New York software startup Traversal has closed a $120 million Traversal Series B funding round at a $1 billion valuation, the company confirmed on Tuesday. The investment funds the development of autonomous artificial intelligence agents designed to find and repair corporate infrastructure outages caused by surging volumes of machine-written code.
Key points
- Kleiner Perkins, Redpoint Ventures and Khosla Ventures co-led the $120 million round, taking Traversal’s total funding to $173 million.
- The startup’s platform uses causal machine learning to map IT infrastructure and investigate production failures when alerts fire.
- Enterprise downtime costs have jumped 50 percent over two years as coding assistants accelerate software deployment.
- Traversal’s revenue multiplied ninefold in 2026, with customers conducting more than 150,000 incident investigations last quarter.
Automating site reliability engineering
Site reliability engineering involves keeping web applications, databases and network services running without interruption. When a system fails, on-call engineers must sift through mountains of logs, performance metrics and deployment histories to locate the source of the crash. Traversal aims to replace that manual diagnostic routine with automated agents.
The company announced that Kleiner Perkins, Redpoint Ventures and Khosla Ventures co-led the round. Existing backer Sequoia Capital took part alongside Battery Ventures and Amex Ventures. Both Kleiner Perkins and Sequoia increased their ownership stakes in the business, which emerged from stealth mode in June 2025.
The deal reflects intense investor appetite for operations automation. Venture firms have backed similar agent efforts across business workflows, including when Serval raised funding to expand its enterprise AI agents earlier this year. Traversal intends to use the capital to recruit engineers, researchers and sales staff.
Instead of relying on human operators to inspect observability dashboards, the startup deploys an autonomous site reliability engineer directly inside an enterprise cloud environment. The tool ingests real-time operational telemetry, establishes a baseline of normal application behaviour, and steps in as soon as an alert triggers.
The infrastructure cost of rapid coding
The rush to adopt generative coding tools has created an unintended operational bottleneck. Developer tools let software teams produce code roughly 180 percent faster than two years ago. However, the systems that host, test and run those programmes have become fragile under the increased volume of daily releases.
A study conducted by Splunk and Oxford Economics found that global downtime costs have risen by more than 50 percent across the past two years. Unplanned outages now drain up to $600 billion annually from businesses worldwide as infrastructure teams struggle to diagnose faults.
Anish Agarwal, co-founder and chief executive of Traversal, told reporters that automated code generators produce software that fails in unfamiliar ways. When thousands of microservices interact, pinpointing why a single database query stalled can freeze operations across entire divisions.
Agarwal cited an incident at a major retail bank where thousands of technical staff gathered in a remote crisis room because account holders could not log in. An obscure database delay caused the outage. When a similar fault returned months later, Traversal identified the root cause in five minutes.
Causal models over standard language models
General-purpose large language models often struggle when deployed on production infrastructure. Traversal argues that frontier language models rely on statistical correlations rather than true cause-and-effect reasoning, making them prone to hallucinations during high-pressure IT emergencies.
To measure that gap, Traversal researchers partnered with academic teams at Cornell Tech and Columbia University to create the ORCA-bench benchmark. The test evaluates AI models across 1,079 root-cause analysis tasks drawn from realistic production failures.
The study found that leading frontier language models correctly diagnosed only 25.3 percent of medium-difficulty incidents. When incoming alert descriptions were vague, model accuracy slipped to 10 percent, while hallucination rates climbed past 40 percent. Those errors present substantial operational risks during a live outage.
Traversal bypasses standard language prompts by applying causal machine learning. Its system constructs what the company calls a Production World Model, turning telemetry, source code and release pipelines into an interactive dependency map. A causal search engine then simulates thousands of hypothetical interventions simultaneously to verify which service change triggered the collapse.
Once the engine identifies the fault, it drafts a remediation step for an on-call engineer to review. A companion protocol, Traversal MCP, passes those incident findings back to automated developer agents so they do not write identical architectural bugs in future releases.
Customer growth and market traction
Operational complexity has encouraged enterprise software buyers to dedicate larger portions of their infrastructure budgets to automated monitoring. Analysts tracking enterprise AI spending have reported consistent demand for tools that demonstrate immediate reductions in labour overhead.
Traversal reported that its annual recurring revenue grew ninefold over the past year, though it declined to disclose absolute turnover. The volume of monthly investigations handled by the software surged from 513 in early 2026 to nearly 73,000 in recent weeks.
Companies using the software include digital bank Nubank, cloud infrastructure provider DigitalOcean, food and beverage giant PepsiCo, legal tech firm Harvey, and American Express. Forbes reported that private equity firm Blackstone has also deployed the platform across its internal systems.
David Vélez, chief executive of Nubank, said in a corporate statement that Traversal became an integral part of how the bank’s on-call teams manage live incidents within five weeks of starting its rollout.
The company notes that roughly two thirds of client usage now extends beyond post-crash investigation. Customers run the system to triage incoming alerts, evaluate deployment risks before merging code, and monitor cloud environments for security regressions.
Academic roots and what happens next
Traversal was founded by researchers with backgrounds in statistics, algorithmic trading and distributed computing. Agarwal and co-founder Raj Agrawal completed doctoral degrees at the Massachusetts Institute of Technology, while co-founder Raaz Dwivedi completed his doctorate at the University of California, Berkeley. Agarwal and Dwivedi retain faculty positions at Columbia and Cornell Tech.
A fourth co-founder, Ahmed Lone, joined after working in quantitative trading at Citadel Securities. The founders applied mathematical techniques originally developed for high-frequency market data to cloud observability streams, enabling models to isolate signal anomalies from routine operational noise.
With $120 million in fresh capital from its Series B round, Traversal plans to expand integrations with major cloud vendors and developer platforms over the coming quarters. The startup intends to double its technical headcount in New York by mid-2027.
