Databricks Data + AI World Tour London
Over 7,000 people registered for the London stop of the Data + AI World Tour this month. EMEA GM Samuel Bonamigo called it Databricks’ biggest regional event and its second biggest anywhere, behind only the annual summit in San Francisco. One of the lead announcements was a $3 billion investment across Europe and the Middle East over the next three years. It includes a 50% increase in the UK workforce and a commitment to train 500,000 people across EMEA in data and AI by 2030, and a new European headquarters in London that will house more than a thousand Databricks staff. Behind that, there was more to learn about a product called Genie that has grown into a family of agents, and a strong line-up of customers already running those agents in production.
Genie – the next generation
Genie started as a conversational interface for querying data in natural language. Twelve months on, it is a product collection. Genie One is the business-facing agentic co-worker, generally available on web, iOS and Android since June. Genie Code is the agentic data engineering layer. It writes pipelines from natural-language prompts and, by Databricks’ own figures, is now responsible for upwards of 70% of new pipeline code on the platform. Genie Agents turn conversations into reusable workflows. Genie ZeroOps sits in the background monitoring pipelines, jobs, tables and models, proposing fixes when something drifts. Underneath is Genie Ontology, which has reached general availability this month and is now enabled by default. Ontology is the semantic layer that teaches each agent what the business’s own data actually means before it answers any question, and puts that data, history and IP in a form agents can use. Databricks SVP of Field Engineering and co-founder Arsalan Tavakoli-Shiraji set the tone during the keynote:
Models are becoming commodities. Your competitors have exactly the same models. Your advantage is your data, your history, your IP, your people.
Richard Shaw, Technical GM for UK and Ireland, is confident enough about Databricks’ internal use of the product to lead with it:
We run Databricks on Genie today. Genie One, even since June, has been evolving very rapidly. We’ve all been naturally gravitating to using Genie because it’s become that much more effective. Being able to now produce slides and documents – it’s bringing a whole different spin on what Genie was originally considered for. And to now doing the ontology, it’s measurably different how we’re operating Databricks now.
Shaw is most interested in the shift from asking to doing – “Genie One, and Genie Agents, is also about creating workflows, which changes processes”. He sees workflows as where customers move from pilot to production. His sales teams now routinely bring Genie up on a phone mid-customer conversation. Shaw gives an example from the evening before the London event when a colleague pulled Genie up during dinner with an executive at a franchise business. The executive asked which Databricks customers were themselves franchisees. Genie answered, with the size of each business – “because it had the context, it gave more information around the scale of those operations”.
Most customers already have something nominally agentic in place, Shaw notes:
Many customers will have something in place today by one of our competitors or partners. They’ve gone, okay, everyone in the organisation should use this. And what we’re seeing is when they truly see what Genie can do, you get that light bulb moment. Our ambitions and our aspirations for them is that every business user could be using Genie – and doing what they can do with it that’s different from the other tool.
There was a live-demo of this during the event, using story came on the main stage, from Amber Roberts on the Databricks AI team, who walked the audience through a fictional company called BobaBricks. BobaBricks had run through $10 million of coding-agent budget in four months before its finance team noticed. Amber turned on smart routing in Unity AI Gateway – which Arsalan described as “one control plane for your entire AI estate”, not just the agents running inside Databricks. Simple tasks like drafting Slack messages moved from frontier models (Fable 5.2, Claude Opus 4.5, GPT-4o, Astra) to Claude Haiku 3.5. The reported outcome was a 12% spend reduction, roughly $150,000 a week, holding steady, with no tool taken away from any engineer. Arsalan threw in the deadpan observation that Fable 5.2 on high effort was “absolutely needed to draft my emails” which may be a bit too relatable judging by the emails I receive these days.
The second-generation Genie has the same governance pattern built in by default. Databricks describes the new generation as an always-on autonomous agent for enterprise work, one that takes ownership of the work, delegates to subagents when needed, tracks progress, follows up on missing inputs, and reaches out when a person needs to make a decision. A new inbox highlights what needs attention, and Genie emails users when they are away. Every action is governed within Unity Catalog permissions and Unity Gateway approval rules and budgets, and is fully auditable.
Shaw credited the ontology layer as the piece that makes the shift workable:
Through Unity Catalog with Genie Ontology, agents can retrieve answers rather than guess them. Ontology is helping move it from a GenAI product that’s referring to the relationships – and helping it give high-quality answers – to becoming the full ontology for the business, thinking about the relationships, and ultimately really changing how a business operates.
Another announcement was LTAP – Lake Transactional/Analytical Processing – which writes transactional data in the row form transactions expect and transcodes it into columnar form for analytics underneath, on one copy of the data. Lakebase, the serverless Postgres database LTAP sits on, reached general availability on Google Cloud on 2 November, completing a multi-cloud rollout. To customers who have spent years building ETL pipelines between transactional systems and the lakehouse, Shaw observes:
You can seamlessly go from operational data to analytical processing without all those pipelines.
Where the UK money goes
Of the $3 billion investment, $850 million was already earmarked for the UK back in March. Shaw explains:
The investment isn’t just in the buildings, because that allows us to, from what I expect, hire more people, build our teams out. The new office in London will be able to support two thousand people, from where we are around five, six hundred today. And a lot of what’s driving that is not just the health of Databricks and our growth. When you look at the products like Genie One, customers are looking at Lake Base and going, ‘great, we have this fantastic evolution of Postgres’ – and the investment is absolutely to fuel that next stage of growth.
The 500,000 EMEA skills target by 2030 broadens the UK-only 100,000-by-2028 number announced this time last year. Shaw emphasized that “the investment back into people will continue” – which makes sense given that field engineering sits with a customer for long enough to help them get agents into production rather than into another pilot.
Some of the UK digital-native customers included Skyscanner and Octopus Energy. The broader UK listDatabricks has published across 2026 includes Unilever, Rolls-Royce, Nationwide, Virgin Atlantic, the Department for Education, the RNLI, Octopus Energy and Flo Health. The customer who closed the main keynote shows what field engineering is funded to support.
Meenakshi Burra, Chief Data Officer and Chief Digital & Information Officer at Unilever, described a data platform supporting 3.7 billion people – roughly half the world’s population. The numbers underneath it: 16 petabytes of data, 55,000 production objects, 40,000 regular users. The in-house Forecast Engine Utility – FEU – covers more than five million product-customer combinations across 40 markets, roughly half of Unilever’s annualized turnover, on a rolling 104-week forward horizon. Eight cross-functional workflows are being redesigned from proven-first to AI-first. Burra described the governing principle as: “AI-driven but human-led. Every AI agent has a Unilever owner, much like every colleague would have a Unilever manager.” Unilever has trained 40,000 people on AI over three years, and has done responsible-AI work “from about 2019”.
My take
Gen 1 Genie let business users ask questions of governed data. Gen 2 lets Genie do work – delegating to subagents, chasing missing inputs, deciding what needs attention, emailing you when you’re out. The mental model’s moved from chatbot to colleague, and Unilever’s line (“every AI agent has a Unilever owner, much like every colleague would have a Unilever manager”) suggests this is heading into the organization chart, not the search bar.
Genie Ontology is doing at the semantic layer what good database modelling has always done at the schema layer – teaching the system what the business MEANS by its own terms before it answers anything. The difference between a model that knows REGION as a concept, and one that only knows the specific regions it’s seen. Without that, agents answer confidently from the wrong data. Pair it with Unity Gateway – which lets finance cap and audit what the agents cost – and you have the two halves that make the shift workable. LTAP is adjacent as it collapses the pipeline between transactional and analytical data, so the data an agent needs is already where it’s already governed. CIOs weighing Databricks against Snowflake, Microsoft Fabric or an in-house build should be looking for ontology, gateway, and the LTAP-style collapse of OLTP/OLAP all together – not just the clever-looking one. They should also be asking vendors about their own internal usage data for proof points.
The UK investment, while it contributes to opening new buildings and increasing headcount is also buying field engineering time so that Databricks staff can sit with Unilever for long enough to work out what “AI-driven but human-led” looks like on a 16-petabyte data platform, and with Compare the Market until agents are put to work in a regulated-sector production environment rather than another pilot deck. More to come on Compare the Market who ran an interesting breakout session with plenty of lessons to share.