Trends, tools and best practices in enterprise AI.
En pragmatisk køreplan til IT-leder eller compliance-koordinator der skal omsætte EU AI Act til handling uden dedikeret compliance-team. De 20 ting, prioritering og hvad der er realistisk.
A pragmatic roadmap for the IT manager or compliance coordinator who needs to translate the EU AI Act into action without a dedicated compliance team. The 20 things, prioritisation, and what is realistic.
De otte kategorier i Annex III gennemgået med konkrete eksempler fra nordisk midmarket. Hvornår er jeres rekrutteringsværktøj, kreditscoring eller OT-system high-risk under EU AI Act?
The eight Annex III categories explained with concrete examples from Nordic midmarket. When is your recruitment tool, credit scoring, or OT system high-risk under the EU AI Act?
Hvad skal en AI-politik indeholde? De otte obligatoriske sektioner, typiske fejl og hvad der adskiller en politique der faktisk bruges fra én der lever i en PDF-mappe ingen åbner.
What must an AI policy contain? The eight mandatory sections, common mistakes, and what separates a policy that is actually used from one that lives in a PDF folder nobody opens.
Forskel og overlap mellem GDPR's DPIA og AI Act's FRIA. Hvornår skal I lave hvad, hvem er ansvarlig, og hvordan undgår I dobbeltarbejde med en koordineret workflow?
The difference and overlap between GDPR's DPIA and the AI Act's FRIA. When do you need which, who is responsible, and how do you avoid duplication with a coordinated workflow?
Hvordan bygger man en AI inventory fra nul? Discovery-processen, 15 spørgsmål per system, ofte oversete AI-systemer (Copilot, Shadow AI), og hvornår man skifter fra regneark til dedikeret tool.
How to build an AI inventory from scratch. The discovery process, 15 questions per system, often-overlooked AI systems (Copilot, Shadow AI), and when to move from spreadsheet to a dedicated tool.
Et politik-PDF gør dig ikke compliant. Her er de fire leverancer, der faktisk rykker — et register, en risikoklassificering, en beslutningsmatrice og en one-pager.
A policy PDF doesn't make you compliant. Here are the four deliverables that actually move the needle — a register, a risk classification, a decision matrix, and a one-pager.
80% af AI-besparelser kommer fra at sende den rigtige forespørgsel til den rigtige model — ikke fra prompt-optimering. En praktisk guide til model routing i produktion.
80% of AI cost reduction comes from sending the right request to the right model — not from prompt engineering. A practical guide to model routing in production.
Anthropics ephemeral cache-rabat er mekanisk simpel, men operationelt svær. Placement-mønsteret, 1024-token grænsen og hvad der kan og ikke kan caches.
Anthropic's ephemeral cache discount is mechanically simple but operationally hard. The placement pattern, the 1024-token threshold, and what can and cannot be cached.
AI-funktioner uden traces er ikke funktioner — de er forpligtelser. Trace-mønsteret, ADR-0001 metadata-felter og EU AI Act artikel 15 i praksis.
AI features without traces are not features, they are liabilities. The trace pattern, ADR-0001 metadata fields, and EU AI Act Article 15 in practice.
You don't need an AI ethics board. You need a register, a risk classification, a decision matrix, and a one-pager. Two weeks, four deliverables.
Every organisation has scattered AI pilots. Few have direction. Prioritisation and an operating model turn experiments into capability.
AI-assisted scoring, capability generation, strategy parsing. APM in 2026 is AI-native or it's a spreadsheet.
ISO 42010 is viewpoint-driven. TOGAF ADM is too slow for the AI era. Modern EA is about stakeholder concerns and decision support — not diagrams.