London hosts a new edition of Current 2026, the reference event organized by Confluent around data streaming and artificial intelligence applied to data in real time. Under the motto “Building Intelligent Systems on Real Time Data”, the event has brought together developers, engineers, data experts and technology leaders from across Europe to explore how organizations can move from passive analysis systems to architectures capable of acting autonomously on instantly updated information.
The starting signal was given by Jay Kreps, CEO and co-founder of Confluent, with a presentation that set the tone for the entire day. Kreps began by thanking the momentum of the open source community that supports the company’s projects: more than 400 assets and more than 80 new Kafka Improvement Proposals (KIPs) for Apache Flink developed over the last year, in addition to advances such as Kafka Queues and the native Flink Agents.
The CEO also officially confirmed that Confluent becomes part of IBM, stating that “IBM has a long tradition as a guardian of open source, demonstrated with Red Hat and Linux. This integration allows us to accelerate our investment in the community and in the capabilities of the product,” said Kreps, who also highlighted that joining IBM will be a new impetus to maintain innovation.
The bulk of his intervention focused on the paradigm shift that the industry is experiencing: the transition from traditional Business Intelligence systems, where people interpret data to make decisions, towards autonomous systems that close the cycle on their own and act accordingly. “The problem is not the AI model, but having the real context of the company’s messy data,” Kreps said. In this sense, he pointed out that in the development of modern AI, data and code are inseparable, and that the quality of the context that an agent receives directly determines the quality of its decisions.
Kreps also warned about two frequent but problematic patterns in the company: directly exposing agents to production systems, which entails security risks and makes evaluation difficult; and curate the context in data lakehouses and then load it into operational databases, a safe approach but one that generates data that is too stale for real-time systems. The solution proposed by Confluent is a streaming-first architecture that unifies batch and streaming processing into a single flow from development to production.
AI ready to go into production
Shaun Clowes, Chief Product Officer of Confluent, took over to present the specific innovations that materialize that vision. His intervention revolved around the general availability of various capabilities within Confluent Intelligence and Confluent Cloud, under a clear premise: eliminate the technical and regulatory obstacles that prevent organizations from bringing their AI projects to the production environment.
“In many companies in EMEA, the complexity of data privacy regulations ends up stopping AI projects before they reach production. “Confluent removes these barriers by integrating enterprise-level governance directly into data flows,” explained Richard Timperlake, Senior Vice President of Sales EMEA, who accompanied the presentation with a relevant market perspective: according to McKinsey, eight out of ten companies consider data limitations to be the main obstacle to scaling agentic AI.
Among the most notable announcements is the general availability of the Real-Time Context Engine, a fast SQL layer with caching over streaming-backed tables, exposed through the MCP protocol so that agents and applications can query it with very low latency. This engine avoids costly transfers between systems and delivers iterations on context data sets in a secure and evaluable way, critical to improving the performance of AI agents.
Confluent is committed to a streaming-first architecture that unifies batch processing and streaming in a single flow, from development to production
Also announced was the general availability of Confluent’s managed MCP server along with Agent Skills, which enable development tools such as Cursor or Claude Code to create, manage and debug streaming operations using natural language. Added to this is the free and open source dbt adapter for Flink, which integrates Flink SQL into the standard workflow of data engineers, significantly reducing the barrier to entry to adopting real-time processing.
On the security and privacy front, Confluent introduced in early access a new automatic anonymization of personal data (PII) function integrated directly into Flink SQL, without the need for custom code or external services. It’s complete with support for Azure Private Link, which lets you keep AI workloads off the public network using private connections to services like Azure OpenAI or Cosmos DB.
The challenge of voice agents in the company
During the day, the conversation between Sean Falconer, Head of AI at Confluent, and Alex Holt, Head of Forward Deployed Engineering at II ElevenLabs, stood out. Both agreed that the two big challenges for deploying voice agents in enterprise environments are latency and context accuracy.
It’s not enough for the model to be capable: it needs access to up-to-date and relevant corporate information in real time. “Agents must be treated as code, with configuration, versioning, testing and monitoring,” Holt stressed, appealing to the engineering rigor necessary to reliably bring these systems to production.
A platform for the present and future of streaming
The new edition of Current 2026 is taking place on May 19 and 20 in the London capital, with a clear message: data streaming has ceased to be background infrastructure and has become the operational core of enterprise AI. With capabilities such as Kora—the cloud-native Kafka engine with elastic auto-scaling and 99.99% SLA—, Tableflow to replicate streams in open formats such as Iceberg or Delta Lake, and Confluent Private Cloud for hybrid and on-premise environments, the company presents a platform designed to accompany organizations at every stage of their data and AI maturity. Those organizations that embrace streaming as a strategic priority will be better positioned to compete in the era of autonomous AI.
