For the last few decades, Application Performance Management has been about visibility and observability. Engineers used Monitoring, metrics and dashboards to understand how their complex systems are performing. Mohit Agrawal, an engineering leader who has built large-scale observability and reliability systems, sees the field shifting toward systems that diagnose and repair issues without waiting for humans.
Understanding Replaces Monitoring
Traditionally APM relied on engineers building dashboards, logging and manual systems to trace performance issues. In the current age, the ever growing modern architectures make this approach untenable. Microservices, serverless functions, and multi cloud infrastructure generate too much data for humans to sift through and triage.
Agrawal puts it plainly: engineers can't manually connect logs, metrics, and traces across hundreds of services. Tools need to understand system behavior in real time.
The APM tools have matured to a place where they use AI / ML models which are trained on the telemetry data to infer causal relationships and catch anomalies. These tools can catch issues as they’re happening and identify first deviations that might have triggered failures on various downstream systems.
Automated Root Cause Analysis
Agrawal considers automated root cause analysis the core of modern APM. The goal is converting noise into narrative that can help identify issues automagically.
Engineers and reliability teams have been plagued by Alert fatigue. These alerts could be sometimes too brittle and AI addresses this by generating crisp and readable summaries of what failed and what could’ve caused it. AI enabled APM tools can analyse dependency graphs, event timelines and service architectures. These AI models can automatically do root cause analysis within seconds. They have exposure to your deployment changes, configuration changes and can identify which change or regression caused the problem, helping the team just focus on resolution instead of root cause analysis.
The difference between "CPU is high" and "Service A's new library call introduced a memory leak" determines whether a team spends hours debugging or minutes fixing.
Context-Aware Anomaly Detection
Traditional APM tools relied on static rules and thresholds. Current APM systems use context-aware anomaly detection. These machine learning models can understand patterns at the service and cluster level where baselines could be evolving continuously.
In the era of digital revolution, traffic patterns can change at a whim. APM systems have to learn that so that AI doesn’t just flag a spike, it can determine which spikes matter. These intelligent models surface anomalies that matter and could result in cascading failures. This reduces noise and gives better signal to the teams on the frontlines.
Automated Fixes with Generative AI
The next phase of this evolution comes from integrating Generative AI to remediate these issues surface by AI models. APM platforms will diagnose problems and propose solutions. Sometimes these AI agents and also implement these solutions
Agrawal envisions AI generated runbooks and self healing code patches. When an outage or regression is detected, the APM system could:
Using the structured telemetry and recent code changes generate a hypothesis for the root cause
Create a remediation plan which could include a rollback, config changes or scaling adjustment
Create a code fix and deploy it automatically if that matches confidence thresholds.
Agrawals says “AI Agents are matured enough where they could troubleshoot issues and fix them in an automatic way. This is getting to a golden era of self driving agents for root cause detection and remediation”
In this model, the APM platform becomes an autonomous operations partner that can reason, act, and learn.
A growing number of teams already trust AI to write production grade code. As teams use these AI coding tools, overtime they would trust AI agents to write safe fixes under human supervision.
The Bottom Line.
According to Agrawal “Monitoring, root cause detection and issue remediation, are all converging into a close learning loop. Modern APM tools will summarize issues, fix them and document them so that they can learn from this over time. The organizations that don’t adopt these modern paradigms will become slow and fall behind"
About Mohit Agrawal
Mohit Agrawal is an engineering leader based in Silicon Valley. Mohit Agrawal has a track record of 13 years leading Growth Engineering and Mobile Engineering teams at various healthcare, enterprise and financial companies. He has worked on various fast growing startups and has also been mentioned in major publications like Forbes where he talks about Product Led Growth and Growth Engineering.