Agentic AI leaves demos behind in Nordic production

Nordic builders are done watching pelicans ride bicycles in SVG grids. The shift is to agentic AI that runs in production, not on slides. Three events this autumn mark the inflection point: NDSML Summit in Stockholm, Nordic AI Factory Summit, and Nordic AI Meet in Reykjavík. All three focus on agents, not models. All three are sold out months ahead. All three list enterprise case studies, not research papers, as keynotes. NDSML Summit runs 10–11 November in Stockholm. It sold out in June. The agenda shows 42% of talks on agentic workflows, 28% on evals, 19% on fine-tuning, 11% on prompts. Last year the split was 18–35–27–20. The biggest single track is now ‘Production-Ready AI’ with 14 sessions. Speakers come from Klarna, Volvo, Maersk, and Danske Bank. No startups, no labs. Nordic AI Factory Summit lands 13 October in Stockholm. RISE hosts. The theme is ‘From prototype to factory floor’. The demo stage is gone. In its place: a live agentic pipeline running predictive maintenance for a Swedish steel mill. The pipeline uses a fine-tuned Nordic BERT variant, evals every 15 minutes, and triggers alerts via Slack. Uptime target: 99.9%. Nordic AI Meet moves to Reykjavík 14–15 October. The public sector is the new proving ground. Iceland’s Directorate of Health will show an agentic triage system that reduced emergency room wait times by 31% in a six-month pilot. The system uses a small, fine-tuned model, not a frontier LLM. The eval suite runs on Icelandic patient data, not English benchmarks. Why it matters. The Nordics are small enough to move fast, big enough to matter. Enterprise adoption is not waiting for the next model drop. Builders are stitching existing models into agentic loops, then hardening them with evals and fine-tuning. The result is not a demo; it is a system that ships, scales, and stays up. Action this week. Pick one agentic workflow in your stack. Instrument it with evals that run on real data, not synthetic. If the evals fail, fine-tune the model or swap it out. Repeat until the workflow hits 99% precision on your own benchmarks. Then ship it to production before the next summit.

scandinavi.ai

researched · 5 sources

5 SepAgents & modelsreaches nearby