The Summer Gathering — AI in practice, and where we play next.

The room was full of momentum. Two big raises led the way — Nearfield's, the largest deep tech round the Netherlands has ever seen, and General Intuition's — plus a raise for Leyden Labs, and other community wins were shared along the way. Both the Nearfield and General Intuition founders were in the room.

The mood in one line: this is the best time to build, and the most demanding. AI is pushing people back toward physical proximity and real community — which is exactly how this community is built to work.

The Stack also launched as the home for AI builders in the Netherlands — a highly curated space. Eva was on-site for office and membership questions.

Erwin Buckers, co-founder of Chainels — the tenant-experience platform used across European real estate, spanning 25 countries with 35 people, €3M raised in 2023, and back to profitable — walked us through the real journey from experiment to operating system.

It started small: AI notetaking and translation in January 2025. Engineering skeptics were won over through hackathons. By March 2026, the full commercial team was on Claude.

Onboarding the non-engineering team was the real unlock. A gamified Slack skill (/help my boss wants me to use AI) walked people through 12 steps and MCP connections; finishing it unlocked a champions channel. A visible leaderboard created healthy competition, and champions emerged in week one.

Just three house rules: you own your output — AI mistakes are still your responsibility; no sensitive data in (now being revisited with guardrails); and ask Claude first before asking a colleague. One output rule kept it human: a five-minute brief on top of any AI-generated document, because no one reads longer.

The stack itself: around 30 MCP connectors, with Notion as the hub. Sales reports run through Claude skills. The website moved off WordPress to a Notion database synced to GitHub. Localization dropped from days to 90 minutes across 15 languages. Legal reviews of RFIs, NDAs, and redlined contracts come back in 10 minutes. A shared library of roughly 50 daily-use skills lives in its own repo, managed by the AI lead. And by staying on the Team plan plus API usage rather than Enterprise, they run about 3x cheaper for their size.

Three things made it stick:

Ozgun Erdogan — co-founder and co-CEO of Ubicloud, and previously co-founder and CTO of Citus Data (acquired by Microsoft) — made the case for open source in enterprise deployments, framed around what it takes to build an OpenAI alternative in the Netherlands. His focus: enterprise, not consumer or startups.

The bottleneck isn't model intelligence; any top model can serve with the right deployment. It's everything around the model: capacity and inference serving, safety, tooling, and the harness. The case for open source comes down to price/performance, risk management (a hedge if token access ever gets restricted), and long-term sovereignty and resilience.

Capacity is the hard part — getting sufficient rate limits from providers. The answer is clustering machines together (InfiniBand, speculative decoding) for roughly 10x GPU performance. And an underrated problem: tool-call accuracy sits around 91% on most providers, and errors compound fast across multi-step agent chains. Guardrail models help filter and validate inputs, and a study on Dutch-language safety is on the way.

Two founders, two brand-new unicorns, one message for the room: dream big, and go after the problems everyone else calls too hard.

Iggy Harmsen (General Intuition) put the bet plainly — breadth wins. Take on more of the problem than anyone else dares, and the moat builds itself.

Hamed Sadeghian (Nearfield Instruments) was laughed out of the room when he first pitched, back in 2012. He kept building — and the market that laughed became the market he now leads.

Their advice landed the same way: pick a problem big and complex enough that the market laughs at first — then out-build everyone who doubted you. The hardest problems are exactly the ones worth spending a decade on.

Three themes kept surfacing: sovereignty and building, go-to-market, and talent.

On GTM, we play Switzerland — a neutral connector who gets founders in front of CXOs. Send us your target customers or ICPs and we'll make the introductions.

On talent, an employer branding report is coming. We're working with the Minister of Digital Affairs and the Ministry of Social Affairs on a regulatory carve-out for strategic tech sectors — covering labour costs, dismissal rules, and works councils — with the 30% ruling and Box 3 also in scope. This is longer-term work, and we'll keep you posted.

Circles are expanding. The AI Circle is opening up to CTOs, CMOs, and CPOs — small recurring groups of six to eight peers — and a CFO circle is launching. One or two CTO spots are open now.

City trips in planning: Paris in September, plus San Francisco, Berlin, Stockholm, and China. The goal is simple: get every member on at least one trip.

The real magic wasn't on stage — it was in the room. Founders who'd never met were handing each other the exact playbook the next one needed: how to get a team onto AI, how to price it, who to call. No panel and no deck can replace a peer who's three months ahead of you and happy to share the map.

That's the whole idea — founders backing founders, in the open, with nothing held back. The further each of us gets, the further all of us get. Thank you for being part of it. Keep sharing.

— The TL community team