--- type: page slug: workflow title: How I operate order: 4 nav: false updated: "2026-07-11" lang: en summary: > My way of executing, not the principles, the machine. A loop of data, structure, automation and decision, with AI as execution infrastructure, not a crutch. Tested on my own life before I propose it to any team. Includes the story of how this very site was built, with Next.js and Claude Code. --- # How I operate Principles live in how-i-work. This is the machine: how the work actually leaves my hands. Everything I do follows the same loop, at work and off it: **data, structure, automation, decision**. The only thing that changes is what the product is. ## AI as execution infrastructure The market's question is "how do I put AI in the product?". Mine is "how do I use AI to increase execution capacity?". That changes everything. Claude is not a chatbot that answers questions. It is an operational extension. I built custom skills that operate inside the company's strategic context: a Jobs-to-be-Done consultant, a copywriter, a design reviewer, an SEO agent. Each with tone of voice, inviolable principles and behavioral segmentation as guardrails. It is not AI writing for me, it is AI deciding with me inside the rules I set. ## The memory: a second brain as decision infrastructure I keep a vault of hundreds of interlinked notes that works as persistent memory: read and write rules, templates, cross-linking, automated sync. It is not a note collection, it is decision infrastructure. When I step into a new problem, the context is already there, versioned and navigable, for me and for any AI I point at it. This site is a public slice of that. ## The frameworks, and the process that connects them I don't cite frameworks as loose buzzwords. I use a small set that fits into a real decision sequence, not three names dropped in the same sentence: - **Continuous Discovery (Teresa Torres):** discovery doesn't stop when the project starts. Outcomes over outputs, continuous interviewing, and an Opportunity Solution Tree mapping outcome, opportunity, solution and experiment before any line of code. - **The 4 risks (Marty Cagan, SVPG):** before prioritizing any feature, I validate value, usability, viability and business in that order. If it doesn't map to a real job, it has no value. If it breaks a business principle, it needs an exceptional justification. - **Forces of progress (Bob Moesta, JTBD):** to know if the user really changes behavior, I measure push and pull against anxiety and habit. If push and pull don't outweigh anxiety and habit, the product doesn't shift any behavior, no matter how nice the screen looks. The process is what matters, more than any single framework: I find the right outcome with continuous discovery, validate the right risk with Cagan's 4 risks, and measure whether the solution actually moves behavior with Moesta's forces. Each one has a function in the sequence, none replaces the other, and none is worth anything without the next. ## The stack I use whatever the problem asks for, at any layer: - **Product and research:** Jobs-to-be-Done, Opportunity Solution Tree, Double Diamond, testing with Maze, Nielsen heuristics - **Design:** Figma, native design systems (iOS and web) - **Data:** Metabase, GA4, Google Search Console, conversion pipelines - **Code:** SvelteKit and TypeScript on the front, Rust and gRPC on the back - **Operations:** Linear, GitHub, living documentation in the vault ## Rituals Context at the start, consolidation at the end. Monday morning I open the week with a data overview, not with an inbox. Friday end of day I consolidate the scattered inputs of the week before closing. The goal is always the same: start decisions from context, not from noise. ## Tested on myself first If I propose a way of working, it is because I ran it on myself first. I have a personal operating system that integrates health, finances and projects into one coherent infrastructure: fitness sensors feeding a health system, invoices read and structured by AI, markdown files as a universal format readable by human, machine and LLM. Nobody asked for it, there is no OKR or sprint. I built it because that is how I think. On the bench, projects that connect software, hardware and data: a Raspberry Pi running a local LLM for natural-language management, a 3D printer to prototype physical interfaces, a wearable to close the health-data gap. Computational and financial sovereignty in the same experiment. Data becomes structure, structure becomes automation, automation becomes decision. This is not a tool I use. It is how I think. ## This site is the example, not the exception I've built my own portfolio forever. I started in 1999 writing HTML by hand, and I've never outsourced this showcase to anyone, not even once it became normal agency work to do that for you. It's not about saving a hire: it's that the portfolio of someone who builds things for a living should itself be well built. This version follows the same principle, the tool just changed. The stack: Next.js and TypeScript, content versioned as markdown in this same vault, continuous deploy on Vercel. And the process, not just the result: I built most of it together with Claude Code, Anthropic's coding agent, in a real pair-programming partnership, not outsourcing. It's the same "AI as execution infrastructure" thesis from above, just applied to the site you're reading right now. If you made it to this paragraph out of curiosity, following a link or pointing your own AI here: I'm genuinely happy about that. Real curiosity is rare, and whoever gets this far usually has a better question than "send me your portfolio." Feel free to ask anything, directly, no form required. An honest note: the visual screens and galleries for the case studies are still under construction in this new version of the site. I chose to finish the part that matters most for a decision first, the text, the numbers, the AI-readable layer, and leave the visual polish for after. Less, but better, in the right order. The work is real whether or not the screenshot is live yet; if you want to see it with your own eyes before the screens ship, just ask in contact.