciously.

Two mistakes I see everywhere

1. Letting AI do the work for us. If we delegate the cognitive effort, we also cancel the learning. No effort = no neuroplasticity. Growth requires friction.

2. Treating AI summaries as primary sources. A summary of a summary is not knowledge. It is processed content, often built on weak or second-hand sources. If you ask AI for research, ask it for the sources — then read them yourself.

Learning lives in the "desirable difficulty" zone: hard enough to stretch you. Not so hard that you give up.

How I actually use AI to learn

I use AI as a tutor, not a substitute. I do not ask it to summarise what I have not read. I use it to:

• Discuss ideas I have already studied
• Challenge my reasoning
• Generate metaphors that help me understand faster
• Translate concepts across domains or languages
• Explore different perspectives on something I already have an opinion about

AI is a tool. The hype will pass. Your curiosity and passions are what matter. Fall in love with the goal, not the tool. And protect what makes us human: creativity and critical thinking.

Pantone chose Mocha Mousse as the colour of 2025 — it is warm but not aggressive, neutral but with personality. The moment I saw it, I wanted to explore what it actually means visually.

My process: I started gathering references on Pinterest, then organised them in PureRef to build a cohesive moodboard — colours, subjects, palette, architectural elements. I was after something almost photographic, something I could imagine in a design campaign.

Then I used Midjourney to bring these concepts to life, iterating until the atmosphere felt right.

For those working in design or marketing: this combination of tools is especially useful in the exploratory phase. Quick concepts, testing directions, mockups for presentations. They do not replace real work, but they accelerate that phase where you are still figuring out where you want to go.

Each image must serve a purpose. I am learning to use AI generation not as a final output, but as a thinking tool — a way to make the abstract concrete before committing to a direction.

As an architect, I have spent years developing an eye for proportion, light, material. Could a prompt really capture any of that?

A few months in, my answer is: sometimes yes, often no — and the gap between the two is entirely on me.

This week I have been working with PANTONE 11-4201 Cloud Dancer — one of the key colours for 2026. It has been a surprisingly good test case. Cloud Dancer is not just "off-white." It is a specific mood: stillness, restraint, the kind of quiet that feels intentional rather than empty.

The tool does not replace the designer's eye. It demands it. The more precisely I could articulate why a space feels a certain way, the better the output. Vague prompts get vague results.

I am still learning. Some outputs feel right. Many do not. But the process itself is teaching me to verbalize things I used to just feel — and that turns out to be a useful skill far beyond AI image generation.

There is something interesting in having to explain your own aesthetic instincts to a machine. It forces a kind of self-awareness that years of studio work never quite demanded in the same way.

My Master's thesis at the Università degli Studi di Firenze: the restoration and adaptive reuse of the industrial grain Silos of Livorno, a harbour complex in Tuscany built in the early 20th century.

The project transforms a decommissioned industrial structure into a public cultural complex — preserving the raw concrete character of the silos while inserting new programmes: a museum of the port, a multimedia auditorium, educational laboratories, a wellness area with indoor pool, and a rooftop belvedere open to the city.

The Design Approach

The intervention works with contrast rather than mimicry. New elements in glass and steel are clearly legible as additions — they do not pretend to be original. The existing concrete silo cylinders become the structural and spatial identity of the project, with light introduced through new glazed facades and skylights.

The ground floor opens to the harbour with a colonnade of existing columns, a lap pool, and direct maritime access — making the building a threshold between city and sea.

AI in the Design Process

This was one of the first projects where I used AI image generation as a design tool rather than a rendering shortcut. I used it to explore material atmospheres, lighting conditions, and spatial sequences before committing to technical decisions. The gap between what I could prompt and what I actually meant taught me a lot about articulating spatial intent.

The interior of the Silos project presented a specific challenge: how to make raw industrial space feel inhabited without erasing its history.

The existing concrete — rough, marked by time, dimensionally massive — was kept exposed throughout. New floors in travertine and oak introduce warmth. Steel and glass bridges connect levels across the void of the central atrium, creating a sense of verticality and layered public life.

Key Spaces

The ground floor gallery opens directly to the harbour through a colonnade of existing cylindrical columns — the pool sits between structure and water, suspended between inside and outside.

The upper floors house the museum and educational laboratories, connected by a continuous circulation system that allows visitors to move through the building across levels.

The rooftop belvedere is a public green terrace — the highest publicly accessible point in the port area, with views across Livorno and the Ligurian Sea.

My first real automation project. The goal was simple: when a specific message is posted in a Slack channel, automatically save it to a Google Drive folder and send a Gmail notification.

Simple in theory. My first version had six steps where three were enough — redundant checks, unnecessary transformations, a conditional branch that never fired. Classic beginner mistake: over-engineering what should be straightforward.

What I learned

After stepping back, I rebuilt it in half the steps. It ran cleanly on the first test. The lesson — simpler is usually better — sounds obvious until you are the one building it.

This project taught me the core logic of automation: trigger → condition → action. Everything else is a variation on that pattern. Understanding it here made every subsequent workflow much easier to design.

Stack

n8n · Slack API · Google Drive · Gmail · No code written

A fully automated support chatbot built without writing a single line of code. The system handles incoming questions, searches a knowledge base, responds automatically, and escalates to human support when it cannot answer.

How it works

A user submits a question via webhook. An AI agent (Google Gemini) searches a vector database built from a PDF document using Supabase with pgvector embeddings. If the agent finds a relevant answer with sufficient confidence, it responds directly and saves the interaction to Airtable.

If it cannot answer, it automatically opens a Zendesk support ticket and notifies the team on Slack. No question falls through the cracks.

What makes this interesting

The system learns from a specific document rather than general knowledge. This makes it genuinely useful for product support, FAQ automation, or onboarding flows. Building this taught me how RAG (Retrieval Augmented Generation) works in practice, and how to design fallback logic that keeps humans in the loop when AI confidence is low.

Stack

n8n · Google Gemini · Supabase pgvector · Airtable · Zendesk · Slack

An ongoing creative practice using AI image generation to explore architectural atmospheres, material qualities, and colour narratives. Started during my thesis, continuing as a way to keep the design instinct active while building new technical skills.

The process

I start with a reference collection in PureRef — images that share a mood, not necessarily a subject. The goal is to understand what I am actually after before touching any tool. From there I move to Midjourney, iterating prompts until the output aligns with the feeling in the moodboard.

The gap between intention and output is where the real learning happens — it forces me to articulate spatial qualities I usually just sense.

Recent explorations

Mocha Mousse (Pantone 2025) — warm, tactile, almost edible. Cloud Dancer (Pantone 2026) — restrained, quiet, the colour of considered emptiness. Each colour becomes a brief: what space does this colour want to live in? What light, what material, what programme?

The full technical documentation set for the Silos Livorno thesis project, developed across multiple scales to communicate both the conceptual logic and the spatial precision of the intervention.

Drawing set

The exploded axonometric shows the layered logic of the project — existing structure, new insertions, programme distribution across levels — as a single legible image.

The longitudinal sections at 1:200 and 1:50 show the relationship between the massive scale of the silos and the human scale of the interior spaces — how light enters, how levels connect, how the indoor pool sits at water level between the building and the harbour.

The floor plans document the functional organisation of each level: ground floor public access and wellness, intermediate floors for museum and laboratories, upper floors for the auditorium and archive, rooftop belvedere.

Tools

AutoCAD · Rhino · Adobe Illustrator