With no hand-written code, they built a daily system that fits how their mind

Achievements In four months, without writing a single line of code by hand, I built and shipped an ecosystem that now runs my working life, and I use it every single day. Two web applications are in production. The first is a ninety-second regulation ritual that gives me a breathing triangle, a mantra spoken in my own cloned voice, and music matched to whatever state I check in from. It reaches me through push notifications, so it arrives without waiting to be opened. The second holds my projects and my tasks, all measured in a unit of work I call a daisy, which is twenty-five minutes of focus followed by two minutes of meditation and three of movement. The two applications talk to each other. When a focus session ends, the ritual opens already knowing what I was working on and greets me with the line “You’ve just left: Discussion with DG Comp”. When the ritual ends, my energy state travels back with me and lands on the task I had abandoned. I looked everywhere for a meditation app that knew what I had just closed, and in the end I built it myself. I also built a three-dimensional landscape of my work in which every project is an island. A translucent building shows everything still to be built while solid floors show what I have already finished, and a question mark stands over any project I never estimated, which makes my invisible work visible for the first time. Everything in that landscape is green and gold, because a brain like mine abandons an application the day it starts keeping score. An AI chief of staff reads my calendar, my projects, my deadlines and my email every morning, then proposes three trade-offs that I remain entirely free to refuse. Beyond this ecosystem I applied the same way of working three more times. I developed a private Make connector to read my business email into a CRM, because no ready-made integration existed for my provider. I built a card-scanning pipeline where OCR reads the card and agents verify what was read before anything is saved. And I rebuilt the entire accounting of my two companies with Claude, Make, Box and Airtable. Building one working system might have been luck, but building three of them the same way is what convinced me I had found a method. Tools I use Claude in three forms and across three models. Cowork orchestrates the work, Claude Code builds it, and the API powers the assistant that lives inside the application. Mechanical work goes to Sonnet, reasoning to Opus and visual design to Fable, which has become the model I rely on to build my web pages. Learning to match the model to the nature of the task turned out to be one of the real lessons of this programme. I build my presentations with Gamma and with Claude, and my web pages with Claude Fable, which means the whole chain from the first idea to the published page now runs on the same set of tools. Airtable holds everything across three bases covering my projects, my rituals and my email CRM. Make moves work between systems, including through the private connector I developed myself. Box stores the accounting documents until they are needed. ElevenLabs speaks the mantras in my cloned voice, SUNO composed the music beds, and Whisper turns the voice notes I record while moving around into ideas I can find again later. Firebase handles identity and push notifications, Vercel hosts the applications together with their serverless functions, and GitHub Actions runs the scheduling. The interfaces are built with React, Vite and Tailwind, and the landscape runs on Three.js. Methodology I stopped chatting with AI and started onboarding it. Each application carries its own constitution, a standing instruction file holding its vocabulary, its business rules and the things that must never be done, and every agent reads that file before it is allowed to touch anything. Every feature begins as a written brief instead of a prompt. The brief explains why the feature matters, what it has to do and how it should behave, and Claude Code builds from there. The thinking lives in the brief, and the code follows almost on its own. I work by orchestrating several agents at once. Each build gets an agent of its own with a defined scope, explicit guardrails and a review before I accept anything. Over the final weekend three agents worked in parallel on three separate builds while I steered all of them, and each one was told in writing what it must never touch. Every conversation graduates into something durable. A conversation becomes a brief, the brief becomes a skill, and the skill eventually becomes part of an application, so nothing valuable stays trapped in a chat window. And everything gets tested in the real world. Agents take screenshots of their own work and check it in a browser, I test on my own phone, and nothing is ever declared finished simply because it compiles. Learnings The brief turned out to be the product. What an agent builds is almost entirely determined by the quality of what I wrote before it started. My first attempt at the three-dimensional landscape failed because my brief described a metaphor without ever describing a visual language, and the second attempt worked because I supplied a reference image, a projection, a palette and a composition. Vibe coding is accessible without ever being easy, and it means less coding with considerably more thinking. I spent far more hours thinking than an engineer would have spent typing. Constraints turned out to matter more than instructions. Telling an agent what it must never touch has prevented more damage than any positive instruction I have ever written. Beauty has to come before data. My first landscape was judged ugly because we connected the database before the world was worth looking at, so on the second attempt I imposed the order and refused to connect anything until the world itself had been validated. Being my own first customer is what found the real defects. In the final days I discovered that my push notifications were being blocked by a twenty-hour lock inherited from the previous evening, and that Apple silently ignores volume settings on iPhone, which meant my music was drowning my own voice. No automated test would have caught either one, and only living with the thing every day did. The deepest lesson is that AI First turned out to mean Human First. What I built fits the way my mind actually works and this is a system that I have been looking for and never found so far. Thank you all for bringing me to that stage, a first stepping stone to more developments to come! Link to videos: https://vimeo.com/1216836797/e6d5f46155?share=copy&fl=sv&fe=ci https://vimeo.com/1216837997/a39ca98d42?share=copy&fl=sv&fe=ci