The current
https://opentelemetry.io/docs/specs/semconv/gen-ai/gen-ai-agent-spans/
documentation provides guidance on span kind selection (CLIENT vs INTERNAL) but
does not address who is responsible for creating the invoke_agent span when both
client and server are instrumented in a distributed scenario.

Scenario

Consider a LangGraph agent that:

1. Runs locally and emits invoke_agent spans with INTERNAL kind (as per the
   spec)
2. Is now exposed over HTTP as a remote service

When a remote client calls this agent service:

      sequenceDiagram
          participant Client as Remote Client
          participant Service as Agent Service (LangGraph)
      
          Client->>Service: HTTP request
          activate Service
          Note right of Service: invoke_agent (INTERNAL)
          Note right of Service: └── chat spans
          Note right of Service: └── execute_tool spans
          Service-->>Client: HTTP response
          deactivate Service

Questions

1. Should the remote client also create an invoke_agent span with CLIENT kind?
   This would result in two invoke_agent spans for the same logical operation.
2. If yes, how should these spans relate? Should they be linked via trace
   context propagation with parent-child relationship?
3. If no, who is the canonical owner of the invoke_agent span? The client-side
   SDK or the server-side agent framework?
4. Should there be a SERVER span kind for invoke_agent? PR [gen-ai] expand invoke_agent span documentation beyond remote agents #2881 discussion
   mentioned SERVER kind for "instrumenting the agent service itself" but this
   doesn't appear in the current spec.

Current Spec Gaps

The spec states:
"Span kind SHOULD be CLIENT and MAY be set to INTERNAL on spans representing
invocation of agents running in the same process."

"It's RECOMMENDED to use CLIENT kind when the agent being instrumented usually
runs in a different process than its caller or when the agent invocation happens
over instrumented protocol such as HTTP."

This guidance helps with span kind selection but doesn't address:

• Coordination between client-side and server-side instrumentation
• Whether duplicate spans are expected/acceptable
• Which party is responsible for creating the span in distributed deployments

Related Issues

• Allow INTERNAL GenAI/db spans instead of requiring the kind to be CLIENT #1315 - Allow INTERNAL GenAI/db spans instead of requiring the kind to be
  CLIENT
• [gen-ai] expand invoke_agent span documentation beyond remote agents #2881 - Expand invoke_agent span documentation beyond remote agents
• Semantic Conventions for Generative AI Agentic Systems (gen_ai.*) semantic-conventions-genai#35 - Semantic Conventions for Generative AI Agentic Systems