Project
Jarvis
AI support agent resolving ~25% of
- ~25% resolution rate
- Slack-native, no new surface to learn
- Passed formal security review
Jarvis is an AI agent that handles a meaningful chunk of the questions that land in
NerdWallet’s #infra-help Slack channel. It’s built on the Model Context Protocol and
LangGraph, using a Supervisor pattern that routes each incoming question to the right
sub-agent.
The real-time context is what makes it useful: Jarvis pulls live state from PagerDuty and OpsLevel so answers reflect what’s actually happening in production, not what was true when a runbook was last edited.
It lives in Slack, which is where the questions already happen. It passed formal security review. Rafael, a junior engineer on the team, now owns several of the features end to end.
A longer write-up is on the way.