Traversal raises $120M at a $1B valuation for AI that finds why production breaks

Traversal's argument is that coding agents have sped up writing software but made running it harder. It cites a survey in which 42% of developers now use AI to write most of their code, up from 12% a year earlier, while 52% say they spend more time debugging that code once it reaches production[1]. The founders, three of whom came from causal machine learning and reinforcement learning research at MIT, Berkeley, Columbia and Cornell, frame incident response as a cause-and-effect problem that plain language models handle badly, because they confuse correlation with cause[1].

Two pieces sit under the product, both introduced in March 2026. The Production World Model turns telemetry, code, deployment history and team knowledge into a continuously updated map of production that an AI can read[1]. The Causal Search Engine lets agents test thousands of hypotheses in parallel and follow causal links across systems to an evidence-backed root cause[1]. Traversal Workers start investigating as soon as an alert fires, and Traversal MCP passes what the system learns about production back to coding agents so they can avoid causing the next incident[2].

Monthly investigations grew more than 100-fold this year, from 513 to nearly 73,000, the company says[1]. About two-thirds of usage now goes beyond root cause analysis into alert triage, release risk, configuration database checks, security and production support questions[1]. The Next Web reports that annual recurring revenue grew nine times this year, with customers such as PepsiCo and Nubank widening their use[2].

The money goes to hiring across research, engineering and go-to-market[1]. The stated goal is "self-driving production": systems that detect issues, find their causes, take corrective action, check that it worked, and prevent repeats[1].

Most AI money in software has gone to the tools that write code. Traversal's unicorn round is a bet that the bigger bill now sits on the other side, in outages and on-call time, and that the same agents flooding codebases will need agents of their own to keep production standing.