added multiagent semantic conventions along with single agent additio…

[PRATIBHA-Moogi]

To address - #1961 (comment).
This PR addresses the need for defining multiagent semantic conventions

Fixes # NIL

Changes

Added multiagent application key attributes in addition to that added extra attributes for single agent application.

Merge requirement checklist

• CONTRIBUTING.md guidelines followed.
• Change log entry added, according to the guidelines in When to add a changelog entry.
  • If your PR does not need a change log, start the PR title with [chore]
• schema-next.yaml updated with changes to existing conventions.

[linux-foundation-easycla]

• ✅login: PRATIBHA-Moogi / (660103e)

The committers listed above are authorized under a signed CLA.

[lmolkova]

I don't believe we're ready to define multi-agent semantic conventions. We don't even have a clear definition of a single agent ones, especially when it comes to client-side agent frameworks.

Multi-agent support is tight to general-purpose workflow #1688 and as suggested in the #1961 (comment) may need a few attributes and events or spans defined (e.g. for transition between agents).

From general semconv perspective, we don't encourage defining attributes without a corresponding span/event/metric or at least an instrumentation that would use them. I don't believe these new attributes would apply to existing GenAI spans the conventions cover.

TL;DR: The suggestion is:

• please help us finalize conventions for a single agent
• if you're interested in multi-agent conventions:
  • consider how to express it in a more general way - any workflow involving agents, humans, or non-AI bots would have similar characteristics as multi-agent one
  • it's best to start by prototyping the actual instrumentation and thinking about information that would be unique and essential for multi-agent flows.

[PRATIBHA-Moogi]

Thanks for your comments @lmolkova . I am working on it along with @gyliu513 . Regarding current gen_ai.agent.xx conventions, it doesn't represent a multiagent set of attributes. We can surely expand that one first. I also find bit of overlap on different concepts such as Workflow, flows, tasks, agents, tools, as well as runnable etc. as expressed by different agentic-application-frameworks e.g. langchain/langgraph, crew.ai, autogen, IBM Bee framework, Llamaindex etc. Is there we are maintaining a document to freeze on a common set of defnitions across these middleware framework ?

[gyliu513]

@lmolkova totally agree with you, I will work with @PRATIBHA-Moogi to improve the single agent first, and then extend to multi-agent