The problem
A production error is only the beginning of an investigation. Someone still needs to gather logs, find the relevant code, understand related issues and turn that evidence into a proposed change.
Those steps cross several tools, and much of the information gathering is repetitive.
What I built
Using RAVEN’s integrations, I built an agent workflow that detects faults, gathers context, investigates possible causes and prepares fixes. It can create tickets and pull requests, giving an engineer a concrete result to inspect.
- 01Detect a fault and collect supporting evidence
- 02Investigate relevant code and related issues
- 03Prepare a proposed fix
- 04Run applicable checks and create a pull request
- 05Human review and release decision
Where judgment stays with the engineer
A plausible explanation still needs evidence. I use source inspection, tests and independent review to check an agent’s conclusions, and revise the diagnosis when the evidence contradicts it.
The workflow prepares changes for review. Preparing a fix, opening a pull request and deploying that fix are distinct steps.
The result
The workflow has detected and created tickets for real production faults. In October 2026, it detected and ticketed 15 faults in one service. Fix preparation extends the workflow, with changes subject to human review before release.
What this can look like for your team: start with one service and a defined class of errors, agree on the evidence and checks required, and make the handoff to an engineer explicit.