We’ll open with a live demo of an agentic workflow running on real investment data at Norges Bank Investment Management - and then spend the rest of the talk explaining how it’s actually possible. The answer isn’t a clever AI trick: it’s a governed data catalogue, clean data models, and an MCP server layer that gives LLMs reliable, machine-readable context. If you’ve tried putting AI on top of messy data and wondered why it didn’t work, this talk is for you.

NAF testet Databricks Genie-agenter på tvers av to domener for å gjøre BI og analyse tilgjengelig gjennom naturlig språk for forretningsbrukere. Nøyaktigheten ble i hovedsak drevet av eksplisitt forretningslogikk, som måltall, filtre, eksempler og instruksjoner, snarere enn generell metadata. NAF arbeider nå med å optimalisere implementeringen av agentene for å støtte organisasjonens strategiske satsing på medlemsengasjement. For å lykkes kreves en todelt tilnærming: å bygge kompetanse hos medarbeiderne i bruk av det nye verktøyet, samtidig som agentenes nøyaktighet og pålitelighet sikres.

This talk presents MUNDAT, Trondheim Kommune’s DataPACT use case on compliant municipal data sharing. We introduce the concept behind LexAlign, a compliance assessment framework combining machine-readable regulatory decision flows, structured system evidence, and and LLM-supported compliance assessment. The talk explores how this approach could be applied in the MUNDAT workflow.

When we used an LLM-generated topic field to narrow document search for a retrieval agent, we discovered the topics were badly broken — one topic covered 40% of documents, duplicates existed under different capitalisation, and over 18,000 near-unique topics had been generated. The agent never errored; it simply retrieved from a skewed subset and quietly gave worse answers. This talk walks through the failure, our LDA-based fix, and the broader governance lesson: AI output feeding another system needs the same scrutiny as any other data source.

This is a hands-on workshop, where you'll spend most of the time actually doing data management. We'll work through real problems in small groups, using the DMBOK as your guide. Along the way you'll see how the pieces of the framework fit together, and why this work matters more than ever now that AI runs on the data you feed it. Later we'll show you the opportunity to become a Certified Data Management Practitioner. The CDMP is the recognized global certification, built on the same DMBOK you've been practicing with during the workshop. We'll explain what it is, why it's worth it, and go through a few mock exam questions together.

This workshop covers how traditional data delivery methods are no longer fit-for-purpose in an AI-driven world. Learn how to provide LLMs with a unified and secure access point to query real-time enterprise data by integrating Denodo's logical data fabric and data virtualisation technology.

Despite heavy investments in traditional data tools, achieving true data literacy has remained elusive. Strawberry is bridging this gap with ""Data Concierge,"" a custom LLM-powered chatbot. The solution enables conversational analytics by layering Google’s Gemini over PowerBI semantic models. The presentation explores these architectural choices and features a demonstration of the tool in action.

Large language models know a great detail about the world, but not enough about your company to be an intelligent partner. What is missing is the context and harness that puts just the right amount of context at the right moment, and an evaluation loop that continuously improves the agent over time. This talk shows how context engineering and evaluation feedback closed that gap for an AI agent used daily by analysts at a Norwegian marketing agency (Brist) with over 5,000 agentic runs since January 2026.

We have built an AI based web application that automates the full process of Occupational Health Service Checks. We are combining health data, client surveys and on site audits with government directives, medical journals and legal documentation to produce professional statistical analytics and customer recommendations to improve safety, wellness and health for the worksites across the Nordics.

Databricks has put a lot more focus on business users and analytics this past year, and there is a growing ecosystem of tools. In this session Julia Førde (Databricks MVP, Senior Consultant at Knowit) walks through what's actually behind Genie One. Genie Ontology, Pages, Domains and Skills, with livedemos of how it all works together, but also what it takes on your end to get good products and answers.

AI models are brilliant at predicting what's probable, but your business runs on what'sallowed.
The AI Harness closes that gap by connecting your models to trusted data, governedcontext and controlled actions.
Learn how to assess your AI readiness, fix gaps in data quality and master data, and cut the hidden cost of AI.
Leave with a practical framework for moving AI agents from pilot to production withconfidence.

Mange tror at data eies på samme måte som fysiske gjenstander. Det stemmer ikke. Lover og regler gir overraskende svak beskyttelse av data, som betyr at kontroll som oftest etableres gjennom kontraktsregulering og tekniske sikkerhetstiltak, ikke bakgrunnsrett. Dette foredraget forklarer hvordan data beskyttes, og hvordan man kan strukturere avtaler slik at man har kommersiell kontroll over de dataene bedriften din genererer og bruker - for eksempel hvordan programvareselskaper kan sikre rett til å bruke data for maskinlæring og produktforbedring, og hvordan man kan sikre best mulig uttelling i kommersielle samarbeid hvor data er en sentral verdidriver. Foredragsholderne har jobbet i advokatbransjen og i teknologibransjen in-house (i henholdsvis Cognite og reMarkable), og de skal gi praktiske tips og takeaways som tilhørerne kan benytte direkte i egne kontrakter og strategier.

We began with a seemingly simple question: could AI help improve a complex business process involving many stakeholders, competing incentives, and no ground truth? What followed challenged many of our assumptions about AI, trust, and decision-making. Through experimentation and a few unexpected discoveries, we learned that trust and explainability weren't constraints on the solution. They became the solution.

When people ask ChatGPT, Copilot, Gemini or Claude for a recommendation, the
engines pick a handful of companies and leave the rest out. We trace that
conversation end to end, across close to 100,000 AI answers: what the engines
read, which sources they trust, why the four engines disagree, and what decides
whether a brand is chosen or passed over. Examples from Norwegian e-commerce and
B2B.

1 times workshop for ledere, produkteiere, teknologiansvarlige og andre som er nysgjerrige på strategisk teknologiledelse.

Mange virksomheter som investerer i KI, men opplever at det er krevende å skalere og ikke minst skape verdi av investeringene. Selv om utviklingsteamene leverer raskere enn noen gang, står tiden fra idé til faktisk forretningsverdi ofte stille. Flaskehalsen ligger ikke lenger i teknologien, men i organisasjonen, beslutningene og måten vi leder endring på.

I denne workshopen utforsker vi hva som kreves for å lede strategisk i KI-paradigmeskiftet. Hvordan gjenkjenner vi at organisasjonen har blitt «proppen i systemet»? Hvilke strukturer, insentiver og ledelsesgrep må endres? Og hvordan flytter vi fokus fra å gjøre dagens oppgaver litt raskere til å utvikle organisasjonen?

Takeaway: Deltakerne sitter igjen med ulike ledelsesgrep som frigjør verdiskaping, og refleksjoner rundt hvordan egen virksomhet kan skape større effekt av KI-investeringene.