What's Next?#
You've seen the full arc — from a live production error, through automated AI investigation, to a pull request with a fix and evidence. Here are some ideas for taking the demo further.
Extend the Agent#
Add more investigation tools
The agent currently uses logs, traces, and the Live Debugger. You could extend it with metric queries (dtctl query on metrics) or service topology lookups to broaden what it can detect.
Tune the investigation prompt
The prompt lives in agent/templates/agent_prompt.md. Editing it lets you guide the agent toward specific error patterns, restrict it to certain namespaces, or ask it to consider dependencies you know about.
Multi-agent investigation
Split detection, investigation, and remediation into separate Claude agents that hand off to each other — for example, a triage agent that classifies the problem type before a specialist agent investigates.
Production Considerations#
Human approval gates
Add a workflow step that posts the agent's findings and waits for a human to approve before the PR is auto-merged. Dynatrace Workflows supports approval steps natively.
Confidence thresholds
Only auto-create PRs when the agent's confidence score exceeds a configurable threshold (e.g. 0.85). Lower-confidence findings can still be reported as issues for human review.
Notifications
Send a Slack or email alert when the agent opens a PR, so the on-call engineer knows a fix is ready for review.