Models

Resolve AI Launches SRE-1 Model and Primary On-call Agent to Automate Production Incident Response

Resolve AI has unveiled SRE-1, a 27 billion-parameter model trained to diagnose root causes of production alerts, alongside Primary On-call, an AI agent that responds to incidents before human engineers are paged.

·4 min read
Resolve AI’s SRE-1 model finds root causes of production alerts as AI agents go on call
Resolve AI’s SRE-1 model finds root causes of production alerts as AI agents go on call

Resolve AI Inc. unveiled two new tools this week aimed at automating how software teams respond to production incidents. The company introduced SRE-1, a specialized model designed to identify the underlying causes of production alerts, and Primary On-call, a service that deploys AI agents to handle on-call responsibilities in advance of human staff.

The announcement underscores Resolve AI's view that while artificial intelligence has reshaped software development practices, it has made minimal inroads into software operations. The company points out that production environments lack a public dataset comparable to the open-source code repositories that trained most large language models. Each organization's incidents remain confidential, creating a data scarcity problem for training operational AI systems.

Building and Testing SRE-1

SRE-1 is a 27 billion-parameter model built on open weights and developed by 16 AI researchers. The team completed its final training run in 7.4 days using up to 512 graphics processing units and relied on reinforcement learning as its sole training methodology. When measured against the same set of incidents before and after training, the model's success rate for single investigations jumped from 6.9% to 38.8%.

Resolve AI built SRE-1 using Snowflake Inc.'s Cortex Training, a serverless platform for customizing open-weight models. Vivek Raghunathan, Snowflake's Senior Vice President of Engineering, noted that Resolve AI served as the first design partner for the service.

In real-world testing involving 49 difficult live production cases, each tested five times, an ensemble of five SRE-1 investigations performed comparably to the strongest frontier model available while operating at reduced cost. Across a broader benchmark of 178 production alert investigations, Resolve AI's complete system resolved roughly double the number of incidents that general-purpose coding agents handled.

How Primary On-call Operates

Primary On-call puts SRE-1's capabilities into production use. When an alert triggers, the agent investigates across telemetry data, infrastructure configuration, codebase information and operational procedures. Based on its findings, the agent can dismiss the alert, apply a fix within the team's established policies or route the issue to a human engineer. After taking action, the agent validates the solution against production metrics and continues monitoring until stability is confirmed.

The system consolidates related alerts and investigations into a unified view of the root problem. For teams hesitant to immediately hand over full on-call duties, Resolve AI provides a shadow mode that logs what action the agent would have taken on each alert without actually executing it.

According to Resolve AI, Primary On-call handles up to 80% of paged alerts without requiring human intervention. The company reports that customers have achieved a tenfold reduction in mean time to resolution and achieved root-cause accuracy rates reaching 95%.

Salesforce Inc. is among the early adopters testing the system. Bhupesh Patel, a software engineering principal architect at Salesforce, shared that during the initial dark launch phase, the agent resolved more than 70% of alerts for the CRM Analytics team "without waking an engineer."

No engineering team can keep up with how fast code and patches land in production. The answer is agents that are accurate and fast enough to sit on the critical path, and that takes a model built for production.

Spiros Xanthos, Chief Executive of Resolve AI

Availability and Company Background

Primary On-call is now available to all customers. Access to SRE-1 remains restricted to Resolve AI's platform.

Resolve AI operates as a venture-backed startup with cumulative fundraising exceeding $190 million. The company closed a $40 million extension to its Series A in April, led by DST Global and Salesforce Ventures, which valued the company at $1.5 billion. Its customer roster includes Coinbase Global Inc., DoorDash Inc., Figma Inc., MSCI Inc., Nasdaq Inc., SAP SE and Zscaler Inc.