Grep AI is pitching agents that handle repetitive operational tasks and turn recurring steps into reusable code. In an Oct. 5 announcement, founder AJ Asver described a system that reserves more intensive AI reasoning for exceptions and unfamiliar cases.
The work he describes includes gathering information, checking it against a procedure and deciding what happens next. Grep calls its harness AgentRun, the software that manages how an agent carries out a job. Its product site describes using smaller models and code for repeated steps while leaving judgment to frontier models.
Customer results remain company claims
Asver said Grep agents review compliance alerts at IG, a U.K. online brokerage, helping more than 80% of traders onboard without manual intervention. He also said customers including IG, Airwallex, Worldpay, FV Bank and Citizens Bank had saved more than 250,000 hours of operational work across uses such as onboarding, document verification and underwriting.
Those are company-reported figures. The announcement does not give their measurement period or calculation method, and the cited material does not independently verify the results.
Y Combinator's profile lists Grep in its Fall 2026 batch and identifies Asver as a founder. The company describes on-demand, scheduled and batch operation through its console, MCP, API or SDKs, with sandboxes, audit logs and cost controls. Its site says candidate workflows are tested before activation and each run records the workflow version it used.
Comments