What OpenTelemetry Looks Like in 2027
I caught up with Ted Young — a co-founder of OpenTelemetry and long-time CNCF maintainer, now at Grafana Labs — at the How to roll out OpenTelemetry at scale event in Amsterdam on Tuesday, September 8. The room was small but the stakes are high: every engineering organization is moving toward OpenTelemetry, and the question isn’t whether but how fast you can ship it to production.
Ted walked me through the four focus areas he sees driving OpenTelemetry forward over the next 12 months. Spoiler: the project is shifting from “make it work” to “make it default.”
1. Clients: From Services to the Full Stack
For the first five years, OpenTelemetry was almost entirely server-side. The Collector, the instrumentation libraries, the SDKs — all focused on the backend.
“We’re now moving to instrument iOS, Android, and Browser. That’s in beta right now across all three environments, and we want to move that to stable production within the year.”
This is the biggest inflection point. Mobile and web instrumentation has historically been the domain of proprietary agents — and OpenTelemetry is closing that gap. Package management is the other half: instead of wrestling with manual installs, you can now run a single command:
apt install opentelemetry…and get complete instrumentation — automatic, across all languages — including the Collector, auto-instrumentation, and Helm charts for Kubernetes. The experience is now equivalent to spinning up a Datadog or New Relic agent.
2. Stability and Security: Doubling Down
The second push is what Ted called “doubling down” on what’s already there: hardening the instrumentation, stabilizing semantic conventions, and tightening security.
“There’s currently a gap between how easy it is to install OpenTelemetry relative to, say, proprietary agents. We’ve now closed that gap.”
It’s the boring work — standardizing APIs, freezing breaking changes, documenting upgrade paths — but it’s the work that moves OTel from “conference demo” to “production default.”
3. AI: Observing the AI Stack
The GenAI SIG is one of the most active groups in the CNCF right now. As Ted explained, the challenge isn’t just observing traditional services anymore — it’s observing the AI agents, the LLM calls, the RAG pipelines that are being stitched into every application.
“People are adding agentic systems to their stack, and now they want to observe all of these GenAI and agentic parts of their stack.”
OpenTelemetry is extending its service-mesh-style instrumentation to agentic frameworks — the same way it provides auto-instrumentation for Django, Express, and Spring Boot. There’s also a fascinating project brewing around Weaver, OTel’s tool for semantic-convention change management, which the GenAI SIG is exploring for managing instrumentation at scale.
4. Performance and Edge: The OpenTelemetry Arrow Protocol
This is where the rubber meets the road for enterprises running at massive scale. OTAP (OpenTelemetry Arrow Protocol) is a Rust-based implementation designed for the Collector to handle really huge data pipelines — the kind you see at Cloudflare and F5 scale, where you’re egressing millions of telemetry points per second from edge locations.
“OTAP is for handling that. Think about edge networking like Cloudflare and F5 places where you’re trying to egress lots of data.”
The C++ SDK push complements this — providing a performance boost and language bindings that make it possible to bind to native code across every supported language.
The Takeaway
Five years ago, OpenTelemetry was an ambitious spec trying to unify a fragmented ecosystem. Today — listening to Ted in Amsterdam — it’s a platform that’s about to become the default way you observe software, from the browser to the edge.
The biggest shift he highlighted: the first-time experience. A year ago, spinning up OpenTelemetry meant reading docs, configuring a Collector, hunting down the right auto-instrumentation library for your language, then wondering whether you had the right sampling strategy. Now you install it like any other package, and the defaults just work.
Recorded at the “How to roll out OpenTelemetry at scale” event, DoubleTree by Hilton Amsterdam Centraal Station, Tuesday 8 September 2026. Ted Young is a co-founder of OpenTelemetry, a long-time CNCF maintainer, and currently at Grafana Labs.
Discussion questions:
- Are you running OpenTelemetry in production today?
- Which of the four focus areas (clients, stability, AI, edge) is your biggest concern or opportunity?
Let me know in the comments — I’ll carry the best answers into next week’s Production AI Brief.
#OpenTelemetry #GrafanaLabs #CNCF #Observability #Amsterdam