Data + AI in Practice: From Foundations to Agentic Workflows
A recap from Data & AI Stockholm x Lovable
On March 11, we gathered at Lovable HQ in Stockholm for an evening focused on one of the most important shifts happening in our space right now: how data foundations are evolving to support real, production-ready AI systems.
This wasn’t a high-level AI hype discussion. It was a practitioner-driven evening, bringing together engineers, analysts, and builders working hands-on with modern data stacks and applied AI.
And what stood out most wasn’t just the content, it was the depth of the conversations, the curiosity in the room, and the shared sense that we are collectively figuring out what this next generation of systems should look like.
From Hype to Reality: What It Takes to Build AI That Actually Works
A recurring theme across all talks was clear:
We’re moving from experimenting with AI… to building systems that are expected to deliver real business value.
Saif Al-Zobaydee grounded this in the reality that many teams are facing today: most AI initiatives still fail to reach production or deliver ROI. According to industry insights, a large majority of generative AI pilots don’t succeed, not because the models aren’t good enough, but because the foundations aren’t there.
Instead of focusing purely on models, Saif emphasized three core lenses for evaluating AI systems:
Revenue → Does it actually drive business impact?
Risk → Can we trust it and control it?
Growth → Can it scale sustainably?
The key takeaway:
Data trust is becoming the new currency.
Without reliable, well-structured, and governed data, even the most advanced AI systems will fail to deliver value.
The Rise of Agentic Analytics
Andres Vourakis took us deeper into what many are calling the next evolution of analytics: agentic systems.
Rather than dashboards or static reports, we’re moving toward systems where users can simply ask questions and receive answers, powered by AI agents.
But as Andres showed, this shift is not just about adding a chatbot on top of your data.
It introduces a completely new set of challenges.
For example, when a stakeholder says:
“I need to see our revenue”
That question is actually incomplete.
To answer it correctly, a system needs:
Metric clarity → What definition of revenue?
Slice clarity → Which segment (market, product, customer)?
Time clarity → Over what period?
This highlights a key problem:
Humans are ambiguous. Systems cannot be.
To solve this, Andres introduced patterns like question validation, where AI systems evaluate whether a query is well-defined before executing it, improving reliability and trust.
Why Semantic Layers Are Making a Comeback
One of the strongest signals across the talk was the resurgence of semantic modeling.
We’re seeing a clear shift:
From dashboards → to conversational analytics
From fragmented logic → to centralized definitions
From human-only systems → to AI-consumable data layers
As highlighted in the presentation, the industry is moving toward:
Universal semantic layers with centralized metric definitions
AI-driven reasoning through natural language queries
Embedded governance to ensure correctness and trust
This is what enables what many are calling Self-Service Analytics 2.0 — where AI makes data accessible, but strong foundations ensure the answers are actually correct.
The Hard Part: Building Systems That People Trust
While the vision is exciting, the reality is complex.
Some of the biggest challenges highlighted:
Interpreting vague user intent
Maintaining and evolving semantic layers
Querying across multiple datasets reliably
Ensuring correctness — because even small AI errors erode trust quickly
This last point came up repeatedly throughout the evening.
Trust is fragile.
And in AI systems, being right most of the time is not enough.
A Shift in the Role of Data Teams
Perhaps one of the most important insights from the evening was not technical, but organizational.
The role of data teams is changing.
We are moving:
From dashboard builders → system designers
From ad-hoc support → product thinking
From serving humans → serving both humans and AI agents
This also means new responsibilities:
Designing systems with guardrails and governance
Creating clean, well-defined data models
Enabling safe access for AI systems
In short:
Data is no longer just an asset, it’s becoming a product.
Beyond the Talks: The Power of Community
What made this event special wasn’t just the presentations, it was what happened after.
During the group discussions, speakers joined the tables, and conversations went deeper:
How do you actually implement semantic layers in practice?
Where should responsibility sit between data and engineering teams?
What does “good enough” reliability look like for AI systems?
These weren’t theoretical debates, they were grounded in real experiences from teams actively building in this space.
If there was one theme that connected everything from the evening, it’s this:
The future of AI will not be decided by models — but by the quality of the systems we build around them.
Agentic workflows, conversational analytics, and AI-driven systems are already here.
But their success depends on something much less flashy:
Clear definitions
Strong data foundations
Thoughtful system design
And above all, trust
Explore the Talks
If you’d like to explore the topics further, the presentation slides from our speakers are available here:
Andres Vourakis, The Road towards Agentic Analytics
Saif Al-Zobaydee, Building the Data Engineering Foundation for Revenue-Focused AI
Ece Kural, Rethinking Data for Agent-First Products
Stay Connected
If you’d like to stay connected with the community and hear about future events, the best places to find us are:
Discord → where community discussions continue between events (and where we post first access to event sign-ups!)
LinkedIn → where we announce upcoming events and share updates
YouTube → where we post podcasts and interviews with professionals and thought leaders in the Data & AI space
The Data & AI Stockholm 💙