Scaling Data. Accelerating AI. Enabling You.

Software is no longer just an application. It's becoming an autonomous system that plans, executes, collaborates, and improves on its own. As agentic AI advances, the way we write software, use it, and design the infrastructure beneath it is being rebuilt from the ground up. In this keynote, Ed examines why coding itself is being automated, what software looks like when agents are its primary users, and why today's cloud and infrastructure stack was never built for any of this. The next decade of computing won't be a refinement of the last. It will be a new world. The question is what we build first.

Many teams discover TiDB through its promises of scalability and HTAP. But real-world adoption takes more than a strong benchmark or a catchy acronym. In this session, Tarmo Kople, IT Infrastructure Architect at LHV Bank (and formerly Bolt’s DB Team Lead), shares his journey deploying and evaluating TiDB across two very different organizations: from high-growth startups to regulated banking environments.

Bolt grew 400% post-pandemic to become Europe's fastest-growing mobility company, serving 100 million users across 500+ cities. But their MySQL infrastructure couldn't keep up. Adding a single column to a loaded 1TB table took up to a week, and managing hundreds of schemas across thousands of microservices had become an operational nightmare. In this session, Bolt's database reliability team shares how they evaluated alternatives, why TiDB's MySQL compatibility and horizontal scalability won the decision, and how they migrated their most critical workloads to seven TiDB clusters on AWS while achieving 3:1 data compression and five-nines reliability in production.

Most agentic AI architectures bolt a vector database onto an existing stack, adding another system to manage, another sync pipeline to maintain, and another failure point to monitor. But what if your SQL database already spoke vectors natively? In this session, Mattias Jonsson explores how built-in vector search and AI-native capabilities in distributed SQL eliminate the need for a separate vector database. The session covers how vector embeddings, semantic search, and agent memory patterns work alongside transactional and analytical workloads in a single query layer, what that means for developers building agentic applications today, and where the convergence of SQL and vector is headed next.

Most teams evaluating distributed SQL start with one question: How compatible is it with what we already run? The answer is more nuanced than any compatibility matrix suggests. This session digs into what actually happens when MySQL workloads move to distributed SQL: the queries that run identically, the assumptions that quietly break, the transaction semantics that shift, and the Raft consensus and LSM tree internals that explain why. Drawing from real-world production migrations across Europe, Daniel van Eeden uncovers the patterns that transfer cleanly, the gotchas that don't surface until production, and the framework you need to evaluate what your workloads actually require.

Today's AI agents run on duct tape: separate systems for transactions, analytics, and vectors, stitched together with brittle ETL pipelines and growing engineering overhead. That approach won't survive the next wave of agentic AI, where enterprises need to support millions of concurrent agents, each with persistent memory, real-time data access, and consistent state. This session introduces the architecture that supports millions of agents: a converged distributed SQL platform that unifies OLTP, real-time analytics, and native vector search in a single ACID-compliant system, handling relational, vector, and full-text demands in one query layer. We'll walk through the Decide-Validate-Remember loop that production teams use to reduce token costs through persistent context and scale multi-agent workflows without state drift.

Agents need more than an LLM to operate in production. They need fresh data, trusted context, durable memory, and the ability to act the moment business events happen. In this session, we will show how Ververica and TiDB work together to power real-time AI applications. Ververica, powered by Apache Flink, continuously processes live events, computes signals, and triggers decisions, all at the speed of light. TiDB provides the operational, analytical, and vector-aware data layer for context, retrieval, and agent memory. Attendees will learn how to build new streaming analytics capabilities while leveraging the current enterprise data ecosystem and infrastructure.

SQL makes sense. But when it breaks, you reach for EXPLAIN. Vector search offers no such comfort. Multi-thousand-dimension embeddings, approximate nearest-neighbour indexes, and quantisation tradeoffs make it hard to know what your system is doing, and harder still to diagnose when results quietly degrade. Through interactive visualisations, Simon Hearne shows what embeddings look like in high-dimensional space, what quantisation does to your recall, and how to catch retrieval failures before your agents do. You'll leave with a sharper mental model and a diagnostic toolkit for the production problems hardest to see.