Databricks Data and AI World Tour London 2025 Keynote

Summary

The UK and Ireland is now the fastest-growing Databricks hub globally, with Databricks present in over 50% of FTSE 100 companies, more than 500 UK employees, and a commitment to train over 100,000 people in AI in the UK and Ireland.
The keynote demonstrates Agent Bricks for building and evaluating multi-agent systems with MLflow, Lakebase as an open-source Postgres database for transactional applications, and Databricks Apps for secure enterprise application deployment, all governed through Unity Catalog.
Customer sessions show easyJet rebuilding its airline revenue management system using Lakebase and Databricks Apps, and Barclays scaling governed data and AI using Lakehouse, Unity Catalog, and Agent Bricks.

Databricks Data and AI World Tour London 2025 Keynote

Databricks Data and AI World Tour London 2025 explores how enterprises can build production AI agents, intelligent applications and governed analytics on open data. The keynote covers a unified architecture that uses Unity Catalog to govern data and AI assets across systems while making them accessible to developers, analysts and business users.
See Agent Bricks create and evaluate multi-agent systems with MLflow, Lakebase support transactional applications with open source Postgres, and Databricks Apps connect AI and data securely. You will also learn how Lakeflow, Databricks AI/BI, Genie, Genie One, Lakehouse, and Lakebridge support data engineering, analytics, and migration. Customer discussions show how easyJet is modernizing airline revenue management and how Barclays is scaling governed data and AI.
Data intelligence customer use cases: https://www.databricks.com/blog/data-intelligence-action-100-data-and-ai-use-cases-databricks-customers

Chapters

FAQs

What is Agent Bricks and what was demonstrated at the London World Tour keynote?

Agent Bricks is a Databricks capability for creating and evaluating multi-agent systems, demonstrated live at the London keynote as a multi-agent supervisor workflow. It uses MLflow for agent evaluation and is part of the Databricks Data and AI platform's foundation for building production AI agents.

What is Lakebase and how is easyJet using it?

Lakebase is a Databricks-managed, open-source Postgres database designed to support transactional AI applications alongside the lakehouse. easyJet is using Lakebase together with Databricks Apps to rebuild its airline revenue management system.

How is Barclays using the Databricks Data and AI platform?

Barclays is modernizing its data infrastructure using Lakehouse and Unity Catalog for governance across its data and AI assets. The bank is also evaluating Lakebase and Agent Bricks as part of its strategy to scale governed data and AI capabilities.

How large is the Databricks presence in the UK and Ireland?

Databricks has more than 500 employees in the UK, is present in over 50% of FTSE 100 companies, and has over 300,000 monthly active users in EMEA growing double digits quarter after quarter. Databricks is also investing more than 10 million to train over 100,000 people in AI in the UK and Ireland and is participating in the UK's DSIT program.

Full transcript

[02:10] Michael Green, managing director and country leader UK and Ireland, welcomes everyone to the London Data and AI World Tour. He notes UK and Ireland is now officially the largest Data and AI World Tour globally, only eclipsed by the summit itself, reflecting the size of the community.
[04:55] Why is the UK and Ireland the fastest growing hub? The region thrives on energy and creativity to build new things, with a rich culture of connectivity and networking. The world tour is about connecting developers, data scientists, analysts, business leaders and innovators.
[06:32] About 40% of the room is here for their first world tour, up from 10 people last year. Last year we announced our European headquarters in London on Windmill Street, and we're now well over 500 employees in the UK.
[07:19] We're investing over 10 million into UK and Europe for the future of AI talent, aiming to train over 100,000 people in AI in the UK and Ireland, participating in the UK's DSIT program, and partnering with universities like LSE, Southampton, Sheffield and University College Dublin.
[08:26] London is the perfect place to kick off the AMEA Data and AI World Tour because it focuses on innovation. It's not a question of what data and AI can do, but what can we build with it. We're now in over 50% of the FTSE 100 companies, working with Reckitt Benckiser, Shell, Linguish, Virgin Atlantic and Flow Health.
[10:19] You'll hear from customers directly: Barclays on modernizing the bank with Lakehouse and Agent Bricks, and easyJet on rebuilding their revenue management system using Lakebase and Databricks apps. We take that trust and partnership seriously. Today we're not just keeping up with change, we are shaping the change. Over 50 customers are speaking today.
[13:40] Michael introduces Samuel Bonamigo, SVP and GM for AMEA, to talk about momentum across Europe.
[14:32] Samuel: we keep growing and scaling, ending the year with close to 2,000 employees in Europe, including engineering resources in Amsterdam, Belgrade, Berlin and the Nordics. We see great momentum with more than 20,000 customers worldwide and more than 300,000 monthly active users in EMEA, growing double digits quarter after quarter.
[16:10] It's also about the ecosystem: GSIs, ISVs, technology and built-on partners, and data providers. Recent partnerships include the London Stock Exchange to increase data quality within the platform, and a major partnership with SAP creating Business Data Cloud, giving access to Databricks within SAP across Microsoft, AWS and GCP.
[17:46] We also announced a strategic multi-year partnership with OpenAI: Databricks customers can take the best of both worlds with no extra data movement, using OpenAI on the Databricks platform to create new agents and applications. What makes us unique: partnerships, platform strategy, product roadmap, strong innovation DNA, and an open-source strategy.
[19:22] Governance, security and Unity Catalog are a gamechanger. The Gartner Magic Quadrant shows Databricks alone in the top right corner. If you're a customer you're working with a world-class platform. More than 20,000 customers across nine-plus industries, and after the keynote you can go to industry tracks with more than 550 customer testimonials.
[21:04] Michael returns: the momentum is about innovation, product, and change. Scan the QR code for the agenda, 50 customer speakers and packed sessions. A big thank you to sponsors, and two official afterparties, a boat party (already sold out) and a Wharf party. Now let's get into the keynote.
[23:44] Michael welcomes Arsalan Tavakoli, SVP, co-founder and head of field engineering.
[24:27] Arsalan: today is about democratizing data intelligence. Databricks has been around a little over 12 years. Many organizations sit in a world where a proliferation of tools has spread data across many places, typically in proprietary formats and locked in, with high costs.
[25:33] The path forward: start with open formats to unify the data; the era of giving your data to a vendor in a proprietary format should be over. Second, bring unified governance to data and AI, not fragmented, that cuts across all your assets and supports all systems.
[26:51] When we say unified governance with Unity Catalog, it means across all your assets: tables, files, unstructured data, models, notebooks, dashboards, supporting open formats, with access control plus discoverability, sharing and business semantics. The goal of governance is to make all your assets consumable, not just to lock data down.
[28:14] Unity Catalog supports data sitting in many different places, including on-prem, giving a single pane of glass, and lets you bring whatever engine you want to read and write across those data sets. The Databricks Data and AI platform starts with open data formats, then Unity Catalog for governance.
[29:35] Data intelligence brings data and AI together, where the whole is greater than the sum of the parts: use AI to make your data accessible to everybody (Genie for natural language questions, an assistant like a co-pilot), and use your data to build new AI capabilities to power your products.
[31:12] Over 90% of customers use classic ML and a large majority use generative AI. Today's key themes: how do we simplify production agents, how do we prepare the foundation for intelligent applications, and how do we implement data intelligence.
[32:32] First, simplifying production agents. A customer video from Flow Health, the number one women's health and fitness app with more than 80 million monthly active users, describes using Agent Bricks with LLM judges and synthetic data generation to fine-tune open source models like Llama, doubling their medical accuracy and safety at low cost.
[34:30] 2025 is the year everybody talked about agents, but getting high quality agents into production is really hard. Three reasons: understanding how agents are performing (evals, defining what good looks like), improving accuracy (many knobs, fast-moving techniques), and balancing cost versus quality.
[36:53] Most organizations don't have hordes of ML engineers; it's business users and domain experts building applications. Agent Bricks helps: instead of a blank slate, start by telling it the common task (information extraction, knowledge assistant, etc.). Describe the high-level outcome you want.
[38:15] Agent Bricks does the underlying work: it creates LLM judges as a benchmark based on what you said, automatically optimizes everything, and gives you a curve to trade off cost and quality. Organizations get agents out far faster, at higher volume, with accuracy that drives meaningful business outcomes.
[39:39] Amber demos Agent Bricks' multi-agent supervisor for the Novel Ideas bookstore, an independent store that competes via great operations and customer service. An employee asks whether a book is in stock and, if not, wants a recommendation so the customer doesn't leave empty-handed.
[41:01] Under the hood you can add up to 20 agents. The bookstock bot is a knowledge-assistant RAG chatbot over unstructured data in Unity Catalog volumes and vector search (staff recommendations, trending titles, purchase history). A Genie space handles the tabular, structured inventory data.
[42:37] The supervisor pings the inventory bot, finds the book out of stock with a restock date, and returns staff picks. To improve quality, Amber opens a labeling session for subject matter experts with expectation guidelines: only three recommendations, don't recommend the sequel before the original, at least one staff recommendation, and add the book's location.
[44:42] After merging that feedback and re-asking, the agent returns three recommendations, no sequels before originals, staff recommendations, and the science fiction and fantasy section location. You can go into the MLflow experiment, see the traces, and add evaluation metrics like LLM-as-a-judge to get agents ready for production.
[47:09] Theme two: preparing for intelligent applications. Applications need a database, but databases were built for a different era, with lock-in, no open format, high expense, and built for on-prem. The Neon acquisition data shows the rise of agents and vibe coding is leading to a massive rise in databases created by agents.
[48:45] Inside Databricks this is Lakebase. It's built on open source Postgres, so community extensions work with it; it separates compute and storage for low latency, high throughput and production SLAs at better cost; and it's built for AI, so you get a database in under a second and only pay for what you use.
[50:23] Lakebase use cases: power AI applications from analytics data without reverse ETL; build new apps like order processing directly on Lakebase; and bring transactional data into the lakehouse for analytics, with Lakebase automatically syncing one governed copy of the data.
[51:28] For applications, Databricks Apps. Vibe coding is great for quick prototypes, but in an enterprise you must handle security, state, dev/staging/prod and upgrades, which require large teams. Databricks Apps moves the application to data and AI, combining Agent Bricks, DBSQL and Lakebase under Unity Catalog.
[52:48] Databricks Apps is seamlessly integrated, secure and governed within the Databricks perimeter, and an open framework: you can use popular vibe coding tools like Lovable, Superblocks and Replit. It has organically taken off, with well into the tens of thousands of applications built.
[54:08] Arsalan welcomes Dennis Mishan from easyJet. Dennis was tasked to modernize the revenue management system, which was a time capsule: a .NET desktop app, SQL Server, and brittle integrations. Airlines want to become modern retailers with offers, orders, personalization and dynamic bundling.
[55:44] After a summer proof of concept with Databricks, easyJet made a bold move to refactor a system spanning 386 Git repositories in three months. With Databricks Apps handling infrastructure, and Lakehouse and Lakebase providing performance, flexibility and governance, teams focus on business logic. Six weeks in, they've made amazing progress.
[57:24] The application is becoming the heartbeat of revenue management: a modern NodeJS React front end, FastAPI back end on two Databricks Apps, everything flowing through Unity Catalog. It's the central hub for commercial decision making, ticket pricing, ancillary pricing, network and schedule changes, with Lakebase for transactions and Lakehouse for analytics.
[59:17] Dennis on the future: airline revenue management is built on 30-plus years of science; agents won't replace that in three months, they'll make it stronger. With Agent Bricks and Lakebase the system evolves from a system of record into a modern system of data intelligence where humans and AI collaborate seamlessly.
[01:01:16] Theme three: implementing data intelligence. It starts with great, high-quality data. Lakeflow is the end-to-end data engineering concept: Connect (single-click connectors to sources like Salesforce, Workday, ServiceNow, transactional databases, Google Drive and SharePoint), declarative pipelines, and orchestration of workflows.
[01:02:39] Many business users won't use SQL, so they use their own tool, do data prep in Excel, build a graph, screenshot it and email it around, outside governance and hard to productionize. Existing vendors add structure but the workflows remain completely separate and disjoint, don't use AI, and are hard to put in production.
[01:04:17] Lakeflow Designer gives end users a drag-and-drop interface to build their own pipelines, and because it understands your data context you can describe in natural language what you want, or paste in a slide and it builds the pipeline. Underneath it writes full pipeline code, so it's production-ready and data engineers can edit it.
[01:05:23] To get BI into everyone's hands there's AI/BI. It's seen over 500% growth. It gives dashboards for the modern era at no additional cost, free with Databricks, incredibly fast, running through the unified governance system so you only see data you have access to.
[01:06:47] Enterprises have thousands of dashboards but usage follows a Zipf distribution, a long tail built for one-off questions. You can't anticipate every question, and dashboards are inefficient for interrogating data. Instead, Genie lets you ask questions of your data directly without routing through a data team.
[01:07:53] Genie lets you query data with Q&A, understands your enterprise content and what data you have access to, and maintains context between sessions. It's become incredibly popular for exposing data to end users without needing heavy teams. Marjorie demos AI/BI.
[01:08:56] Marjorie tours AI/BI and Databricks One, a new experience for business users with a clean entry point, Genie spaces and apps. She asks which marketing tactics drove the most qualified leads last quarter; Genie responds in seconds with a chart, understanding business language like qualified lead, governed by Unity Catalog.
[01:10:33] Asking how to optimize the funnel, the Genie Research Agent builds a step-by-step plan, chooses data, runs the analysis, highlights where conversions are dropping and gives next steps with citations, so it can be trusted. For regularly tracked insights, AI/BI dashboards have visualization, cross filtering and scheduled email snapshots built in.
[01:11:53] Marjorie asks Genie to explain a spike in the marketing impressions chart, analyzing key drivers and enabling a deep dive into audience engagement. Databricks One, Genie, the research agent and AI/BI dashboards let everyone go from a question to the why to a shared trusted view, turning curiosity into confident action.
[01:13:07] Great BI requires a powerhouse data warehouse underneath, which is DBSQL, with massive growth and continually improving price performance. Databricks has the best TCO in the market across concurrency and data sizes, driving incredible adoption and migration from other systems.
[01:14:26] Migrations are complex: lots of code, underlying systems, and the people who built them have left. They're high complexity, labor intensive, and automated LLM solutions lack accuracy without context. Lake Bridge is a fully open system: it scans the environment for an assessment, then uses an advanced LLM-based code converter.
[01:16:05] Lake Bridge uses Lakeflow Connect to migrate the data and a reconciler to validate that the new system's output matches the old one. Over 20 legacy warehouses are supported. Benefits: higher conversion accuracy, faster migrations at lower cost, a modernized end state, and it's completely free, proven on hundreds of migrations.
[01:17:12] Many of you already use Databricks, and there's a Free Edition, free forever, with virtually all of Databricks to learn and experiment, plus self-served content, already used across universities and institutions. Arsalan welcomes Maggie Chung, head of enterprise data platform at Barclays.
[01:18:35] Maggie: Barclays' data strategy is driven top-down from the board to build a future-ready data platform with industry-leading capabilities, from data management and governance to advanced analytics, real-time and BI insights, and the application of AI. They saw Databricks as a market leader in end-to-end data intelligence.
[01:19:58] Barclays has delivered tangible benefits from exiting legacy technology to improving customer service and real-time fraud insights. They migrated an investment banking legacy warehouse into Databricks in just four and a half months, a record, and built a Service 360 data product for the US consumer bank that improved service while reducing cost to serve.
[01:21:53] Governance is critical for a large regulated bank. Cataloging and metadata is at the heart of Barclays' strategy; to do agility safely, metadata is king. They built an overarching enterprise metadata hub bringing together business, technical, operational and data product metadata, with Unity Catalog at the heart of moving from metadata that knows what they've done to metadata that tells them what to do.
[01:23:30] Maggie's vision: after the June conference they ran immersion sessions across the bank. Lakebase is gamechanging in bringing structured operational data together with analytics data, a problem she doesn't think anyone else has solved, and business units are fighting to be the front runner. With Agent Bricks the focus is using the capability to scale AI efficiently and safely, integrating LLM ops onto any model.
[01:26:46] Arsalan sums up: simplify production agents with Agent Bricks, built on open data and unified governance; prepare for intelligent applications at the database (Lakebase) and application (Databricks Apps) level; and implement data intelligence with high-quality data through AI/BI, Genie, dashboards and Databricks One, underpinned by a powerful data warehouse.
[01:27:46] Michael Green closes, thanking Maggie, Dennis, and the demos by Marjorie and Amber. He points to lunch, the expo, the networking session, and the afterparties, and reminds everyone that next year's Data and AI Summit is 15th to 18th in San Francisco, so book hotels and flights early.

Learn more about the Databricks Data and AI platform.

The information provided herein is for general informational purposes only and may not reflect the most current product capabilities or configurations.