Automating Production Fixes with Agentic Observability

The Shift to Agentic Observability

Traditional observability relies on human-driven dashboards and manual investigation. The next generation of observability treats telemetry as a data stream for autonomous agents. By logging and tracing significantly more data than human-centric systems require, developers can provide agents with the exact code paths taken during failures, enabling them to move from simple responders to reviewers of agent-generated fixes.

Automating the Fix Loop

The core breakthrough is the use of file-based context injection. Instead of forcing agents to query complex dashboards, the system uses custom skills to pull production traces and logs into temporary files directly within the repository. This allows coding harnesses like Claude Code to operate in a familiar environment, using the same local debugging workflows developers use on their laptops. The process follows a continuous loop:

• Trigger: An event or periodic schedule initiates the agent.
• Context Gathering: Skills pull relevant production traces and logs into the repo as local files.
• Diagnosis: The agent analyzes the code path and telemetry to identify the root cause.
• Remediation: The agent generates a fix or a pull request for human review.

Deployment and Security

To address enterprise requirements, the system supports deployment within a private VPC, ensuring production data does not leave the customer's infrastructure. Users can configure their own sandboxes and harnesses, maintaining control over the agent's environment while leveraging pre-baked skills for common observability platforms like Pyroscope. This approach allows teams to treat AI-based incident response as an open-box process where the agent's actions, transcripts, and session data remain fully auditable.