Sentry's engineering team built an automated debugging workflow using their own Seer AI agent and Claude routines. Seer monitors issues, performs root cause analysis, proposes fixes, and opens GitHub pull requests automatically. A Claude routine runs hourly, checks a Slack channel for new Seer-generated PRs, identifies the best reviewer based on commit history, verifies the PR is still open, and pings the reviewer with a request for action. Early results show a 21% increase in PR action rate and a 13% increase in 48-hour response rate. The workflow took several weeks of iteration to refine, and Sentry is exploring further automation such as automatic PR assignment.
Table of contents
It all starts with knowing what’s broken and whyCreating a routine for reviewEarly resultsTry it yourselfQuestions this post answers
What early metrics did Sentry see after rolling out their automated Seer debugging workflow?
After rolling out the workflow in mid-July, Sentry saw roughly a 21% increase in PR action rate, a 13% increase in 48-hour response rate on open PRs, and a 12.5% increase in close-without-merge PRs. The close-without-merge increase was explained by duplicate fixes or engineers opting for more comprehensive solutions, not by poor fix quality. Teams evaluating AI debugging tools track real adoption numbers like these on daily.dev.
How does Sentry's Claude routine decide who should review a Seer-generated pull request?
The Claude routine sends a prompt to the Seer agent asking who is best suited to review the PR. Seer examines the GitHub commit history for the files flagged in its root cause analysis and identifies the engineer who has worked most closely with that code. The routine then tags that person in the Slack thread and asks them to merge, close, or leave feedback. Engineers building similar PR routing automations find implementation patterns for this kind of workflow on daily.dev.