[CNCF LFX Proposal] OpenTelemetry — Zero-code AI Agent observability for otelc

### CNCF Project

OpenTelemetry

### Term

2026 Term 3 (Sep-Nov)

### Program Name

Zero-code AI Agent observability for otelc

### Program Description

Design and build an **AI provider adapter interface** in `otelc` (`opentelemetry-go-compile-instrumentation`) — OpenTelemetry's compile-time, zero-code instrumentation for Go — that emits the `gen_ai.*` semantic conventions (spans, metrics, events) through a single shared abstraction, so AI providers and libraries plug into a common contract instead of being instrumented ad hoc.

The already-merged OpenAI instrumentation (#604) is migrated onto this interface as the reference implementation, then the adapter is extended to additional providers and libraries: `anthropic-sdk-go`, Gemini, and the MCP / LangChain frameworks. MCP/LangChain require non-HTTP hooks (following the Kafka/Redis instrumentation patterns) rather than the HTTP-based path.

This design-first framing gives the mentee real architecture and problem-solving work — defining the adapter contract, mapping heterogeneous provider APIs onto shared semconv, and validating it across libraries — rather than repetitive per-library instrumentation.

Community demand is concrete and vendor-neutral: the OpenAI instrumentation (#604) closed community issue #327, and the request pipeline is active (#706 for aws-sdk-go-v2, #553 MongoDB shipped).

Deliverables follow a week-by-week plan where each milestone ships a runnable demo + green CI so progress is visible at a glance: adapter interface + OpenAI migration → anthropic-sdk-go → Gemini → MCP → LangChain, each with instrumentation rules, tests under the paired `test/` directory, and an example app emitting `gen_ai.*` spans.

Privacy posture: message bodies are off by default and strictly opt-in via `OTEL_INSTRUMENTATION_GENAI_CAPTURE_MESSAGE_CONTENT` — nothing is captured unless the operator turns it on.

### Technologies

Go, OpenTelemetry, compile-time instrumentation, gen_ai semantic conventions, API/interface design

### Skills same as Technologies?

- [x] Yes, the required skills are the same as the technologies listed above.

### Required/Desirable Skills

_No response_

### Mentors

Haibin Zhang | @NameHaibinZhang | namehaibinzhang@gmail.com | zhanghaibin1990
Xabier Martinez | @txabman42 | x42.martinez@gmail.com | x42.martinez

### Upstream Issue URL

https://github.com/open-telemetry/opentelemetry-go-compile-instrumentation/issues/709

### Application Prerequisites

- [x] Resume
- [x] Cover Letter
- [ ] School Enrollment Verification
- [ ] Participation Permission from school or employer
- [ ] Coding Challenge
- [ ] Custom Prerequisite (fill in details below)

### Coding Challenge URL

_No response_

### Custom Prerequisite Name

_No response_

### Custom Prerequisite Description

_No response_

### Custom Prerequisite — File Upload

- [ ] Yes — completion of this task requires the mentee to submit a file.

---
**LFX program:** [CNCF - OpenTelemetry: Zero-code AI Agent observability for otelc (2026 Term 3)](https://mentorship.lfx.linuxfoundation.org/project/91cab14a-42aa-4e08-9144-3b243309e68c)

Start by reading upstream issue #709 and the already-merged OpenAI instrumentation (#604) in opentelemetry-go-compile-instrumentation. Define the shared adapter contract, then validate each provider or framework in its paired test/ directory and example app. Done means the planned providers emit gen_ai.* telemetry with green CI and message capture remains opt-in.

Written by the indexing model from the issue text.