Resolve AI says production systems have no public training corpus comparable to published code, because incidents stay private at each company. It trained SRE-1 with reinforcement learning alone, over a final run of 7.4 days on up to 512 GPUs. On the same incidents, the pass rate for a single investigation rose from 6.9% to 38.8%.
Primary On-call investigates each page using telemetry, infrastructure, code and operational knowledge. Depending on what it finds, the agent can close the alert, mitigate the problem where team policies allow, or escalate to a human. After acting, it checks the fix against the production baseline and groups related alerts into a shared view. A shadow mode records the calls the agent would have made without acting on them.
Resolve AI says up to 80% of paged production alerts are handled without escalation, with up to a tenfold improvement in mean time to resolution and root-cause accuracy of up to 95%. Salesforce reported that in the first few weeks of dark launch testing, the agent closed more than 70% of its CRM Analytics team's paging alerts without waking an engineer.
Why it matters if you build agents
Teams building incident-response agents now have a model tuned for production investigations, though the performance figures come from the vendor and have not been independently verified. Before handing over the pager, teams can use shadow mode to compare the agent's planned actions with what engineers would have done. Teams should also check which mitigations their policies permit the agent to take on its own.
Get stories like this every weekday
One short email with the day's agent tooling news.