watsonx Orchestrate

Hi everyone,

while building a project with watsonx Orchestrate, I noticed relatively high latency when executing workflows.

The delay does not appear only in steps involving LLM calls (for example in Generative Prompt nodes), which would be expected. I also observe noticeable latency between nodes that should be lightweight, such as:

• calls between workflows

• Python tools with very small deterministic logic

• simple data-processing steps

Because of this, a workflow with around 10 nodes and a few LLM calls can easily take ~10 seconds to complete, even when the individual operations themselves are trivial.

My initial idea was to design the system in a modular way, splitting logic into multiple small workflows and tools. In practice this seems to introduce a significant latency overhead between steps.

So I wanted to ask the community:

• Is this expected behavior with Orchestrate workflows?

• Have others observed similar latency when chaining multiple nodes?

• Are there recommended best practices to reduce execution time (for example limiting workflow depth, merging steps, avoiding nested workflow calls, etc.)?

• Have you found any workarounds that help keep workflows modular without incurring large latency costs?

Any experience or suggestions would be very helpful.

Thanks.