Traversal raises $120m at a $1bn valuation for its AI site reliability engineer

Why it matters

Traversal is betting that keeping production systems running is a separate AI problem from writing code. It cites ORCA-bench, a benchmark it co-developed with Cornell Tech and Columbia researchers, on which frontier models reach about 25% accuracy on medium-difficulty reliability tasks with hallucination rates often above 40%. Its agent learns normal behaviour from logs, metrics and traces and uses causal machine learning to find root causes. Customers named include PepsiCo, DigitalOcean, Nubank and Harvey. Watch whether the nine-fold revenue growth holds as the money goes into hiring.