# OpenTelemetry Celebrates Graduation and the Next Era of Agentic > [!question] Lingering Questions > - Is a tracing schema built for deterministic systems adequate for non-deterministic agent execution? The talk asserts yes and tests nothing. > - "Second-largest CNCF project" — by what measure? Contributors, commits, and adoption all imply different rankings, and none is specified. > - Cardinality: instrumenting every agent tool call at machine speed raises a volume-and-cost question the talk never mentions. > [!tip] Article Ideas > - The three-factor account of why a standard succeeds (rip-and-replace aversion, unified signals over silos, welcoming contributor community) is a reusable frame — worth testing against standards that failed, not just this one that succeeded. > [!link] Cross Reference > - LLM token usage as a first-class telemetry signal is what would make model-routing cost arguments measurable rather than anecdotal — cross-link to any vault notes on AI cost/tokenomics. > - [[3. Resources/MOCs/AI Cost & Tokenomics MOC]] > [!quote] Vocabulary > - **GenAI semantic conventions** — OpenTelemetry's standardized schema for describing LLM/agent telemetry (tokens, prompts, tool calls) across vendors, positioned in this talk as the target for any new AI instrumentation work. ## Bibliographic capsule A KubeCon Japan 2026 keynote from two OpenTelemetry Governance Committee members announcing the project's graduation from the CNCF — its highest maturity tier — paired with a forward argument that vendor-neutral instrumentation is the right substrate for observing agentic AI systems. ## Author's thesis OpenTelemetry succeeded because it solved a problem organizations couldn't solve themselves (unify observability signals without re-instrumenting on every vendor switch), and that same rip-and-replace aversion argument now extends to agentic AI, where GenAI semantic conventions are pitched as the neutral schema before proprietary tooling fragments the space. ## Key arguments (≤3) 1. **OpenTelemetry graduates as CNCF's second-largest project after Kubernetes** — a real milestone, reported by the project's own governance committee in a celebration slot, which invites retrospective survivorship bias in the success-factor account. 2. **Standards win where re-instrumentation is the expensive part** — the transferable mechanism behind the graduation, not specific to observability. 3. **Diagnosis should be organized by question, not by data type** — logs, metrics, and traces as one correlated signal rather than three silos, because "we have system problems," not separately a logging problem and a metrics problem. ## Claim atoms ### OpenTelemetry graduates as CNCF's second-largest project > [!quote] ID:rw_01kysr4w2xw9hdsm5w300wvtyg > today with a big announcement. Open Telemetry has graduated from the CNCF. Opening of the keynote. The milestone itself, plus the size claim made shortly after — note the size metric is unspecified. ## Permanent candidates - **Standards win when re-instrumentation is the expensive part** — a neutral standard succeeds in domains where switching vendors would otherwise require rewriting integration code. → promoted. - **Diagnose by question, not by data type** — tooling organized around the kind of data it holds forces the investigator to reassemble the system from silos at exactly the moment they can least afford to. → promoted. ## Processing checklist - [ ] Test the three-factor standard-adoption account against at least one standard that failed - [ ] Cross-link LLM token usage as a telemetry signal to any vault notes on AI cost measurement - [x] Connect into the relevant MOC ## Links - [[3. Resources/MOCs/Open Source Governance MOC]] — CNCF graduation and the standard-adoption mechanism are open source governance case studies - [[3. Resources/MOCs/AI Agent Safety MOC]] — the agentic-observability argument bears directly on how agent systems get monitored and audited - [[3. Resources/Readwise/Full Document Contents/Articles/OpenTelemetry Celebrates Graduation and the Next Era of Agentic - Alolita Sharma & Ted Young]]