The Next Challenge for AI Is Software Reliability

AI is making it easier to write software.But writing software and operating it reliably are two very different problems.

As more code is generated and deployed, the responsibility of monitoring applications, identifying failures and responding to incidents becomes even more important.

PagerDuty introduced Paige in November, an SRE agent designed to help engineering and operations teams detect, investigate and remediate production issues.The announcement is interesting because it focuses on a part of software engineering that does not always receive as much attention as code generation.

What happens after the software is deployed?

Building Software Is Only the Beginning

A production application depends on many interconnected components.There might be a frontend, several backend services, databases, queues, external APIs and infrastructure providers.A problem in any one of these components can affect the entire application.

For example, a slow database query could increase API response times. This might cause requests to time out, which could eventually affect the user experience.Identifying the actual cause requires understanding how these components interact.Simply looking at the first error message is rarely enough.

Why Incident Investigation Is Difficult

When an application fails, engineers often need to examine multiple sources of information.Logs provide details about what happened. Metrics help identify unusual behaviour. Traces show how requests move through different services.

These sources are useful, but connecting them can take time.An engineer might need to identify when the problem started, compare it with recent deployments and determine which service is responsible.AI agents could help bring this information together.

They can investigate relevant signals, identify possible causes and help engineers understand what changed.

From Monitoring to Investigation and Remediation

Traditional monitoring systems are generally designed to detect unusual behaviour and notify the responsible team.An SRE agent can go further.

It can help investigate an incident, connect information from different sources and propose possible solutions.In appropriate environments, it may also carry out predefined remediation actions.

This creates the possibility of reducing the time between detecting a problem and restoring the application.However, automated remediation needs careful boundaries.

Restarting a service might be relatively straightforward. Modifying production database records or changing access permissions is a different level of responsibility.The system needs to understand which actions are safe to perform automatically and which require human approval.

Observability Becomes More Important

For an AI agent to investigate a production issue effectively, it needs access to useful operational information.This makes observability increasingly important.

Applications should produce structured logs, meaningful metrics and traces that connect related operations.Without these signals, an AI agent has very little reliable information to work with.This is also a reminder that observability should be designed into applications from the beginning rather than added only after something goes wrong.

The Future of Reliable Software

I think reliability engineering will become an increasingly important part of AI driven software development.If AI helps us ship software faster, we also need better systems to make sure that software continues working after deployment.

The goal should not be to remove engineers from incident response entirely.It should be to reduce repetitive investigation, surface useful information faster and help engineers make better decisions during production incidents.

Generating code is one part of software engineering.Keeping that code reliable is another.