Despite the disruption AI represents to many fields of tech, it's a good time to be among the core maintainers of OpenTelemetry, an open source project that might be poised to bring order to the upheaval.
Austin Parker, one of the co-founders of the 2019 Cloud Native Computing Foundation observability project and a member of its governing board, looked back on the project's -- and the industry's -- evolution from traditional logging to a more comprehensive approach that integrates metrics, traces, events and contextual data in this IT Ops Query interview with Informa TechTarget senior news writer Beth Pariseau.
The data collection standard is now positioned as a potential source for consistent semantic conventions among AI observability tools, according to Parker, who also works as director of open source for observability vendor Honeycomb.io.
"Being able to have something that says, 'Hey, these are the things that you should care about, encoded in the data … and you can go and read a read a document that explains what all these attributes mean' … prevents [developers] from having to reinvent the wheel every time [they] want to do something with generative AI," Parker said.
Being able to have something that says, 'Hey, these are the things that you should care about, encoded in the data' … prevents [developers] from having to reinvent the wheel every time [they] want to do something with generative AI.
Austin ParkerOpenTelemetry core contributor
But while OpenTelemetry can provide data collection and labeling standards, the community has less consensus on approaches to instrumenting code for AI agent frameworks. Some OpenTelemetry proponents favor telemetry built into the project's code, while others favor separately publishing instrumentation libraries to GitHub.
Parker prefers a hybrid approach but leans toward built-in standards, comparing them to the standard setup in most cars.
"I fundamentally don't believe that it's good for us as a project, and I also don't think the user community expects that we always optimize around the high-performance F1 car," he said. "Over the 10 years, there is a much stronger expectation from the user community that open source is treated not just as a project, but as a product. Part of being a product is making decisions like that."
AI agent frameworks could also incorporate separate observability instrumentation, but Parker said he expects them to follow the example of newer programming languages such as Deno, a runtime for JavaScript, TypeScript and WebAssembly, which has pulled in OpenTelemetry rather than building its own.
"One of the things that I see a lot of people talk about when it comes to agentic AI, is this idea of TypeScript being the lingua franca for AI development because it's easy to build, it's easy to run, the models know a lot about it, and it's well documented," he said.
Parker said that over the next year, the OpenTelemetry project will continue to develop semantic conventions for generative and agentic AI, along with events and a new declarative configuration language. It will also explore its own AI assistant to improve the developer experience.
In the meantime, the AI upheaval in the industry is far from over, he predicted.
"The tech is here. People are getting value out of it. It is starting to have some transformative impacts," Parker said. "But who is actually going to be able to make a business out of this stuff? That is a much bigger question."
Beth Pariseau, senior news writer for Informa TechTarget, is an award-winning veteran of IT journalism covering DevOps. Have a tip? Email her or reach out @PariseauTT.
