Resolve AI launches a model and an AI agent for the on-call pager
Resolve AI says the pass rate of SRE-1 for one investigation went from 6.9% to 38.8% after training.
Resolve AI is a start-up that repairs faults in software while it operates. It launched SRE-1, a model that finds the root cause of an alert. It also launched Primary On-call, a tool that sends AI agents to handle each alert first. The company says the agent handles as much as 80% of alerts with no human help. It gives no external test.
How it works
The agent checks each alert across telemetry (system data), infrastructure, code and team information. It can close the alert, repair the problem where team policy lets it, or send the alert to a human. Then it monitors the system until recovery.
The numbers
- Pass rate for one investigation: 6.9% before training, 38.8% after.
- On 49 hard incidents, 5 SRE-1 investigations gave the same result as the top frontier model, at a lower cost.
- On 178 alerts, the full system solved 2 times more incidents than general agents for code.
What is not known
Salesforce says its agent closed more than 70% of the alerts of one team in first tests. Resolve AI says customers report 10 times less time to repair and as much as 95% precision of the root cause.
Insight: Vendors now compete on how much control the agent has
Gremlin and Monte Carlo make agents that wait for approval from an engineer for a test or a change. Resolve AI lets its agent close alerts and repair problems without a human where policy lets it. It gives a shadow mode to let teams check it first. Each vendor seems to want to be the first agent that teams trust to work without a human. Find which changes its agent makes without approval.
Sources
Posted