Flywire Platform Team Discusses Orbit: AI Code Review in Deterministic Workflows

Eduardo Simon and Javier Casal from Flywire's platform team introduce their internal Orbit platform in a video published on October 12, 2026, by the ai.engineer channel. They demonstrate how to embed large language model judgment into deterministic engineering workflows.

According to the publisher's description, Orbit uses YAML definitions for workflows that undergo merge request review like regular code, explicitly declaring tools and access permissions for each step. This design keeps predictable work in code while letting the model supply judgment. The platform is expanding across approximately 400 engineers.

The publisher states that more than 9,700 code review runs in the past 30 days came from a workflow allowing the model to read but never modify code. Beyond code review, the pattern supports translation updates, dependency fixes, ticket triage, and bulk repository changes. The architecture separates orchestration, execution, and session state, adding timeouts, rate limits, and locks against duplicate runs.

The speakers are candid about remaining work: new workflows initially receive only a few real case checks, golden datasets run on demand, online evaluations are still being built, and full sandboxing remains a goal. The publisher notes that reported execution success rates do not establish answer correctness. For engineering readers interested in safely integrating AI steps into existing infrastructure, this video offers a concrete adoption path and risk disclosures.