Ramp OCA On-Call Assistant
Suggested workflow: An assistant that provides investigation progress and reports in incident channels, requiring approval for destructive operations. Task: The Ramp OCA On-Call Assistant monitors incident channels to provide investigation updates and reports. It operates in read-only mode, requiring human approval via Slack buttons for any destructive actions like restarts or rollbacks. The workflow integrates with alert systems to join incidents automatically.; tools: Tool described in the source. Human role: Source evidence is unavailable; specific human review roles and operational divisions are not detailed.; limitations: The agent is strictly read-only for system changes, relying on human approval for all destructive operations. Source reliability is secondary, derived from a public technical case summary. Integration requires specific alerting tools (e.g., PagerDuty, Datadog) and Slack for approval workflows..
Tools
Tool described in the source
Published on Agent Work Atlas:
Overview
Suggested workflow: An assistant that provides investigation progress and reports in incident channels, requiring approval for destructive operations.
Workflow
The Ramp OCA On-Call Assistant monitors incident channels to provide investigation updates and reports. It operates in read-only mode, requiring human approval via Slack buttons for any destructive actions like restarts or rollbacks. The workflow integrates with alert systems to join incidents automatically.
Reported result
In four weeks, the assistant supported 575 of 1,220 incident-fix PRs, reducing the number of engineers pulled into incidents by 37%. It posts initial updates in approximately 5 minutes and full reports in about 13 minutes. These metrics are attributed to the source description of Ramp's implementation.
Human role
Source evidence is unavailable; specific human review roles and operational divisions are not detailed.
Limitations
The agent is strictly read-only for system changes, relying on human approval for all destructive operations. Source reliability is secondary, derived from a public technical case summary. Integration requires specific alerting tools (e.g., PagerDuty, Datadog) and Slack for approval workflows.
Try it
Act as an SRE agent developer. I need to build a prototype for an on-call assistant similar to Ramp's OCA. Inputs: [replaceable: Incident channel name, Alert source (e.g., PagerDuty/Datadog), Slack workspace for approvals]. Steps: 1. Set up a listener for new alerts in [replaceable: Incident channel]. 2. Upon trigger, perform read-only checks: fetch recent deploys, config changes, and similar past incidents. 3. Generate a preliminary investigation update and post it to the channel within 5 minutes. 4. Continue parallel hypothesis investigation using sub-agents or tools. 5. Post a full report within 13 minutes, citing specific log lines or queries for every conclusion. 6. For any proposed fix (restart/rollback), do NOT execute. Instead, post a Slack approval button. 7. Only proceed if a human clicks approve. Deliverables: A script or agent configuration that implements this read-only investigation and approval-gated workflow. Quality Check: Verify that no destructive commands are executed without human input. Measure time-to-first-update and time-to-full-report against the 5/13 minute targets. Safeguards: Ensure the agent has zero write permissions to production systems. Require human confirmation before deploying to any live incident channel. Ask clarifying questions if the alert source or approval mechanism is undefined.
Source reports are not independently verified. Publication date is the catalog publication date, not the original article date.