add gen_ai.main_agent.name for nested agent and workflow inocations

[lmolkova]

Fix #300
This is an alternative to #355 and complements #270

In #270 we're adding a notion of main agent (local within the process). The root_operation name is pretty much the same except it's not resource/entity attribute, but still identifies the thing this operation is nested under.

Is there a conflict with entity? No:

• main_agent is limited to in-process propagation. It's interesting to have origin agent (or maybe end-to-end task name), but it's additive and should not break 'my local boss' identification. I.e. if we do this, it should come through some other attribute(s)
• since main_agent is local, the worst case is that main agent would show up as entity and signal attribute, but they will have the same value
• entity can only be set by infra / application code and can only apply to cases when main agent is global for the whole process.
• signal attributes can be set by all agentic / inference instrumentations

Are we blurring the line between agent and workflow Yes:

• main_agent is either agent or workflow - this is a logical thing rather than something related to agentic library API
• this line is effectively blurred for the agent registries and other places where we call something an 'agent' from the outside and internally it can be a single llm call or a giant graph of graphs.
• As a consequence, we should consider demoting invoke_workflow to operation name on an invoke_agent span and reporting its metrics as invoke_agent metrics. To follow up on Clarify workflow span criteria and add reference scenarios for langchain and openai agents #354

Motivation

• ability to differentiate main agents/workflows from nested ones
• as a follow up, we can reference this attribute from inference, tool calls and use it as a filtering mechanism for them (give me inference token usage by main agent)

Checklist

• Motivation section filled in above
• Reference scenarios updated for affected libraries
• Towncrier fragment added under changelog.d/ for any change to the conventions that a consumer would care about. Editorial changes (typos, pure rewording, repo tooling) don't need an entry.

See [CONTRIBUTING.md]

[opentelemetry-pr-dashboard]

Pull request dashboard status

Waiting on the author · refreshed 2026-08-20 16:07 UTC

Move out of draft to request review.

Status above doesn't look right?

• Just replied or pushed? Anything around or after the refresh time above may not be picked up yet — give it a few minutes.
• Anything look wrong? Report it with what you expected; it helps us improve the dashboard.