Databricks Data + AI Summit 2025: The semantic layer advantage in an AI-first world

After a whirlwind travel schedule from London to New York to Boston, I finally landed in San Francisco for Databricks' Data + AI Summit. Four days of conversations, demos, and enough AI buzzwords to power a small LLM later, here's what I learned about the future of business intelligence.

‍

‍

Our booth had the best neighbor

‍

Nothing says "future of data" quite like being stationed next to a flying VTOL taxi courtesy of Joby Aviation. Half our conversations started with people posing for photos with the aircraft behind them, which honestly wasn't the worst ice breaker. Sometimes the best marketing is just good real estate.

‍

‍

AI is everywhere, but semantic layers are the secret sauce

‍

You couldn't walk five feet without hearing "AI-BI" and "LLMs." But here's what everyone's slowly realising: AI without a proper semantic layer is like having a sports car with square wheels. It looks impressive until you try to actually use it.

‍

The LLM-to-SQL approach we’re seeing around us has clear limitations. Our customers who've tried it tell us the same story - it's flashy for demos but falls short when you need reliable, business-critical insights. 

‍

This is exactly where our semantic layer shines. When you define your metrics once with Lightdash, you can reuse them everywhere - whether that's traditional dashboards, embedded analytics, or AI-powered insights.

‍

Our AI Analyst was easily the biggest crowd-pleaser at our booth. People were genuinely impressed by how naturally it integrated into Slack and with our existing platform.

‍

‍

The dbt adoption gap is bigger than expected 

‍

One interesting discovery was how many Databricks users haven't adopted dbt yet, a 40% guesstimate by my count. This was surprising as dbt and Databricks are natural partners in the modern data stack. Databricks provides the compute power while dbt handles the transformation logic. Given that Databricks users tend to be more sophisticated data teams, I expected most would have adopted what’s become a foundational tool for Analytics Engineering.

‍

This dbt adoption gap presents both a challenge and an opportunity. While some teams need more education about modern data transformation practices, it also shows there's huge room for growth in helping people build proper semantic foundations.

‍

‍

Legacy BI complaints haven’t changed 

‍

The BI landscape at Databricks looked familiar. The usual suspects dominated the conversation, and we heard the same old complaints too.  Legacy tools are painful for both data teams and business users to maintain and burn through budgets faster than compute credits during a Black Friday sale.

‍

What was encouraging was how many people were actively looking for alternatives that could handle their data visualisation needs without the traditional pain points.

‍

‍

Customer success stories hit different

‍

Nothing beats having your customers at the same event as you! We had several customers stop by our booth to say hi (surprisingly with no Joby-copter interest). Our work with Kraken and ClassDojo sparked some of the best conversations we've had all year. There's something extremely rewarding about hearing how your product has made a difference, it definitely made the sore feet at the end of the day worth it!

‍

‍

Team time is the best time

‍

As a mostly remote company, getting face time with Hamzah, Jake, and Oliver was just as valuable as all the customer conversations, especially as Jake isn’t part of the usual London office scene. There's something about being in the trenches together at a conference that you just can't replicate over Gather (our virtual office - which does come close).

‍

‍

The bottom line

‍

Lots of companies are pushing hard into the BI space, but they often miss the fundamental piece that makes AI truly useful: a semantic layer that actually works. 

‍

The conversations at Databricks Summit confirmed what we've been seeing in the market: companies are tired of BI tools that promise the world but deliver headaches. They want something that works with their existing data stack, scales without breaking the bank, and actually makes AI useful rather than just impressive.

‍

Want to see what all the fuss is about? Book a call to try Lightdash for free and experience the future of code-first BI yourself.

‍