Flywire's Orbit: 9,700 Code Review Runs in 30 Days, Workflow Not Autopilot
AI Engineer · youtube · 2026-10-13
Engineers at payments company Flywire presented their internal AI agent platform Orbit: over 9,700 code review runs in 30 days, serving roughly 400 engineers. The philosophy is "workflows, not autopilot"—predictable work stays in code, the model supplies judgment, and it can read code but never modify it.
- Workflows are defined in YAML, reviewed via merge requests like any code, declare per-step tools and access, and run on existing deployment infrastructure
- The architecture separates orchestration, execution and session state, with timeouts, rate limits and locks against duplicate runs
- Per-step restrictions cover network access, tools, skills and MCP servers; structured outputs drive subsequent branches
- Use cases: code review, translation updates, dependency fixes, ticket triage, bulk repo changes
- Honest gaps: new workflows get only a few real-case checks initially, golden datasets run on demand, online evals are still being built, and full sandboxing is a goal; execution success rate doesn't equal answer correctness
- Adoption path: build a useful skill, prove the recurring task, then turn it into a background workflow with explicit ownership and review
More from coding & agent
- Daniel Jeffries: only ~2% can actually use AI agents, mass adoption is 10-20 years away — Dan_Jeffries1 · 2026-10-13
- Fully local AI short film mimics Wes Anderson style with a detailed open pipeline — Mist3r-Robot · 2026-10-13
- Controlling his Mac's terminal from a phone via t3 code beta while waiting at speech therapy — NERDDISCO · 2026-10-13
- LLM cache verifier drift: reject-only calibration tripled error rate; 5% audits fix it — Reasonable_Royal_621 · 2026-10-13
- A style guide for making Claude-like models write normally, and its linguistic traps — gleech · 2026-10-13
- Open-source desktop agent dodaios runs local Qwen models via ollama — and shows where 16GB models fail — InternationalGap3698 · 2026-10-13