Data Streaming Summit 2026: StreamNative Bets on Agent Infra

What happened

StreamNative's Data Streaming Summit returned to San Francisco on October 7–8, 2026, at the Hotel Nikko, the venue where the event started 16. The company behind Apache Pulsar organized it 5. This year the conference took on a new identity. Promotional materials called it "The Data Streaming + Agent Infra Conference" 148. The event site and one ticket listing used the shorter "Data + Agent Conference" 27. Either way, the message is clear. The streaming community now sees AI agents as its next big workload.

The call for proposals closed June 30 8. Selected speakers received a free pass and promotion 4. The program ran across three tracks: streaming engines, agent infrastructure, and agent runtimes 9. The listed lineup included engineers from Apple, OpenAI, and Uber, along with StreamNative's co-founder and CTO Matteo Merli and RisingWave founder Yingjun Wu 2. Tickets topped out at about $268, went on sale September 1, and were non-refundable 7.

The pitch: agents are a streaming problem

StreamNative's case for the rebrand is simple. The company argues that agent builders have stopped talking about prompts and demos and started talking about production 9. Production agents need several things:

• real-time context
• memory that persists across sessions
• event-driven triggers
• reliable, observable orchestration of tool calls 9

The summit framed these needs as real-time decision-making, durable memory, continuous context, and operational reliability at scale. It called them problems the streaming community has been working on for more than a decade 16. StreamNative admits the problems are partly new. It argues their underlying shape is familiar: durable state, event ordering, replayable history, multi-tenancy, and observability 9.

This framing is a smart move. It turns Kafka- and Pulsar-era expertise into a claim on the agent stack, a space that orchestration frameworks and model vendors would otherwise define.

What was announced

Yaroslav Tkachenko attended and presented. His write-up says Pulsar news took up much of the keynote, which is unsurprising given the organizer 5. The headline feature was "scalable topics" in Pulsar 5.0. These let topic capacity grow or shrink without manual repartitioning. Pulsar splits and merges segments on its own while keeping message key ordering intact. Tkachenko notes the design draws on LinkedIn's Northguard and Pravega 5. He also highlighted three other items:

• Oxia, StreamNative's replacement for ZooKeeper as Pulsar's metadata store
• Streamhouse, discussed as a source of real-time context
• Broader sessions on Flink and AI-native streaming 5

The 2026 event builds on the previous year's. At the September 2025 summit, StreamNative launched two products [12]:

• Ursa Everywhere, which brought its lakehouse-native streaming engine to all deployment models
• Orca, an event-driven engine for production AI agents

The company described this pairing as the industry's first unified platform for real-time AI and claimed it could cut streaming costs by 90% 12. Those are vendor claims, not independent benchmarks. StreamNative's product page makes a more modest pitch of "up to 10x lower cost." It describes Ursa as writing directly into Iceberg and Delta Lake tables, with no connectors or duplicated data 11. The page also points to Ursa winning Best Industry Paper at VLDB 2025, and to an open-sourced "Streaming Proof" harness that tests ordering and delivery guarantees under failure conditions 11.

Context: a crowded calendar

The summit is not the only streaming event in town. A 2026 roundup of data conferences named Confluent's Current, held November 4–5 in San Francisco, as the top pick for streaming and real-time practitioners 10. The same list includes much larger and pricier events. Databricks' Data + AI Summit, for example, listed $1,895 tickets 10. Against that field, StreamNative's roughly $268 admission makes the summit a niche, practitioner-focused event 7. It describes itself as an "open, ecosystem-wide gathering" 3. It is also, clearly, a launch platform for its host.

Our reading

The rebrand matters more than any single feature. Pulsar 5.0's scalable topics address a real operational pain point, and Oxia shows that ZooKeeper's long retirement is still under way 5. The strategic story, though, is positioning.

StreamNative is betting that two forces will push streaming platforms into the middle of AI architecture:

• agents' need for replayable, ordered, durable event histories
• lakehouse-native storage that removes duplicated pipelines

Whether enterprises buy that argument from StreamNative or from Confluent remains open. Still, the bet that agent infrastructure looks a lot like streaming infrastructure is a credible one. The rest of the industry will likely adopt the same framing soon.