The Duality of AI-Powered Observability

As the pace of AI innovation continues to accelerate, the latest focus is on agentic orchestration and workflows. The move to agentic is a natural evolution from the LLM use cases of a year ago, and agentic AI unlocks compelling new ways to increase productivity across a variety of domains, from incident response to research to software development. But these new capabilities come with a substantial increase in complexity. It is becoming more difficult to monitor, debug, and govern agentic systems vs. the relatively linear systems of LLMs. This creates a dual opportunity that entirely transforms how we think about observability.

There are two complementary revolutions happening simultaneously:

1. Observability for Agentic AI: Providing the visibility needed to facilitate AI adoption by making black-box AI systems transparent and debuggable.
2. AI for Observability: Using intelligent insights and automated actions across complex workflows to transform traditional reactive monitoring into proactive, predictive operations.

Observability for AI: Making Black Boxes Transparent

Consider a modern AI-powered application where each agent interaction involves multiple of the following:

• Tool calling with external APIs
• Context passing between agents
• Parallel processing branches
• Error handling and retry logic
• Performance optimization decisions

Traditional Application Performance Monitoring (APM) is not able to provide transparency into this complex web of interactions. You may be able to see HTTP request durations but miss the AI decision-making process entirely. 

To unlock AI’s full value, you need deep visibility into these agent-level interactions. At the same time, their performance within the APM entity is also needed in order to quickly identify and resolve issues specific to your AI components. In order to advance your AI adoption and not fall behind your competitors, you need to reduce and ultimately eliminate the unknowns about how these systems are performing and why.

With Agentic AI Monitoring, we are expanding New Relic’s existing AI Monitoring capabilities to provide that same automatic, detailed visibility, but extending it into every agent and tool call within your multi-agent collaborations. You get granular insights into tool utilization, performance, and errors with a view that shows which agents and tools were called, in what order, and key performance data. No manual instrumentation required. This will come with Distributed Tracing with AI Agent Spans to allow you to track every interaction between AI agents and the tools they’re calling. It will also highlight changes in how your AI components communicate – and all of this will be available in one single view.

Additionally, we are releasing AI Response Filtering, where you can instantly filter AI-related calls in your catalog view to pinpoint, address, and fix issues faster. Gone are the days when you had to manually sift through thousands of logs to find a problematic prompt or response. And of course, we are always continuing to enhance our LLM Monitoring to include new APIs and protocols, now including support for ConverseAPI (AWS Bedrock’s API for building AI applications with LLMs).

Consider the scenario where you are working hard to rapidly adopt AI, building powerful but complex multi-agent systems. The problem? You're basically flying blind. When these AI apps fail, there is limited to no visibility into what went wrong, putting your massive AI investments at risk. Take the example of Devin, an engineering leader at an e-commerce company. His team is building a new AI recommendation engine that is quite complex, essentially a team of AI agents. Each agent has a specific job:

• One agent finds relevant products
• Another agent analyzes customer reviews
• A third agent personalizes the final recommendation based on user preferences

When the recommendation engine provides a user with a slow or incorrect recommendation, troubleshooting is a frustrating guessing game for Devin and his team. They can't see the complex interactions between their AI agents, which means slower root cause analysis and longer downtime. The business is at risk of losing revenue, and Devin's team struggles to innovate while buried in debugging tasks.

Unlike siloed tools or generic APM, New Relic AI Monitoring eliminates the guesswork and provides complete visibility into your agentic AI applications. New Relic Agentic AI Monitoring correlates AI performance with your entire stack automatically. It maps and monitors every agent and tool call in your system so you can instantly pinpoint bottlenecks, identify errors, and understand exactly how your tools are being used. This is observability designed for AI.

Our new Agents Service Map provides a complete view of all interactions between your AI agents. The AI Trace View lets you see every step, including which agents were called, latency, and errors. You can drill down into specific agent and tool traces to find the root cause of issues in seconds. With Autogen monitoring, you can see agent-to-agent communications in the Waterfall view you already know. And the AI Inventory gives you a single place to see all the AI agents & services you're managing. This simplifies governance and helps engineers quickly focus on the specific tools they are responsible for.

This is a game-changer. Instead of spending days debugging, you can identify and solve issues in minutes, and get back to building great features. By accelerating issue resolution for agentic AI services, New Relic can help prevent lost revenue and reduce operational costs. This de-risks your AI investments and empowers your business to make better decisions based on reliable AI outputs.

AI for Observability: Intelligence at Every Layer

The same AI techniques that power these agentic workflows can be applied to observability itself, allowing you to create intelligent systems that think about your infrastructure the way your engineers do—but at machine speed and scale. The New Relic MCP Server is the answer to bringing our AI capabilities to any MCP-compatible AI agent. You should have the freedom to choose your tools, not be locked into them. That’s why, unlike other solutions tied to a single AI agent, our MCP Server is built to be completely agnostic to any specific AI client. It’s an extensible toolkit designed to work with MCP-compatible agents from GitHub CoPilot to ChatGPT to Claude—the choice is yours.

This means you can bring all your observability data directly to your preferred GenAI tool and take action, right from within your existing workflow. This allows you to:

• Fetch alerts and incidents with natural language
• Generate intelligence reports without leaving the IDE or AI assistant
• Analyze deployments using conversational queries across multiple tools

At New Relic, we are always about meeting our customers where they actually work—not forcing you to context-switch, tab-switch, or window-switch. We are designing the New Relic MCP Server to enable seamless, standardized interaction between New Relic and the growing ecosystem of AI tools that support MCP. AI assistants like GitHub Copilot have become second nature for developers, but they typically operate in a silo. This means engineers have to context-switch to their observability tool to troubleshoot, which costs them additional time and effort. New Relic is the only platform that brings observability directly into the conversational workflow developers are already familiar with.

As an example, Amy is an engineering leader at a major social media company. Her team leverages AI assistants to write and deploy code faster. When there’s a production issue, her engineers are forced to constantly switch contexts. They leave their AI assistant, log into separate observability tools, run queries to find the right data, and then toggle back to their code. This disjointed workflow of data silos is slow and inefficient, draining productivity and slowing down incident response. For a social media platform where user experience is everything, this delay risks impacting user engagement and ad revenue.

Amy's team explored other tools, but found that Datadog's approach to MCP monitoring required manual “skill packs” just to connect to their AI, creating friction. Dynatrace was too restrictive, trying to lock them into a proprietary AI instead of the assistants her developers already use. Neither option offered the seamless, conversational workflow they needed.

New Relic AI MCP Server eliminates this friction by bringing observability directly into the AI assistants your team already uses. It integrates New Relic AI into any MCP-compatible agent, so engineers can query data and get insights in their natural workflow. Imagine your engineers asking their AI assistant in plain English, "What's our platform's error rate since the last deployment?" and getting an instant answer, powered by New Relic, without ever leaving their coding environment. This is observability at the speed of conversation.

By embedding New Relic insights directly into AI agents, engineers can troubleshoot and resolve issues in a single, unified workflow, leading to immediate productivity gains. Engineers can use natural language to instantly query telemetry data, analyze deployment impacts, and generate alert reports, dramatically reducing Mean Time to Resolution (MTTR). This makes observability data accessible to everyone on the team, regardless of their skill level with query languages. If they can ask a question, they can get insights from New Relic, leading to smarter, faster decisions across the board.

By accelerating incident response and embedding observability directly into the Software Development Life Cycle (SDLC), New Relic improves developer productivity and protects user experience. This seamless integration streamlines key IT Service Management (ITSM) and IT Operations Management (ITOM) workflows, reduces operational costs, and empowers your team to innovate faster.

The Strategic Transformation

Observability for AI removes the barriers to AI adoption by making complex systems understandable and reliable. AI for Observability transforms your operations from reactive to predictive, from manual to automated. Together, they create an intelligent feedback loop: AI systems become more observable and reliable, while observability platforms become more intelligent and proactive. This is how New Relic is building confidence to allow our customers to innovate at the speed of software in an AI-accelerated world.