Dapr 1.19 Cuts Workflow Overhead for AI Agents
The open-source runtime's new release reduces data transfer and retires a control-plane service for long-running agent workloads.
Updated on Nov. 9, 2026 in Artificial Intelligence
Diagrid Inc. released Dapr 1.19, changing its workflow engine so workers receive only new events instead of the full execution history on each step. The release also lets operators retire the standalone Placement service by enabling a Scheduler placement mode.
Why it matters
Previously, the workflow engine sent the entire execution history to the worker on every step, increasing latency and memory use on long-running jobs. The changes target the durable workflows that long-running AI agents rely on.
In Diagrid's engine benchmarks, median step latency dropped from 553 milliseconds to 153 milliseconds with the opt-in WorkflowsFastPath preview enabled. The smaller data transfers apply only to workers built on Dapr's Java and Python SDKs in 1.19.
The players
Diagrid Inc.
Diagrid Inc. is the company that released Dapr 1.19 and publishes the engine benchmarks cited in the release.
Dapr
Dapr is an open-source runtime for building distributed applications, widely used for workflows and actor-based services.
The details
Workers in 1.19 keep the execution history they already hold and receive only new events, though they still replay the full history locally. The Scheduler placement mode is enabled through a Helm setting once every sidecar runs 1.19, and it can be switched on or off without restarting the sidecar.
Timeline
Dapr 1.19 was released on November 9, 2026.
The Tech Race
Dapr 1.19 changes the prior workflow engine, which sent the full execution history to workers on every step, by sending only new events instead. The release also adds alpha Vector and Binary Store building blocks, extending the runtime's component coverage for AI-oriented workloads.
Developers building AI agents on Dapr's Java and Python SDKs can use the smaller data transfers in 1.19 without changing their code. Teams can also retire the standalone Placement service by enabling the Scheduler placement mode through a Helm setting.
The takeaway
Developers running long-running AI agents on Dapr should review the 1.19 workflow changes, since less data moving per step can reduce latency and memory use. Testing the opt-in WorkflowsFastPath preview on representative workloads will show whether the reported gains hold in practice.
Further reading
Explore more Artificial Intelligence coverage to see how agent infrastructure is evolving.
