New Relic launches an AI agent platform and new OpenTelemetry tools

As companies begin deploying AI agents en masse, an unavoidable question arises: who watches the watchers? New Relic, one of the key players in the observability market, has decided to answer it first. It has introduced a platform for building and managing AI agents, along with a set of tools for deeper OpenTelemetry data integration.

To grasp why this announcement matters, look at what is happening in enterprise AI right now. 2025 became the year of agents: autonomous AI systems that don't just answer queries but carry out chains of tasks on their own, analyzing data, calling APIs and making decisions. But the more complex the agent, the harder it is to understand what exactly it is doing and why. Classic monitoring tools, built for microservices and containers, are a poor fit. Agents are not just code. They are sequences of reasoning steps, model calls and tool invocations, any of which can go wrong.

That is the gap New Relic is aiming at. The new AI agent platform lets enterprises not only build and deploy agents but also observe their full lifecycle. That means tracing call chains, monitoring the latency of calls to language models, tracking token costs and detecting anomalies in agent behavior. In essence, New Relic is carrying the philosophy of APM (Application Performance Monitoring) over to agentic AI, where instead of HTTP requests you have to track prompts, and instead of database response time, inference delays.

The second part of the announcement is just as important, even if it sounds more technical. New Relic is expanding its integration with OpenTelemetry, the open standard for collecting telemetry data that has become the de facto lingua franca of observability in recent years. OTel makes it possible to collect metrics, logs and traces from any system in a unified format, without being tied to a specific vendor.

New Relic's improved tools for working with OTel data mean companies will be able to combine information about traditional infrastructure and AI agents in a single dashboard. This is critical: an agent that calls an external API with a three-second delay can look "broken" at the AI level, when the problem actually lies in the network or the backend.

For the market, this move by New Relic signals the emergence of a new segment: observability for AI. Until now, monitoring of AI systems has remained fragmented. Developers used LangSmith to debug LangChain chains and Weights and Biases for model experiments, and for production monitoring they often made do with homegrown solutions. New Relic offers a single entry point, built on an open standard, which lowers the barrier to entry and the risk of vendor lock-in.

Competition in this space will intensify quickly. Datadog is already experimenting with AI monitoring, Dynatrace is developing its causal analysis tools, and startups such as Arize AI and Langfuse are targeting exactly the same niche. But New Relic has the advantage of scale: thousands of enterprise customers who already use the platform to monitor infrastructure and can add AI observability without changing their stack.

There is also a broader context. As AI agents take on ever more responsible tasks, from managing cloud infrastructure to handling customer requests with real financial consequences, their transparency becomes not just a technical question but a regulatory one. The European AI Act and similar initiatives require explainability and auditability of automated decisions. Observability tools for AI agents are not a luxury but necessary infrastructure for complying with future rules.

New Relic is betting that the era of agentic AI will demand the same mature monitoring infrastructure we are used to seeing in the world of cloud applications. If that bet is right, the company is in the right place at the right time. If adoption of agents slows down, or the market fragments into dozens of incompatible frameworks, unifying all the data streams will become significantly harder. But one thing is already clear: AI observability is ceasing to be a niche topic and is becoming a mandatory layer of the enterprise stack.