From Human Dashboards to Agentic Observability
Agentic observability is the emerging practice of designing observability platforms so AI agents, not humans, can monitor systems, understand incidents, and take action in production through APIs, automated workflows, and governed telemetry access. This shift turns monitoring data into a live control surface for autonomous operations, where agents diagnose failures, propose fixes, and learn from past incidents without relying on engineers to watch dashboards. It demands new tooling that explains what agents saw and did, exposes their reasoning, and feeds those insights back into development so agent behavior becomes a managed, auditable part of cloud operations instead of an opaque side process. The key takeaway is blunt: if you plan to run serious AI agents in production, your observability platform must evolve from a visual reporting tool into an automation substrate. New Relic’s latest launch makes this explicit, extending its observability platform so AI agents, as well as human engineers, can investigate incidents and retrieve system context. Latitude’s open-source release pushes in the same direction, treating AI agent monitoring as a loop that reports what went wrong and points to the fix instead of a dashboard to be watched. This is less a feature race than a change in operating model for cloud operations automation.
Latitude Shows Why Open-Source AI Agent Monitoring Matters
Latitude’s new open-source platform is a pointed answer to a growing pain: teams shipping autonomous agents to real users can no longer tolerate opaque behavior in production. The system is built to show what an agent is doing once it meets real users, catch where it breaks down, and route the fix back to the editor where the code already lives. That is AI agent monitoring designed for intervention, not observation. At its base is a discovery layer that gathers thousands of live conversations and clusters them into a single picture of what people ask for and where they hesitate, escalate, or drop off. Usage can be broken down by power users, one-time visitors, and accounts hitting failures most often, with individual sessions inspected alongside their cost, latency, and problems. A semantic search lets teams ask in plain language where users mention a feature the agent does not offer and immediately see matching conversations. This is agentic observability built to catch real-world failures and convert them into concrete signals a team can act on before problems spread.
Closing the Loop: From Signals to In-Editor Fixes
Latitude’s more interesting move is not just surfacing agent failures; it is closing the loop inside the developer workflow. When an agent keeps failing the same way, Latitude collapses those moments into a single signal that names the problem, counts how often it occurs, and attaches the likely reason. Signals come from automatic flaggers, annotations, or manual creation, and each receives an evaluation. Saved searches can be promoted into monitors that run against every new conversation so a pattern reaches the team before it reaches more users. This is what modern AI agent monitoring should look like: timed, contextual, and automatable. An MCP server delivers projects, traces, signals, searches, and datasets directly to a coding agent, so work happens where engineers already operate instead of in a separate console. Production conversations can be turned into datasets and reused as test sets, allowing teams to confirm that a fix holds before shipping. The platform is distributed under an MIT license and can run on a team’s own infrastructure, with a free tier and full source access for anyone who wants to read or modify it. That openness is not a side note; it is what gives enterprises the transparency and control they now demand as agent reliability becomes a defining concern at scale.
New Relic Turns Telemetry Into a Control Plane for AI Agents
While Latitude tackles agent transparency, New Relic is moving observability into the heart of autonomous incident response. The company announced the next evolution of its platform, introducing Autopilot and Ground Truth to power agentic AI-first businesses. On June 23, it formally extended its observability platform for an operating model where AI agents, not only human engineers, investigate incidents and retrieve system context. This follows its push to simplify enterprise OpenTelemetry adoption, giving platform teams a less disruptive path toward open, mixed-mode observability. Once telemetry is standardized and governed, AI agents need a way to use it. According to New Relic, “operations are going headless. AI agents won’t log in to view dashboards. They’ll pull what they need through APIs, reason about it, and act”. Autopilot is an out-of-the-box automated SRE agent built on its observability data substrate that automatically triages incidents, identifies root causes, and scopes possible remediations. It starts analysis the moment an alert fires, helping teams triage incidents, identify root causes, scope remediation paths, and improve early incident response. The system includes domain specialists in Kubernetes, tools optimized for Kafka troubleshooting, and cross-stack root-cause analysis, with more domain agents planned.
Governed Agent Access and the Road to Autonomous Cloud Operations
Autopilot is only half of New Relic’s agentic observability story. Ground Truth is designed for organizations that already run their own agents or orchestrators. Instead of forcing teams to adopt its agent, the capability gives tools such as GitHub Copilot, Claude Code, AWS DevOps, or custom orchestrators direct access to the deepest insights inside their observability data through agent-optimized tools that are difficult to get via public APIs or basic query layers. This is cloud operations automation designed around APIs and workflows, not user interfaces. New Relic Knowledge grounds conclusions in an organization’s specific runbooks and retrospectives, while External Model Context Protocol connections to Jira and GitHub pull in code and work-tracking context to pin down problems. Long-term memory captures operational knowledge that would otherwise stay trapped with individual engineers. Agentic operations will depend on data quality, not just model quality; an AI agent can only investigate or recommend action if it can retrieve trusted telemetry, understand system state, and connect errors to deployments, dependencies, infrastructure, and service-level objectives. A large enterprise measured a 1.1% error rate across more than 1,300 users, pointing to the kind of proof customers will look for before allowing agents deeper into operational workflows. As governance and security integrations such as Amazon Bedrock AgentCore rise, the message is clear: enterprises will not scale autonomous incident response without strong, auditable control over what agents see and do.
