Production alerts turned into fix pull requests
An agent picks up production alerts from chat, investigates logs and code, and opens a fix pull request for engineers to review.
The use case
Engineering teams get flooded with production alerts in Slack, and many recur because nobody has time to fix the root cause. An agent connected to the alert channel, logs, and code repository investigates each alert, traces it to a likely cause, and drafts a fix as a pull request with an explanation. Carvana reported cutting recurring alerts this way. Engineers still review every pull request, run tests, decide whether the diagnosis is right, and merge. The agent should not deploy changes or touch production systems on its own.
Example workflow
- Connect the agent to the alert channel, logs, and repo
- Agent investigates new alerts
- Agent posts its diagnosis in the thread
- Agent opens a fix pull request
- Engineer reviews, tests, and merges
Tools
- Claude
- Slack
- GitHub
- Datadog
Risks and guardrails
- Wrong root cause diagnoses
- Fixes that break other behavior
- Access to production logs
- No auto-deploys
Sources
Written by Agents Lookup. Credits and official pages for the tools this use case names:
- Anthropic customer story: Carvana: claude.com/customers/carvana
- Claude (official page): claude.ai (checked 2026-11-09)
- Slack MCP server (official page): docs.slack.dev/ai/slack-mcp-server (checked 2026-11-09)
- GitHub MCP server (official page): github.com/github/github-mcp-server (checked 2026-11-09)
- Datadog MCP server (official page): docs.datadoghq.com/mcp_server (checked 2026-11-09)