When an executive asks a question like “What is actually driving customer churn right now?” most data teams face a familiar bottleneck: the decision latency trap. Turning that question into a ticket, prioritizing it in a crowded backlog, building a pipeline, and delivering a report can take weeks. By the time the answer arrives, the business context has already moved on.
At the latest Nordic Data Science & Machine Learning (NDSML) Summit, Theodor Kraft, AI Tech Lead at GlobalConnect, delivered a compelling talk about how one of the Nordic region’s largest digital infrastructure providers is eliminating this friction entirely. The solution was a transition from static reporting to Agentic Analytics.
The Missing Bridge Between AI and Data
While “vibe coding” and AI coding assistants like Cursor and Lovable have transformed software development, analytics has historically trailed behind. Kraft explained: software development operates in predictable, self-contained environments. Data, on the other hand, is inherently messy, contextual, and deeply dependent on evolving business definitions. If you ask an LLM to generate SQL without rich contextual boundaries, it quickly stumbles on table schemas, column meanings, and business logic.
GlobalConnect solved this by integrating LLMs directly with their cloud data warehouse using Model Context Protocol (MCP) tool-use frameworks. By pairing LLMs with a lightweight, tightly maintained semantic layer, they enabled AI agents to discover semantics, understand business context, explore schemas, and execute queries in real time.
The result is time-to-insight collapse from weeks to minutes, enabling executives to get verified data answers during the course of a single meeting.
Three Levers of Value: People, Process, and Product
Integrating AI into the data stack isn’t just about building faster query generators. Kraft outlined how Agentic Analytics delivers value across three distinct horizons: 
- Quality: Assisting non-technical and financial analysts in generating precise, syntactically correct queries that execute reliably without manual data manipulation errors.
- Efficiency: Transitioning data teams out of “ticket-servicing mode” and into true self-service enablement.
- Transformation: The most valuable tier. Senior data analysts with over a decade of experience are now using agents to conduct complex, exploratory analyses they previously never had the bandwidth to execute.
Kraft emphasized that Agentic Analytics is not a quick UI add-on slapped on top of a legacy platform. GlobalConnect has embedded AI throughout their entire engineering lifecycle: from automated CI/CD documentation tools (“AutoDoc”) to assisting data engineers inside VS Code as they build core data models in Snowflake and dbt.
Data Engineering is Still King
Despite the rapid evolution of autonomous agents, Kraft’s message to data leaders was clear: Data engineering is more critical than ever. Agents are only as good as the context they are fed. Building modular, agile semantic models, maintaining clean warehouse architecture, and raising organizational AI maturity remain the essential prerequisites for agentic success.
The Full Technical Plan
This article is part of the NDSML Summit Key Insights Series! Designed for anyone who missed the live sessions, or simply wants to dive deeper into the blueprints presented by top industry leaders, this series brings the summit’s valuable takeaways.
This overview only scratches the surface of GlobalConnect’s production architecture. In the full presentation from NDSML, Theodor Kraft explains:
- The multi-agent framework and tool-use architecture connecting LLMs to Snowflake.
- How GlobalConnect handles data security, governance, and user-level access permissions with AI agents.
- Strategies for overcoming the evaluation and “golden test set” challenge when validating exploratory AI queries.
- How to embed agentic ingestion directly into the lower levels of your data stack.
Watch the full presentation on the Hyperight platform to get the step-by-step breakdown.
Join the Frontline of Applied AI at NDSML Stockholm
The shift toward Agentic Analytics is happening now, and the teams mastering these architectures today are setting the pace for the industry.
Don’t miss the upcoming edition of the Nordic Data Science & Machine Learning (NDSML) Summit in Stockholm. Join hundreds of leading AI practitioners, data engineers, and tech leaders as they share real-world case studies, architectural blueprints, and production-tested strategies that you can apply directly to your enterprise.
Secure your early-bird tickets for NDSML Stockholm and take your place at the forefront of the AI and data transformation.
