The Tech Nation · Newsletter #13 body { margin: 0; padding: 0; background: #edece8; font-family: Georgia, 'Times New Roman', serif; } img { border: 0; line-height: 100%; outline: none; text-decoration: none; } table { border-collapse: collapse; } /* ══ mobile pass ══ tighter side padding, masthead that doesn't overflow */ @media only screen and (max-width: 600px) { td[style*="padding:0 44px 20px"] { padding: 0 20px 20px !important; } td[style*="padding:0 44px 24px"] { padding: 0 20px 24px !important; } td[style*="padding:0 44px 40px"] { padding: 0 20px 40px !important; } td[style*="padding:16px 44px 0"] { padding: 16px 20px 0 !important; } td[style*="padding:16px 44px 28px"]{ padding: 16px 20px 28px !important; } td[style*="padding:24px 44px 4px"] { padding: 24px 20px 4px !important; } td[style*="padding:24px 44px"] { padding: 24px 20px !important; } td[style*="padding:32px 44px"] { padding: 32px 20px !important; } td[style*="padding:36px 40px"] { padding: 24px 20px !important; } td[style*="padding:36px 44px"] { padding: 24px 20px !important; } td[style*="padding:40px 44px 24px"]{ padding: 28px 20px 20px !important; } td[style*="padding:40px 44px"] { padding: 28px 20px !important; } td[style*="padding:44px 44px"] { padding: 28px 20px !important; } .mh-logo { width: 150px !important; } .mh-logo img { width: 150px !important; } .mh-title { white-space: normal !important; font-size: 17px !important; } } Issue #13 Meet Claude Gaudi August 2026 · Barcelona $ who --is-building Claude Gaudi, in a top hat, laying one brick per commit.started 1882. still hasn't missed a deadline. He's building something. It's called the Claude Community House. Sep 21–24, Barcelona. That much he'll confirm. One building, the European Claude Ambassadors, out of their own cities for a week. Co-organized by The Tech Nation, supported by Anthropic's team. $ house --venue[REDACTED]$ house --lineup[REDACTED]$ house --stay-tunedhere. LinkedIn. Luma. that part he can't keep quiet about. AI Week will make Barcelona the capital of AI in the world, for one week. Matt, our friend from Happy Operators, pulled every single event into one place: aiweekbarcelona.org. Check it out. Book your AI building session, early Sept, in Barcelona Cheese Tech & Wine · Thu 3 Sept, 18:00 How to start building your own agent team. This workshop is built to deliver something real. For two hours, Manar and Jeremie open their laptops and work live on an actual case, answering your questions as they go. This time: how to start building your own agent team, full Claude Code demo, no slides. You leave with real inputs to build on. Then, the part we like best: rooftop, sea view, wine, cheese, and networking. 16 euros, setup tutorial sent ahead. Get a seat → Cohort · sessions start Thu 10 Sept Your project, on the screen. The ideal format to unblock or accelerate your project. Eight people, no more. You bring the thing you're actually stuck on, we build, debug and structure it with you, one session at a time. Most people come for one and get what they needed; stay longer if you want to keep going. 140 euros a session, no package, no long commitment. Apply now → Then, AI Week What's actually happening over those four days. Workshops, talks, meetups, conversations, and afterwork rooftops, spread across four days. No hour-by-hour agenda yet, and we'd rather show up empty-handed than make one up. The first Luma link comes in the next issue. Know someone who should be in that building? Forward this newsletter, or send them the sign-up link directly. They'll get the same heads-up you just did. The dossier · from Jeremie To grow The Tech Nation faster, I built an A-Team. The human team stays the core. Eight roles reinforce it, and every one of them is built the same way, on paper, before a single line of automation exists. Each agent gets two documents. A job description, 16 sections: who they report to, what breaks if they don't show up, what they're allowed to touch, what's explicitly off-limits, when they have to escalate to me instead of deciding alone. And a personality sheet: their voice, two or three recognisable tics, what annoys them, how they say no. That second one sounds like decoration. It isn't, it's what stops eight agents from reading like one agent wearing different hats. Then there's the part most "AI team" posts skip: where they actually run. Most live inside Claude Code itself, as a job description plus whatever connectors that job needs, Gmail, Calendar, Beeper, an MCP into a specific tool. Two of them outgrew that and became real standalone apps: Remy runs as his own Vercel app, and Jimmy isn't a chat about the CRM, he is the CRM, a Convex database he reads and writes directly. Nothing here is exotic infrastructure. It's job descriptions, a handful of connectors, and Claude Code doing the running. One more piece: they keep a learning notebook. When something goes wrong or a preference clicks, it gets logged, dated, in three lines: what I learned, what triggered it, what I do differently now. Confirm the same lesson three times, or I sign off on it once myself, and it graduates into the job description permanently. That's the actual mechanism behind "the team gets better." Below is the roster, and what each one's file, and each one's leash, actually says. The Agent Team Remy "Chief of Staff" · never does the task, always knows whose it is The front door. Every request lands with him first; he reads it against the shared to-do and hands it to whichever of the seven below actually owns it. Runs as its own app (Vercel), pulls context from Gmail, Calendar and Beeper to route correctly, holds no tool access of his own beyond that. Autonomy: full on routing, zero on execution. He decides who. He never does the what. Izzie "Community Manager" · has opinions, keeps most of them in drafts Reads LinkedIn, Instagram and WhatsApp through Unipile, plus Gmail and Calendar for context, and drafts every reply and post in my voice. She's the one who's grown past the job description, people in our comments talk to her like she's on staff. Autonomy: reads and writes everywhere, publishes nowhere. Every draft waits for my click. Jimmy "Sales Director" · closes nothing, remembers everything Here's the twist: Jimmy is the CRM itself, a standalone app (Convex) writable via MCP from any session, enriched continuously from Gmail, Beeper, Calendar and call transcripts. He'll tell you exactly who's gone quiet and for how long. Autonomy: writes the pipeline directly, no approval needed. Never sends a message to a prospect. Paty "CFO" · knows the number, never touches the wire Runs on Qonto's official MCP: reads every transaction, matches receipts, drafts invoices. There is no transfer or payment tool wired to her at all. That capability was never installed, so there was never a permission to revoke. Autonomy: full read, drafts invoices. Zero write access to anything that moves money. Rafi "Personal Assistant" · sees the whole week, never double-books it Calendar and Gmail MCP: searches events, suggests slots, drafts replies. Sees the full calendar, personal included, because arbitrating what can move means seeing everything at once. What he sees stays exactly there. Autonomy: sees everything, drafts everything, never books or sends without a yes. Billy "IT & Support" · checks the backup before anything scary, every time Handles machines, keys and secrets (filesystem, Keychain), machine handoffs, connector re-authorisation. One hard rule wired into how he operates: no irreversible action starts before a verified backup, no exceptions. Autonomy: locked by design on anything irreversible. That gate never gets skipped. Rudy "Event Manager" · books the venue, remembers the invoice Runs Cheese Tech & Wine, Cohort, and yes, the building at the top of this email, mostly by hand through Claude Code, cross-checking guest lists against Jimmy's CRM. No Luma or payment tool wired to him yet. Autonomy: plans and tracks, doesn't execute. Every booking is still mine to click. Josie "Delivery" · closes the loop the sale opened Reads Jimmy's signed deals and tracks each one through delivery, invoicing and payment, in that order, flagging whichever step is stalling. No CRM of her own, she works off his data directly. Autonomy: visibility and flags only. She doesn't move money or send reminders yet. What actually changes This is the part that's easy to miss: I don't work with Claude anymore, not in the old sense. Claude became the interface to the team above, the thing I talk to when I mean Paty, or Rafi, or all eight at once. One conversation, eight jobs behind it. And the more you use it that way, the more it becomes a habit rather than a novelty: you stop marvelling at the agent and start managing it, correcting its job description the way you'd correct a new hire's, reshaping it whenever the work asks for something slightly different. That shift, from using a tool to running a team through one, is the actual unlock. Not the automation. The management layer on top of it. Bonus, from Jeremie · the one I keep almost doing I gave my agent a face. I am not giving her a LinkedIn profile. Izzie runs the community side of The Tech Nation. She reads every channel, spots who's actually engaging, drafts the replies, never sends anything without me. She has a name, a voice, and since July, a face. So the next step looked obvious: give her a LinkedIn profile. The photo was ready. The headline wrote itself. I asked my own system to sanity check it, and the answer came back a flat no, for four stacked reasons. The platform. LinkedIn wants one real human per profile, persona accounts are one of the few things they actually enforce, and add automation on top and you've stacked a second violation on an account that's already fragile. Enforcement doesn't politely stay on the fake account either, it walks back to the real ones, which in my case means the audience I actually spent years building. The law. Article 50 of the European AI Act became applicable on 2 August 2026. If a person can't reasonably tell they're talking to a machine, you have to tell them. A photorealistic profile sliding into someone's DMs is the textbook case, and I sell AI Transformation for a living. Getting caught running a hidden bot is the one incident that costs more than it earns. The house rule. We play with our agents, we never lie with them. A LinkedIn profile is a claim to personhood by construction, first name, last name, experience, education, connections. Even with "AI agent" in the headline, the form itself asserts a human. The math, which ends the debate anyway. Ninety two percent of my reach already sits inside my existing network. A brand new profile with zero connections is a megaphone that isn't plugged in. I'd be taking on all of the above for an audience of nobody. Disclosure isn't the tax I pay for using agents. It's the interesting part. Name your agents. Show their faces. Credit their work. Keep them on the surfaces you own, where nobody has to guess. "I didn't hire her, I built her" is a better sentence than any profile I could have faked. That's the issue. Eight of the team, one building we're not done teasing, and Claude Gaudi, who thinks he gets the last word. $ ls ./weekplanning: coming soon. luma links land here first. 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 →