Traversal raises $120 million at a $1 billion valuation for AI agents that investigate production outages

Traversal, a startup that builds AI agents for site reliability engineering (SRE), announced on November 3 a $120 million Series B at a $1 billion valuation. Kleiner Perkins, Redpoint Ventures and Khosla Ventures co-led the round, joined by Sequoia Capital, Battery Ventures and Amex Ventures. The company has now raised $173 million in total, including $48 million when it came out of stealth in June 2025 and a strategic investment from Amex Ventures that December.

What the product does

SRE teams keep production software running and are called in when something breaks. Traversal started with root-cause analysis: when an incident occurs, its system tries to work out what actually went wrong. The company describes two core components. A "Production World Model" turns telemetry, code, deployment history and internal knowledge into a continuously updated map of a customer's systems, and a "Causal Search Engine" tests many possible explanations in parallel. Long-running agents built on top investigate incidents and help with fixes.

According to Unite.AI, about two-thirds of usage now goes beyond root-cause analysis, into areas such as alert triage, release risk and security. The longer-term goal, the company says, is software that can detect a problem, act on it, check the result and prevent a repeat. Named customers include Nubank, Harvey, DigitalOcean, PepsiCo and American Express.

Company-reported growth

Traversal says annual recurring revenue has grown ninefold since January and product usage nearly a hundredfold. It reports more than 150,000 investigations in the last quarter and says it supports over 4,000 engineers. None of these figures has been independently verified, and the company did not disclose absolute revenue.

The founding team comes from academic research in causal machine learning and reinforcement learning: chief executive Anish Agarwal and Raj Agrawal hold MIT doctorates, Raaz Dwivedi a UC Berkeley doctorate, and Ahmed Lone was a quantitative trader at Citadel Securities. The new money will fund hiring in research, engineering and sales.

How far automation can go

The company's own benchmark, ORCA-bench, illustrates the difficulty of the task. Across five general-purpose AI agents, the best reached 25.3% root-cause accuracy on medium-difficulty incidents and 10% on hard ones. Those results measure general models, not Traversal's own system in production, so they say little about how well its product performs. Traversal is betting that a specialised model of each customer's infrastructure closes that gap, a claim customers will judge as deployments such as Nubank's, only a few weeks old, mature.