Traversal Raises $120M to Push AI Operations Toward Autonomy

By Nathan Brooks

Traversal announced a $120 million Series B at a $1 billion valuation on November 3, 2026, accelerating its push toward self-driving production systems. Kleiner Perkins, Redpoint Ventures and Khosla Ventures co-led the round, with Sequoia Capital, Battery Ventures and Amex Ventures participating.

The round lifts total funding to $173 million. It follows the $48 million Traversal announced when it emerged from stealth in June 2025 and a strategic Amex Ventures investment in December 2025; Kleiner Perkins and Sequoia Capital increased their earlier investments. Traversal said the capital will accelerate self-driving production work and hiring across research, engineering and go-to-market roles.

The financing reflects a bet that causal AI can move beyond diagnosis into autonomous remediation. For enterprise technology teams, that is a different proposition from generating incident summaries: the target is software that detects a problem, understands its causes, takes corrective action, verifies the fix and prevents recurrence.

Traversal’s current business remains anchored in investigations and broader production questions. The round therefore expands an existing investigation product while underwriting a more autonomous operating model, a step the company has not yet demonstrated across complex customer environments.

How Traversal connects production evidence to code

Traversal began with incident root-cause analysis, drawing on academic research by its three co-founders in causal machine learning and reinforcement learning. Its central premise is operational rather than fashionable: incidents generate many simultaneous signals, so a plausible correlation cannot establish why a system failed.

The company’s Production World Model converts raw telemetry, code, deployment history and organizational knowledge into a continuously updated, AI-readable representation of a production environment. Its Causal Search Engine tests thousands of hypotheses in parallel, follows dependencies and timing across systems, and converges on causes supported by evidence.

Traversal formally introduced both systems in March 2026 after nearly two years of research and engineering. Its workers now operate as proactive, long-running agents that gather context, investigate incidents or alerts, and assist with remediation when problems appear.

Traversal MCP makes that production intelligence available to coding agents. It connects operational evidence with development work intended to prevent incidents, extending Traversal’s outer loop from diagnosis toward prevention.

Why usage is moving beyond the postmortem

Traversal reports that monthly investigations rose more than 100-fold during 2026, from 513 to nearly 73,000. Roughly two-thirds of product usage now extends beyond root-cause analysis into alert triage, release risk, CMDB validation, security, customer-journey health and production-support questions.

That mix points to a shift from occasional post-incident work into recurring operational decisions. In an accompanying funding release, Traversal reported ninefold year-to-date growth in annual recurring revenue, more than 150,000 customer investigations in the most recent quarter, and support for over 4,000 engineers.

Nubank, Harvey, DigitalOcean and PepsiCo are among the enterprise environments using the platform. Nubank CEO David Vélez said the deployment was “only five weeks old and expanding in waves as results hold up.” That five-week qualification tempers the headline adoption figures and shows why customer results still need to mature.

The benchmark exposes the remaining gap

ORCA-bench, a paper by researchers affiliated with Cornell Tech, Traversal and Columbia University, documents the difficulty of autonomous on-call work. It contains 1,079 root-cause-analysis tasks built around six days and more than 50 GB of metrics, logs and traces from an OpenTelemetry-instrumented system running 19 microservices.

Agents used real observability interfaces and source code while the tasks varied report specificity, detection delays and co-occurring faults. Across five evaluated general-purpose agents, the best root-cause accuracy reached 25.3% on medium-difficulty tasks and 10.0% on hard tasks.

That amounts to roughly one correct answer in four on medium tasks and one in 10 on hard ones. Hallucinated, implausible root causes appeared in 7% to 40% of incident reports, depending on the model. Removing source-code access reduced accuracy and increased hallucinations for every model.

The paper’s finding for practitioners is direct: “fluent code ability does not translate directly into reliable on-call diagnosis.” Code-reading systems can still miss interacting faults or invent causes that fit the available signals without explaining the actual failure.

Those scores do not measure Traversal’s own production performance. The authors caution that ORCA-bench used a controlled public testbed spanning six days, with each task handled separately. Whether the findings carry over to larger, changing proprietary environments remains untested.

What does this change for enterprise technology teams?

Practitioners now have a production-intelligence layer that can join telemetry, deployment history and code context, then carry findings into both remediation and development. They also have a firm boundary: Traversal reports substantial usage, but ORCA-bench neither measures its production results nor proves that software can safely correct faults and prevent recurrence.

The next test for Traversal is whether its causal system can take corrective action, verify the outcome and prevent recurrence across complex customer environments. Until then, the $120 million buys room to test that thesis; it does not settle whether autonomous remediation can work reliably in live enterprise systems.