“Self-service BI is broken. Even when it works.”

The Vision: AI-Driven Data Foundations That Actually Deliver

Let’s be honest: most self-service data setups are clunky, expensive, and require way too much human babysitting. We’ve all seen dashboards multiply and then rot, and data teams burn hours maintaining them.

This newsletter is about changing that.

I’m on a mission to build a 10x data team and data infrastructure: a system that reduces build-up and operational costs by 90%, and increases the delivered business value by 10x.

This update is a behind-the-scenes look into why I believe this is possible, what I’ve learned so far, and how I’m building it with AI.

Why This Matters

Whether you're leading a data team in a fast-moving startup, a growing scaleup, or a more mature organization, you’re facing similar core challenges:

• Building a reliable data foundation that scales
• Empowering stakeholders to get answers without constant support
• Delivering insights that drive action, not just filling dashboards

This newsletter is here to help you navigate those challenges with real, tested strategies. The focus is on increasing your team’s impact without overextending your resources, relying on bloated tooling, or getting stuck in endless maintenance loops.

The Backstory: Productizing Data Foundations in Scaleups

Over the past 17 years, I’ve built and led data teams across high-growth companies, mostly VC-backed scaleups and PE-backed consumer companies in the $50M–$1000M revenue range.

In every case, the need was the same: build the data infrastructure fast, make it usable by business teams, and keep it flexible as the business grows.

After building data foundations as in-house data leaders for a few years, I started building them as a freelancer on a daily/hourly rate. Then I productized my work:

• Reusable frameworks
• Blueprinted systems
• Prebuilt transformation pipelines
• A team of freelancers who could execute like clockwork
• Fixed pricing without surprises

It worked. I could build the first data foundation in fast-growing scaleups in 8–15 days and establish some degree of self-service among stakeholders.

But I wasn’t satisfied.

Everything is systematized, repeatable and blueprinted - why shouldn’t I be able to get this down to 1 day?

The answer: I should. And I believe AI will get me there.

The Missing Semantic Layer

However, I had to make a decision:

Should I focus first on making myself more efficient in building up the data foundation or should I focus on making the interaction of stakeholders with the data foundation more efficient and valuable?

I decided for the second path.

My setup always had one flaw: the absence of a true semantic layer.

My productized setup delivered this:

• Stakeholders access data marts with well-defined metrics in BigQuery via Google Sheets
• Data is granular, clean and structured
• KPIs and Dimensions are defined on a unit-granular level (customer, transaction etc)
• But the final step, business-friendly dynamic aggregation across dimensions, is missing

What happens then?

• Aggregate tables multiply (daily management dashboard, weekly management dashboard, monthly management dashboard)
• Since metric and dimension definitions were pre-defined on a unit-granular level, definition-drift didn't explode but it surely happened
• Self-serve within a governance framework using Google Sheets on top of the DWH worked well for some stakeholders but not for everyone

I’ve always avoided semantic layers because they’re expensive to build and harder to maintain.

But they are absolutely essential for building analytics agents.

The Goal: A Semantic Layer Built for Speed, Simplicity - and AI agents

To make the interaction of stakeholders with the data foundation more efficient and valuable, I’m now focused on two things:

1. Rapid Semantic Layer Construction

I’ve started experimenting with a tool called Connecty AI, which promises something bold: a Day Zero semantic layer.

Their goal is to become the world's first fully autonomously created semantic layer.

Here’s the concept:

• It scans your data warehouse metadata
• Suggests a semantic model automatically
• You can refine it via natural language
• It self-heals and keeps itself updated

It’s not fully working yet, but this is the direction I believe in.

2. Making That Layer Accessible via AI Agents

Once that semantic layer is in place, I want to expose it to AI agents (something that Connecty doesn't support at the time of writing).

But I don't want these agents to just answer questions. I’m thinking in three stages:

→ Stage 1: Reactive Agents

Basic Q&A style. The agent sits on top of the semantic layer and answers questions like:

• “What was our revenue yesterday?”
• “Which channel had the highest CAC last week?”

This is just the beginning.

→ Stage 2: Proactive Agents

These agents:

• Know your business goals
• Understand your active initiatives
• Monitor metrics autonomously
• Alert you when something's off

Example: If the goal is to raise conversion from 2% to 3%, and a landing page isn’t pulling its weight, the agent tells you before the marketing team notices.

For this to work, the agent needs access to both:

• Structured data (metrics, KPIs)
• Unstructured context (Jira, Trello, Slack, emails, Confluence etc)

→ Stage 3: Autonomous Agents

These agents don’t just analyze. They act.

For example:

• Spot an underperforming landing page
• Design a new variant including all the ad copy
• Set up an A/B test
• Roll out the winner automatically

This is the frontier. We’re not there yet, but that’s where I want to go.

Lessons So Far: Where the Leverage Really Is

In dozens of conversations with other data leaders, a pattern emerged:

There is leverage in helping engineers write code. But most leaders see more leverage in helping business stakeholders help themselves.

My feeling is that this view is slightly biased because many leaders have never seen self-service work.

But in my opinion, even well-working self-service setups are flawed.

In 2026, no one should manually build analyses, pivot tables and ad-hoc queries anymore.

I'm not there yet but I'm getting close. And it starts with one thing: a robust, rapidly built, maintainable, AI-ready semantic layer.

Bottom Line: The Real 10x Isn’t in Speed. It’s in Access.

A possible 10x comes from building a data foundation in 1 day, not 10.

But the real 10x comes when stakeholders are self-sufficient. When they can get answers fast, reliably, and in plain language because your semantic layer powers a data agent that works like a colleague, not a dashboard.

That’s where I’m heading.

Next steps for me:

• Continue testing Connecty (and similar tools) for Day Zero semantic layer capabilities
• Building a stage 1 agent on top of that semantic layer
• Progressing that agent through the 3 stages

And I’ll keep sharing what works, what breaks, and what gets us closer to the data stack of the future.

This newsletter was still a little high-level as I haven't had the time to go really deep. In my next episode I try to go a little deeper into the nitty-gritty details.

Until next time—
Stay sharp.

Sebastian

P.S.: This is not a sponsored post. I don't receive any kickbacks from Connecty and I am sharing my neutral, unbiased views.