Confluent launches Streamhouse to give AI agents real-time context
Confluent, an IBM company, has introduced what it calls a Streamhouse: a data architecture designed to give AI agents, applications and analytics a continuously updated, governed view of business state. The launch, announced on 5 November 2026, coincides with the general availability of Lightning Tables on AWS, completing a five-pillar Data Streaming Platform that now spans Stream, Connect, Process, Govern and Query.
The core problem the architecture addresses is well understood by enterprise data teams. Warehouses and lakehouses were built for batch reporting. As autonomous agents proliferate, they increasingly act on information that is stale or inconsistent across systems, producing decisions that look locally valid but fail in practice. Confluent's airline analogy in the release illustrates the risk clearly: a rebooking agent that checks seat availability without simultaneously checking aircraft location can issue itineraries that physically cannot work.
What Lightning Tables adds
Lightning Tables is a read-only serving layer that materialises Apache Kafka data into tables and keeps them current as changes arrive. In Confluent's internal testing, data moves from creation to serving in approximately 100 milliseconds, with query latency of between 10 and 100 milliseconds and throughput exceeding 1,000 queries per second. The company cautions that actual latency varies by configuration and workload. The layer supports SQL and a direct API, and applies schemas, lineage and access policies from data capture through to serving, so governance is not bolted on after the fact.
Alongside Lightning Tables, Confluent is making two further Query capabilities available. Tableflow converts streaming data into open table formats for lakehouse and warehouse analytics. The Real-Time Context Engine serves business context to AI agents via the Model Context Protocol, a standard that has gained traction as a way to supply large language model agents with structured, low-latency data at inference time.
Accenture is cited as a production user of a Streamhouse architecture powered by Confluent. Thomas Mahaffey, SAI Chief Data Strategist at Accenture, said the shift from legacy batch and point-to-point architectures had "dropped our time-to-insight from months to days and slashed reporting times, empowering our teams to focus on delivering high-value AI for our 800,000 employees." That is a substantive outcome claim, though Confluent did not publish independently audited figures to accompany it.
Market context and competitive positioning
The real-time data streaming market is competitive. Confluent's platform is built on Apache Kafka, in which it retains a strong commercial position, but it faces pressure from managed Kafka alternatives (AWS MSK, Azure Event Hubs), from newer streaming engines such as Redpanda and RisingWave, and from lakehouse vendors that are adding their own streaming ingestion layers. The release notes the formation of a Streamhouse Working Group, which reportedly includes Aiven, Redpanda, StreamNative and Ververica: an attempt to establish the Streamhouse concept as a cross-vendor architectural category rather than a proprietary Confluent product, a move that could accelerate adoption but also dilutes Confluent's ability to own the narrative exclusively.
The IBM acquisition brings Confluent closer to enterprise integration and governance tooling but also complicates the go-to-market story for customers already navigating IBM's hybrid-cloud portfolio. EMEA organisations in particular will be watching data-residency provisions as Lightning Tables expands beyond its initial AWS availability; the EU AI Act's requirements for high-risk AI systems are increasingly shaping how enterprises architect the data pipelines that feed their agents.
Shaun Clowes, General Manager for Confluent and Data at IBM, described the ambition plainly: "The next era of enterprise AI will be won by the agents that know what's happening in the business the moment they act." Whether Lightning Tables' performance benchmarks, which are based on internal testing rather than third-party audits, hold at the scale of large enterprise workloads remains to be verified in production deployments.