Splunk Preps Second Open Source LLM for Telemetry Data

Splunk is preparing to release an open-source artificial intelligence (AI) model specifically designed for analyzing telemetry data on the Hugging Face platform. At its .conf26 conference, the company also unveiled plans for a Universal Collector, set to launch in beta in 2027, aimed at simplifying the collection of various telemetry data types using OpenTelemetry.

Raja Mukhopadhyay, vice president of observability cloud for Splunk, explained that an AI model trained to reason across log data will enable DevOps teams to observe AI applications and agents at scale efficiently.

The new open-source AI model is needed because existing general-purpose AI models are trained on text, code, and video data, whereas metrics, a type of numerical data, require an LLM specifically trained to handle such information. While general-purpose AI models could theoretically analyze logs, the vast volume of this data would likely overwhelm context window limits.

Consequently, the second AI model developed by Splunk focuses on efficiently reasoning across log data, a task tailored to the unique nature of this type of numerical information.

As DevOps teams increasingly integrate AI agents into their workflows, the amount of generated telemetry data is surging, making efficient analysis crucial. Mukhopadhyay highlighted that the emergence of agentic engineering raises questions about what DevOps teams will observe in the future, suggesting that AI agents may soon generate code in machine languages beyond human comprehension.

While validation of generated code by additional AI agents is necessary, DevOps teams should anticipate an exponential increase in applications running in production environments.

To manage this growing complexity, software engineers will need to transition from direct coding roles to overseeing fleets of AI agents assigned specific tasks. This shift aims to streamline software development and deployment, ultimately leading to higher quality applications. However, the quality of software generated by AI agents will be paramount, as flawed applications could prove more problematic than the benefits they offer.