Builders Unscripted: Ep. 4 - Pietro Schirano

**TL;DR:** Pietro Schirano demonstrates how to turn creative ideas into reality using GPT 5.5 and Codex — from encoding images into sound and back, to automatic icon generation, to a voice-driven writing assistant. These applications turn AI into an extension of the mind. ## The Versatile Explorer Pietro is a musician, engineer, designer, founder, and early tester of OpenAI's models. His creativity stems from an explorer's mindset: "We were born too late to explore the oceans, too early to explore the stars." In his view, AI is like a new ocean and new stars — a latent space to explore like Magellan or Marco Polo. "You can enter the mind of this incredible new technology and explore it like explorers of the past, figuring out what you can do with this technology," he says. ## Three Dimensions for Evaluating New Models With each new model release (e.g., GPT 5.5), Pietro uses a personal evaluation set to test three key dimensions: 1. **Instruction Following** – How well does the model strictly follow instructions? He cautions that more "safe" updates sometimes sacrifice follow-through ability. 2. **Task Execution** – Can the model actually carry out the operations he requests? 3. **Multi-Agent Guidance** – How well does the model guide other agents? This is a complex task, e.g., having Codex spawn multiple sub-agents and coordinate them. He shows off his custom shortcut system for Codex: for example, the "Spawn" command generates multiple agents, and the "PR" command reviews code and pushes pull requests. A major improvement in GPT 5.5 is its ability to stay focused on a task for longer, which is essential for truly "unlocking wealth." ## Creative Demo: The Mysterious Communication Between Image and Sound Pietro demonstrates the first app built with GPT 5.5 — converting an image into sound, and then reconstructing the image from that sound. He plays a piece of music generated from the OpenAI logo, which sounds harmonious and pleasant. Then he drags the sound back into the app and asks it to restore the image — the logo appears perfectly. He mentions this reminds him of the scene in Steven Spielberg's *Close Encounters of the Third Kind* where humans and aliens communicate in a similar way. Behind this app lies a fusion of cryptography, music understanding, and multimodal reasoning, and version 5.5 completed the entire workflow almost in one shot. ## From Icon Design to App Icon When asked about his design process, Pietro shows the just-generated "Odysseus" app icon — a Greek ship combined with a quill, featuring gradients, textures, and border effects. He marvels: "It would have taken me hours in Photoshop 15 years ago to get this effect, and now it's done in one shot." ## Writing Assistant Odysseus: Extension of the Mind Odysseus is a writing autocomplete tool, named after *The Odyssey* and *Ulysses*, symbolizing "an ocean of infinite possibilities." It can: - Autocomplete the next likely text as you type. - Via the GPT real-time voice version, convert spoken words directly into text, capturing thoughts in real time like a second brain. - Support stub commands that let the model automatically generate content, and allow insertion and editing in the middle. "We humans are incredible input machines (vision, sound, smell), but we're kind of terrible at output," Pietro says. Odysseus makes the interface feel like a natural extension of oneself — "I speak to it, it writes in real time, I can stop, autocomplete — it's insane." ## Programming as a New Medium Pietro compares AI-assisted programming to photography: in the past, not everyone could be a good painter, but nearly anyone can press a shutter and take a photo. Now with Codex and GPT, programming is no longer an unattainable skill — it has become a "moldable clay," a snapshot of an idea in your mind. This shift exhilarates him. **Source:** Builders Unscripted: Ep. 4 - Pietro Schirano | YouTube (https://www.youtube.com/watch?v=SPC_yCe1cUw)