From 5c29d2eaa7879511129cb27b7e1faced4c8a2c8a Mon Sep 17 00:00:00 2001
From: rkoster <hi@rkoster.dev>
Date: Thu, 2 Jul 2026 12:06:46 +0200
Subject: [PATCH 1/2] Add research note: OpenTelemetry semantic conventions for
 GenAI

---
 research/opentelemetry-genai.md | 50 +++++++++++++++++++++++++++++++++
 1 file changed, 50 insertions(+)
 create mode 100644 research/opentelemetry-genai.md

diff --git a/research/opentelemetry-genai.md b/research/opentelemetry-genai.md
new file mode 100644
index 0000000..26023cf
--- /dev/null
+++ b/research/opentelemetry-genai.md
@@ -0,0 +1,50 @@
+---
+title: "OpenTelemetry Semantic Conventions for GenAI"
+author: Ruben Koster (@rkoster)
+date: 2026-07-02
+tags: [observability-governance, ecosystem-survey]
+cf_areas: [loggregator]
+status: draft
+sources:
+  - https://opentelemetry.io/docs/specs/semconv/gen-ai/
+---
+
+## Summary
+
+OpenTelemetry defines semantic conventions for generative AI operations (currently in
+Development status) covering all four OTel signals. The conventions standardize span
+attributes for LLM model calls, agent framework operations (planning, tool selection),
+structured request/response events, and GenAI-specific metrics such as token usage and
+operation duration.
+
+## Key findings
+
+- **Model spans**: Standardized attributes — `gen_ai.request.model`, temperature,
+  `max_tokens`, and `gen_ai.usage.*` token metrics.
+- **Agent spans**: Conventions for agent framework operations above individual model calls —
+  planning, tool selection, multi-step reasoning.
+- **Events**: Request/response payloads and tool call inputs/results as structured OTel
+  events.
+- **GenAI metrics**: `gen_ai.client.token.usage`, `gen_ai.client.operation.duration`,
+  `gen_ai.client.response.status`.
+- **MCP integration**: Semantic conventions also defined for Model Context Protocol
+  operations.
+- **Vendor coverage**: Conventions cover OpenAI, Anthropic, Azure AI Inference, and AWS
+  Bedrock.
+
+## CF relevance
+
+Platforms already collecting OTel traces and metrics have a clear path to supporting GenAI
+observability — the conventions plug into existing pipelines without new infrastructure. The
+coverage gap is at the agent-framework layer: current conventions focus on model-level spans,
+while multi-step agent operations (tool orchestration, planning loops) are still developing.
+A platform surfacing agent workloads would need to decide how to correlate agent-level spans
+with platform-level routing and request traces.
+
+## Open questions
+
+- How to correlate agent spans with platform-level routing traces (request IDs, trace
+  propagation)?
+- Should agent-span conventions be extended at the platform level, or left to frameworks?
+- Can platform health checks be extended to include model connectivity and token-limit
+  headroom?

From 3fe7acde3113881bb5876758d864b32487934141 Mon Sep 17 00:00:00 2001
From: rkoster <hi@rkoster.dev>
Date: Thu, 2 Jul 2026 12:11:44 +0200
Subject: [PATCH 2/2] Fix OTel note: correct metric name, signal count, and
 source URL

---
 research/opentelemetry-genai.md | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)

diff --git a/research/opentelemetry-genai.md b/research/opentelemetry-genai.md
index 26023cf..9a47cb2 100644
--- a/research/opentelemetry-genai.md
+++ b/research/opentelemetry-genai.md
@@ -6,13 +6,13 @@ tags: [observability-governance, ecosystem-survey]
 cf_areas: [loggregator]
 status: draft
 sources:
-  - https://opentelemetry.io/docs/specs/semconv/gen-ai/
+  - https://github.com/open-telemetry/semantic-conventions-genai
 ---
 
 ## Summary
 
 OpenTelemetry defines semantic conventions for generative AI operations (currently in
-Development status) covering all four OTel signals. The conventions standardize span
+Development status) covering spans, metrics, and events. The conventions standardize span
 attributes for LLM model calls, agent framework operations (planning, tool selection),
 structured request/response events, and GenAI-specific metrics such as token usage and
 operation duration.
@@ -26,7 +26,7 @@ operation duration.
 - **Events**: Request/response payloads and tool call inputs/results as structured OTel
   events.
 - **GenAI metrics**: `gen_ai.client.token.usage`, `gen_ai.client.operation.duration`,
-  `gen_ai.client.response.status`.
+  `gen_ai.client.operation.time_to_first_chunk`.
 - **MCP integration**: Semantic conventions also defined for Model Context Protocol
   operations.
 - **Vendor coverage**: Conventions cover OpenAI, Anthropic, Azure AI Inference, and AWS