The Tech Nation · Newsletter #9 body { margin: 0; padding: 0; background: #edece8; font-family: Georgia, 'Times New Roman', serif; } Issue #9 · Summer edition 2026 People · Tech · Impact ☀️ The summer edition · Barcelona ⚡ Claude Community Event #5 · this is the one Our fifth Claude Community Event. Catch it before the city empties out. The room where the Barcelona AI scene actually shows up. Launches, builders, real conversations. The summer one you do not want to miss. Save your spot → The summer issue. I went hunting for a creative direction and ended up back on GPT for images, with one feature that changed how I work. Then the CRM that maintains itself, and the actual plumbing behind it. The privacy line, and the open source tools starting to draw it. Plus our events before the break: the Founder AI Buildathon with Karisma and Itnig, the fifth Cheese, Tech & Wine, and Crewly Cohort #5, on top of our fifth Claude Community Event. Main Feature For images, I went back to GPT. One feature is the real game changer. Last week I was hunting for a new creative direction on a project. I started where I usually start, with Claude and Claude Design, and kept landing on the same safe, familiar look. Warmed up déjà vu, competent and a little dead. On Manar's advice I reopened ChatGPT with the exact same brief, one change: I asked for finished Hero visuals as images, not code, not a design system. The range opened up at once. More directions, more willingness to push the brief somewhere I had not asked for, instead of one tidy answer. But the thing that actually changed how I work is smaller and easy to miss: commenting directly on an image. It is worth two minutes of your time, so here is how it works. 01 · Point, don't re-prompt Generate an image, then click straight on the part you want to fix and describe only that change. Everything else stays exactly as it was. No rewriting the whole prompt and rolling the dice again, no losing the 90 percent that was already right to repair the 10 percent that was not. 02 · Lift elements out, clean Ask it to pull a single element out of the image on a transparent background. A logo, an object, a character, extracted and ready to drop into anything. For anyone who has ever fought a cutout, that one alone is worth the trip. "Image generation stops being a slot machine. It becomes editing." The honest version. I build on Claude every day and that has not changed. For reasoning, writing, and agents, it is still where I work. But for raw image creativity right now, GPT is the sharper tool, and pretending otherwise would only cost me good work. Pick the best tool per job, then re-check every quarter, because this ranking keeps moving. Build note I never filled a CRM in fifteen years. So I wired one to fill itself. In fifteen years I never kept a CRM up to date. Almost nobody does. The data goes stale because keeping it fresh is manual work, and manual work nobody enjoys simply does not happen. So instead of trying to be more disciplined, I removed the typing. The piece that does the work is an autonomous agent I call Jimmy. Jimmy itself is not the interesting part. How it is wired is, so here is the actual chain. Jimmy sits on top of Claude. Claude is connected to my own knowledge base, the place that holds my context, my methodology, my commercial history. That base is plugged into my live sources: email, my meeting notetaker, LinkedIn, WhatsApp. Information climbs the chain on its own. A call gets transcribed, a thread moves, a mail lands, and the right record updates and enriches itself. I never open a form. Once the data maintains itself, the CRM stops being a filing cabinet and becomes something you talk to. I tell Jimmy what I expect from him, and he turns proactive: who to follow up, what changed since last week, what deserves my attention today. 01 The sources Email, meeting notetaker, LinkedIn, WhatsApp, feeding in live. 02 The brain My context, methodology and history, where Claude reads and reasons. 03 The agent Jimmy on top, updating records and turning proactive. Solo it already runs. In a small team it holds. At company scale it gets heavier, and the honest open question is the line between efficiency, GDPR and privacy. I am turning the whole build into a workshop in September, so you can wire your own. The hard part was never the software. It is the line between efficiency and privacy. Tool to test That privacy line. Tools are starting to draw it. The honest tension in the story above: the more your AI reads your real data, mails, messages, calls, the more you have to answer for what it sees. GDPR is not a footnote. It is the line. Good news. A small wave of open source tools now sits between your data and your model, detecting and stripping personal information before it ever reaches the AI. Presidio · Microsoft The most established. Detects and anonymizes personal data in text and images, customizable to your own entities. github.com/microsoft/presidio → DataFog Lighter, developer first, built to scan and redact PII inside your pipelines. github.com/DataFog/datafog-python → Phileas · Philterd A redaction engine focused on filtering sensitive data out of text at speed. github.com/philterd/phileas → Our take at The Tech Nation. If you are building anything that feeds real customer data to a model, put one of these in front of it now, not after the first incident. Privacy by design is a lot cheaper than privacy by lawsuit. Plug in Building in Barcelona? Here's how to plug in. Four altitudes. Pick the one that matches where you are right now. Level 01 · Claude Community Events The room. Fifth edition, this summer. We host the Claude community in Barcelona. Come for the launches, stay for the people who are actually building. Best entry point if you want to know what is going on. Save your spot → Level 02 · Cheese, Tech & Wine #5 Masterclass format. Smaller room. Hands-on. Around 80 people, one focused theme, real practice. You leave with concrete inputs, not a tote bag. Cheese and wine included. The fifth edition before we break for summer. Reserve a seat → Level 03 · Crewly Cohort #5 Tiny group. Real protos. Built together. Cohorts of 8. Two hours. You bring an idea, we build the prototype together. By the end you have something that actually works. Fifth cohort, last before summer. Apply → Level 04 · Crewly · The Tech Nation A real tech team. At your service. Day one. You want to learn, you also want to ship. You need a tech team behind you, not a course. That is Crewly. Product, tech, AI, delivery, integrated. Plug into it instead of building it. Book a demo → Agenda · before summer 🔨 This summer · Barcelona · half a day Founder AI Buildathon · Karisma × The Tech Nation × Itnig Half a day to actually ship a project and get going with Claude Code. Real teams, a hard deadline, no slideware. Bring an idea or bring your hands, you leave with something that runs. Use our code TTN100 on Luma. Join with code TTN100 → 🛠 This summer · #5 Crewly Cohort Fifth and last cohort before the break. Cohorts of 8, two hours, bring your idea and leave with a prototype that holds. Apply on Luma → 🧀 This summer · #5 Cheese, Tech & Wine Fifth edition before the break. Around 80 people, one focused theme, hands on. Cheese and wine after. Reserve on Luma → This is our last issue before summer. Catch one event before the city empties out, and if you are building something serious, let's talk. See you on the other side of August. You're building something.You want to move faster. Let's talk. LET'S TALK LinkedIn Instagram the-tech-nation.com People · Tech · Impact Got this from a friend? Subscribe →