In October 2026, David Fabritius, a product marketing manager at New Relic, argued: “As enterprise software architectures transition toward highly ephemeral environments and autonomous Agentic AI workloads, legacy monitoring methods become completely obsolete.” This assertion serves as the foundation for the company’s recent push into Advanced APM, a suite that combines automated dependency mapping with self-maintaining business entities. While the marketing framing emphasizes the shift from system metrics to business outcomes, the underlying mechanism reveals a more fundamental change in how infrastructure vendors are adapting to the agent stack.
The shape of telemetry has changed. Traditional application performance monitoring focused on request and response traces. Agent workloads, however, produce complex tool-call chains, multi-agent handoffs, reasoning steps, and variable token spend. To capture this, New Relic launched Agentic AI Monitoring in November 2025. This tool provides visibility into agent and tool calls, featuring an AI Inventory view and an Agents Service Map to visualize interactions. Notably, the company states that this capability is only possible by building Agentic AI Monitoring on top of tried-and-true APM and infrastructure monitoring tools. This reveals the incumbent distribution wedge: existing observability platforms are leveraging their deep integration into current infrastructure to become the primary record for new agentic workflows.
There is a second, more subtle move occurring. By launching the New Relic AI Model Context Protocol server, the company allows AI assistants like GitHub Copilot, ChatGPT, Claude, and Cursor to query observability data directly within developer workflows. This transforms the observer into a participant in the agent loop. As Stephen Elliot, an IDC Group Vice President, noted in November 2025, “Observability platforms will need to fill that gap by making observability capabilities available to any MCP-compatible agent.”
The forcing function for this transition is what New Relic calls agent debt. In a vendor-commissioned survey of technology leaders released in June 2026, 94 percent of respondents rated AI-generated code as higher quality at review time, yet 78 percent report more incidents once it ships, and 62 percent ship AI-generated code to production without line-by-line manual verification. Brian Emerson, New Relic Chief Product Officer, was direct: “Agent debt is the hidden operational tax of the agentic era and a threat to enterprise uptime.” He added: “The promised productivity gains of AI coding assistants are a mirage if they simply shift the bottleneck from developers writing code to SREs fixing it.”
The financial stakes are high. According to vendor research, high-impact outages carry a median cost of $2 million per hour, or more than $33,000 per minute. To address this, the company released New Relic Preflight, an open-source observability assistant for agentic coding that tracks token spend, call volumes, and tool-selection quality. The Advanced APM suite further embeds custom business KPIs, such as checkout duration or revenue throughput, allowing leaders to quantify financial risk, such as $50,000 per hour at risk. These figures are vendor-reported or vendor-commissioned.
This observability layer connects to a broader infrastructure ecosystem. The AI gateway category is becoming a protocol-aware security enforcement layer, while the agent runtime safety layer acts to constrain agent behavior. Furthermore, cloud vendors are shipping protocol-native agent gateways. Observability is the layer all of these depend on, as enforcement and audit require a record of what agents actually did.
A tension remains between open standards, such as the Model Context Protocol and OpenTelemetry GenAI semantic conventions, and the platform gravity of proprietary telemetry enrichment. Monitoring what agents do is not the same as governing it. For builders, the implication is to verify whether agent monitoring covers tool calls and cross-agent handoffs in their specific stack, and to determine if telemetry remains portable. Decision-makers should also monitor the status of features moving from preview to general availability, the progress of FedRAMP High and Department of Defense Impact Level 4 authorizations, and treat performance claims, such as the vendor-reported claim of visibility up to 5x faster than manual tracking, with appropriate caution.
Verification note: All product figures and comparisons, including the 5x faster claim and the 15-minute competitor window, are vendor-reported. The survey percentages were commissioned by New Relic. The Advanced APM blog post is vendor marketing authored by a New Relic product marketing manager.